Stefano Nichele

• Ramstad, Ola Huse; Sandvig, Axel; Nichele, Stefano & Sandvig, Ioanna (2026). Influence of neural network bursts on functional development. bioRxiv. ISSN 2692-8205. doi: 10.1101/2025.01.21.633559. Full text in Research Archive Show summary

  Network bursts or synchronized burst events are a typical activity seen in most in vitro neural networks. Network bursts arise early in development and as networks mature, activity becomes dominated by bursts propagating across the entirety of the network. The reason for this developmental plateau in vitro is unknown, but to bypass it would confer a significant advantage in the use of in vitro networks for computation. As most neurons in a network participate in network bursts, burst onset supersedes any ongoing activity thereby placing a limit on short term computations equal to the recovery period between bursts. By assessing 521 multielectrode array recordings from day 4-39 in vitro we find that network bursts influence the connectivity, but this change is only weakly associated with the origin of the network bursts. The impacts of bursts on functional integration and segregation in neural networks are discussed along with approaches to mitigate the development and propagation of network bursts in vitro. Additionally, we hypothesize that burst initiation zones or pacemakers are viable targets for stimulation for computation in the context of control and reservoir.

• Nichele, Stefano; Peters, Carl-Hendrik & Reimers, Felix Simon (2026). Benchmarking the State of Networks with a Low-Cost Method Based on Reservoir Computing. In Beushausen, Hans; Kioumarsi, Mahdi & Shafei, Behrouz (Ed.), Proceedings of the 2nd International Conference on Net-Zero Built Environment. Springer. ISSN 9783032276988. p. 992–1004. doi: 10.1007/978-3-032-27698-8_87. Full text in Research Archive Show summary

  Using data from mobile network utilization in Norway, we showcase the possibility of monitoring the state of communication and mobility networks with a non-invasive, low-cost method. This method transforms the network data into a model within the framework of...

• Heiney, Kristine; Józsa, Mónika; Rule, Michael E.; Sprekeler, Henning; Nichele, Stefano & O’Leary, Timothy (2026). Information theoretic measures of neural and behavioural coupling predict representational drift. PLoS Computational Biology. ISSN 1553-734X. 22(2), p. 1013130–1013130. doi: 10.1371/journal.pcbi.1013130. Full text in Research Archive Show summary

  In many parts of the brain, population tuning to stimuli and behaviour gradually changes over the course of days to weeks in a phenomenon known as representational drift. The tuning stability of individual cells varies over the population, and it remains unclear what drives this heterogeneity. We investigate how a neuron’s tuning stability relates to its shared variability with other neurons in the population using two published datasets from posterior parietal cortex and visual cortex. We quantified the contribution of pairwise interactions to behaviour or stimulus encoding by partial information decomposition, which breaks down the mutual information between the pairwise neural activity and the external variable into components uniquely provided by each neuron and by their interactions. Information shared by the two neurons is termed ‘redundant’, and information requiring knowledge of the state of both neurons is termed ‘synergistic’. We found that a neuron’s tuning stability is positively correlated with the strength of its average pairwise redundancy with the population. We hypothesize that subpopulations of neurons show greater stability because they are tuned to salient features common across multiple tasks. Regardless of the mechanistic implications of our work, the stability–redundancy relationship may support improved longitudinal neural decoding in technology that has to track population dynamics over time, such as brain–machine interfaces.

• Pontes-Filho, Sidney; Nichele, Stefano & Lepperød, Mikkel Elle (2025). Reservoir Computing with Evolved Critical Neural Cellular Automata. In Witkowski, Olaf; Adams, Alyssa M.; Sinapayen, Lana; Baltieri, Manuel & Khosravy, Mahdi (Ed.), ALIFE 2025: Ciphers of Life: Proceedings of the Artificial Life Conference 2025. MIT Press. doi: 10.1162/isal.a.832. Full text in Research Archive Show summary

  Criticality is a behavioral state in dynamical systems that is known to present the highest computation capabilities, i.e., information transmission, storage, and modification. Therefore, such systems are ideal candidates as a substrate for reservoir computing, a subfield in artificial intelligence. Our choice of a substrate is a cellular automaton (CA) governed by an artificial neural network, also known as neural cellular automaton (NCA). We apply evolution strategy to optimize the NCA to achieve criticality, demonstrated by power law distributions in structures called avalanches. With an evolved critical NCA, the substrate is tested for reservoir computing. Our evaluation of the substrate is performed with two benchmarks, 5-bit memory task and image classification of handwritten digits. The result of the 5-bit memory task achieved a perfect score and the system managed to remember all 5 bits. The result for the image classification task matched and sometimes surpassed the performance of the best elementary CA for this task. Moreover, the proposed critical NCA may operate as a self-organized critical system, due to its robustness to extreme initial conditions.

• Lindell, Trym; Ramstad, Ola Huse; Sandvig, Ioanna; Sandvig, Axel & Nichele, Stefano (2025). Information Encoding and Decoding in In Vitro Neural Networks on Micro Electrode Arrays through Stimulation Timing. International Journal of Unconventional Computing. ISSN 1548-7199. 20(3), p. 155–195. doi: https:/www.oldcitypublishing.com/wp-content/uploads/2025/09/IJUCv20n3p155-195.pdf. Full text in Research Archive Show summary

  A primary challenge in utilizing in vitro biological neural networks for computations is finding good encoding and decoding schemes for inputting and decoding data to and from the networks. Furthermore, identifying the optimal parameter settings for a given combination of encoding and decoding schemes adds additional complexity to this challenge. In this study we explore stimulation timing as an encoding method, i.e. we encode information as the delay between stimulation pulses and identify the bounds and acuity of stimulation timings which produce linearly separable spike responses. We also examine the optimal readout parameters for a linear decoder in the form of epoch length, time bin size and epoch offset. Our results suggest that stimulation timings between 36 and 436ms may be optimal for encoding and that different combinations of readout parameters may be optimal at different parts of the evoked spike response.

• Bhandari, Shailendra; Nichele, Stefano; Denysov, Sergiy & Lind, Pedro (2025). Optimizing Quantum Circuits with Evolutionary Algorithms for Stable Boolean Gates, Elementary Cellular Automata, and Highly Entangled Quantum States. International Journal of Unconventional Computing. ISSN 1548-7199. 20(3), p. 225–257. doi: 10.32908/ijuc.v20.270425. Full text in Research Archive Show summary

  We investigate the potential of bio-inspired evolutionary algorithms for designing quantum circuits with specific goals, focusing on two particular tasks. The first one is motivated by the ideas of Artificial Life that are used to reproduce stochastic cellular automata with given rules. We test the robustness of quantum implementations of the cellular automata for different numbers of quantum gates The second task deals with the sampling of quantum circuits that generate highly entangled quantum states, which constitute an important resource for quantum computing. In particular, an evolutionary algorithm is employed to optimize circuits with respect to a fitness function defined with the Mayer-Wallach entanglement measure. We demonstrate that, by balancing the mutation rate between exploration and exploitation, we can find entangling quantum circuits for up to five qubits. We also discuss the trade-off between the number of gates in quantum circuits and the computational costs of finding the gate arrangements leading to a strongly entangled state. Our findings provide additional insight into the trade-off between the complexity of a circuit and its performance, which is an important factor in the design of quantum circuits.

• Pietropolli, Gloria; Nichele, Stefano & Medvet, Eric (2024). The Role of the Substrate in CA-based Evolutionary Algorithms. In Li, Xiandong & Handl, Julia (Ed.), GECCO '24: Proceedings of the Genetic and Evolutionary Computation Conference. Association for Computing Machinery (ACM). ISSN 9798400704949. p. 768–777. doi: 10.1145/3638529.3654112. Full text in Research Archive
• Valderhaug, Vibeke Devold; Ramstad, Ola Huse; Wijdeven, Rosanne Francisca van de; Heiney, Kristine; Nichele, Stefano & Sandvig, Axel [Show all 7 contributors for this article] (2024). Micro-and mesoscale aspects of neurodegeneration in engineered human neural networks carrying the LRRK2 G2019S mutation. Frontiers in Cellular Neuroscience. 18, p. 1–15. doi: 10.3389/fncel.2024.1366098. Full text in Research Archive Show summary

  Mutations in the leucine-rich repeat kinase 2 (LRRK2) gene have been widely linked to Parkinson’s disease, where the G2019S variant has been shown to contribute uniquely to both familial and sporadic forms of the disease. LRRK2-related mutations have been extensively studied, yet the wide variety of cellular and network events related to these mutations remain poorly understood. The advancement and availability of tools for neural engineering now enable modeling of selected pathological aspects of neurodegenerative disease in human neural networks in vitro. Our study revealed distinct pathology associated dynamics in engineered human cortical neural networks carrying the LRRK2 G2019S mutation compared to healthy isogenic control neural networks. The neurons carrying the LRRK2 G2019S mutation self-organized into networks with aberrant morphology and mitochondrial dynamics, affecting emerging structure–function relationships both at the micro-and mesoscale. Taken together, the findings of our study points towar

• Glover, Tom Eivind; Jahren, Christian Ruben; Francesco, Martinuzzi; Lind, Pedro & Nichele, Stefano (2024). A sensitivity analysis of cellular automata and heterogeneous topology networks: partially-local cellular automata and homogeneous homogeneous random boolean networks. International Journal of Parallel, Emergent and Distributed Systems. ISSN 1744-5760. 40(1), p. 59–99. doi: 10.1080/17445760.2024.2396334. Full text in Research Archive
• Jain, Sanyam & Nichele, Stefano (2024). Frequency-Histogram Coarse Graining in Elementary and 2-Dimensional Cellular Automata. Nordic Machine Intelligence (NMI). 3(2), p. 15–110. doi: 10.5617/nmi.10458. Full text in Research Archive Show summary

  Cellular automata and other discrete dynamical systems have long been studied as models of emergent complexity. Recently, neural cellular automata have been proposed as models to investigate the emerge of a more general artificial intelligence, thanks to their propensity to support properties such as self-organization, emergence, and open-endedness. However, understanding emergent complexity in large scale systems is an open challenge. How can the important computations leading to emergent complex structures and behaviors be identified? In this work, we systematically investigate a form of dimensionality reduction for 1-dimensional and 2-dimensional cellular automata based on coarse-graining of macrostates into smaller blocks. We discuss selected examples and provide the entire exploration of coarse graining with different filtering levels in the appendix (available also digitally at this link: https://s4nyam.github.io/eca88/. We argue that being able to capture emergent complexity in AI systems may pave the way to open-ended evolution, a plausible path to reach artificial general intelligence.

• Lindell, Trym; Ramstad, Ola Huse; Sandvig, Ioanna; Sandvig, Axel & Nichele, Stefano (2024). Chaotic Time Series Prediction in Biological Neural Network Reservoirs on Microelectrode Arrays. In Hirose, Akira & Ishibuchi, Hisao (Ed.), 2024 International Joint Conference on Neural Networks (IJCNN). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9798350359312. p. 1–10. doi: 10.1109/IJCNN60899.2024.10650567. Full text in Research Archive
• Farner, Jørgen Jensen; Ramstad, Ola Huse; Nichele, Stefano & Heiney, Kristine (2024). Local Delay Plasticity Supports Generalized Learning in Spiking Neural Networks. In Villani, Marco; Cagnoni, Stefano & Serra, Roberto (Ed.), Artificial Life and Evolutionary Computation. WIVACE 2023.. Springer. ISSN 9783031574306. p. 241–255. doi: 10.1007/978-3-031-57430-6_19. Full text in Research Archive Show summary

  We propose a novel local learning rule for spiking neural networks in which spike propagation times undergo activity-dependent plasticity. Our plasticity rule aligns pre-synaptic spike times to produce a stronger and more rapid response. Inputs are encoded by latency coding and outputs decoded by matching similar patterns of output spiking activity. We demonstrate the use of this method in a three-layer feed-foward network with inputs from a database of handwritten digits. Networks consistently showed improved classification accuracy after training, and training with this method also allowed networks to generalize to an input class unseen during training. Our proposed method takes advantage of the ability of spiking neurons to support many different time-locked sequences of spikes, each of which can be activated by different input activations. The proof-of-concept shown here demonstrates the great potential for local delay learning to expand the memory capacity and generalizability of spiking neural networks and offers new perspectives on how to configure neuromorphic hardware.

• Jain, Sanyam; Shrestha, Aarati & Nichele, Stefano (2024). Capturing Emerging Complexity in Lenia. In Villani, Marco; Cagnoni, Stefano & Serra, Roberto (Ed.), Artificial Life and Evolutionary Computation. WIVACE 2023.. Springer. ISSN 9783031574306. p. 41–53. doi: 10.1007/978-3-031-57430-6_4. Full text in Research Archive Show summary

  This work investigates the emergent complexity in Lenia, an artificial life platform that simulates ecosystems of digital creatures. Lenia’s ecosystem consists of a continuous cellular automaton where simple artificial organisms can move, grow, and reproduce. Measuring longterm complex emerging behavior in Lenia is an open problem. Here we utilize evolutionary computation where Lenia kernels are used as genotypes while keeping other Lenia parameters, such as the growth function, fixed. First, we use Variation over Time as a fitness function where higher variance between the frames is rewarded. Second, we use Auto-encoder based fitness where variation of the list of reconstruction loss for the frames is rewarded. Third, we perform a combined fitness where higher variation of the pixel density of reconstructed frames is rewarded. Finally, after performing several experiments for each fitness function for 500 generations, we select interesting runs for an extended evolutionary time of 2500 generations. Results indicate that the kernel’s center of mass increases with a specific set of pixels and the overall complexity measures also increase. We also utilize our evolutionary method initialized from known handcrafted kernels. Overall, this project aims at investigating the potential of Lenia as ecosystem for emergent complexity in open-ended artificial intelligence systems.

• Nichele, Stefano; Sayama, Hiroki; Medvet, Eric; Nehaniv, Chrystopher & Pavone, Mario (2024). Editorial: Special Issue "The Distributed Ghost"-Cellular Automata, Distributed Dynamical Systems, and Their Applications to Intelligence. Artificial Life. ISSN 1064-5462. 31(1), p. 65–67. doi: 10.1162/artl_e_00450. Full text in Research Archive
• Glover, Tom Eivind; Lind, Pedro; Yazidi, Anis; Osipov, Evgeny & Nichele, Stefano (2023). Investigating Rules and Parameters of Reservoir Computing with Elementary Cellular Automata, with a Criticism of Rule 90 and the Five-Bit Memory Benchmark. Complex Systems. 32(3), p. 309–351. doi: 10.25088/ComplexSystems.32.3.309. Full text in Research Archive Show summary

  Reservoir computing with cellular automata (ReCAs) is a promising concept by virtue of its potential for effective hardware implementation. In this paper, we explore elementary cellular automata rules in the context of ReCAs and the 5-bit memory benchmark. We combine elementary cellular automaton theory with our results and use them to identify and explain some of the patterns found. Furthermore, we use these findings to expose weaknesses in the 5-bit memory benchmark as it is typically applied in ReCAs, such as pointing out what features it selects for or solving it using random vectors. We look deeply into previously successful rules in ReCAs such as rule 90 and explain some of the consequences of its additive properties as well as the correlation between grid size and performance. Additionally, we present results from exhaustively exploring ReCAs on key parameters such as distractor period, iterations and grid size. The findings of this paper should motivate the ReCAs community to move away from using the 5-bit memory benchmark as it is being applied today.

• Lindell, Trym; Hudcová, Barbora & Nichele, Stefano (2023). Canonical Computations in Cellular Automata and Their Application for Reservoir Computing. ALIFE : Proceedings of the Artificial Life Conference. 16. doi: 10.1162/isal_a_00594. Full text in Research Archive Show summary

  Cellular Automata (CAs) have potential as powerful parallel computational systems, which has lead to the use of CAs as reservoirs in reservoir computing. However, why certain Cellular Automaton (CA) rules, sizes and input encodings are better or worse at a given task is not well understood. We present a method that enables identification and visualization of the specific information content, flow and transformations within the space-time diagram of CA. We interpret each spatio-temporal location in CA’s space-time diagram as a function of its input and call this novel notion the CA’s Canonical Computations (CCs). This allows us to analyze the available information from the space-time diagrams as partitions of the input set. The method also reveals how input-encoder-rule interactions transform the information flow by changing features like spatial and temporal location stability as well as the specific information produced. This general approach for analysing CA is discussed for the engineering of reservoir computing systems.

• Nadizar, Giorgia; Medvet, Eric; Nichele, Stefano & Filho, Sidney Pontes (2023). An experimental comparison of evolved neural network models for controlling simulated modular soft robots. Applied Soft Computing. ISSN 1568-4946. 145. doi: 10.1016/j.asoc.2023.110610. Full text in Research Archive Show summary

  Voxel-based soft robots (VSRs) are a type of modular robots composed by interconnected soft and deformable blocks, i.e., voxels. Thanks to the softness of their bodies, VSRs may exhibit rich dynamic behaviors. One open question is what type of neural controller is most suitable for a given morphology and sensory apparatus in a given environment. One observation is that artificial neural networks with state may be able to cope with the dynamical nature of VSR bodies and their morphological computation. In this work, we consider four types of controllers, i.e., multilayer perceptrons (MLPs, stateless), recurrent neural networks (RNNs), spiking neural networks (SNNs) without homeostasis, and SNNs with homeostasis. We consider three robot morphologies tested for locomotion, where each morphology is investigated in simulation with three different types and number of sensors. Neural network controllers are optimized with neuroevolution, and the experimental results are compared in terms of effectiveness, efficiency, and generalization ability. In addition, we analyze the resulting behavior of the robots systematically. Our results show that RNNs are typically more effective while MLPs are often the weakest controllers, particularly for robots with few sensors. However, SNNs are more capable in terms of generalization and the mechanism of homeostasis is often beneficial. Finally, we show that RNNs and SNNs with homeostasis produce a more wide variety of behaviors.

• Pierro, Alessandro; Heiney, Kristine; Shrivastava, Shamit; Marcucci, Gulia & Nichele, Stefano (2023). Optimization of a Hydrodynamic Computational Reservoir through Evolution. In Silva, Sara & Paquete, Luis (Ed.), GECCO '23: Proceedings of the Genetic and Evolutionary Computation Conference. Association for Computing Machinery (ACM). ISSN 9798400701191. p. 202–210. doi: 10.1145/3583131.3590355. Full text in Research Archive Show summary

  As demand for computational resources reaches unprecedented levels, research is expanding into the use of complex material substrates for computing. In this study, we interface with a model of a hydrodynamic system, under development by a startup, as a computational reservoir and optimize its properties using an evolution in materio approach. Input data are encoded as waves applied to our shallow water reservoir, and the readout wave height is obtained at a fixed detection point. We optimized the readout times and how inputs are mapped to the wave amplitude or frequency using an evolutionary search algorithm, with the objective of maximizing the system’s ability to linearly separate observations in the training data by maximizing the readout matrix determinant. Applying evolutionary methods to this reservoir system substantially improved separability on an XNOR task, in comparison toimplementations with hand-selected parameters. We also applied our approach to a regression task and show that our approach improves out-of-sample accuracy. Results from this study will inform how we interface with the physical reservoir in future work, and we will use these methods to continue to optimize other aspects of the physical implementation of this system as a computational reservoir.

• Glover, Tom Eivind; Jahren, Christian Ruben; Ramstad, Ola Huse & Nichele, Stefano (2023). Minimum Equivalence in Random Boolean Networks, Elementary Cellular Automata, and Beyond. In Iizuki, Hiroyuki; Suzuki, Keisuke; Uno, Ryoko; Damiano, Luisa; Spychala, Nadine; Aguilera, Miguel; Izquierdo, Eduardo; Suzuki, Reiji & Baltieri, Manuel (Ed.), ALIFE 2023: Ghost in the Machine: Proceedings of the 2023 Artificial Life Conference. MIT Press. p. 100–108. doi: 10.1162/isal_a_00592. Full text in Research Archive Show summary

  Random Boolean networks (RBN) and Cellular Automata (CA) operate in a very similar way. They update their state with simple deterministic functions called Boolean function or Transition Table (TT), both being essentially the same mechanism under different names. This paper applies a concept most known from CA called Minimum Equivalence (ME). ME is applied to RBN and shows how to calculate the number of unique computations for a given number of neighbours. Crucially, it is shown how RBN rules are even more equivalent than in CA, how the set can be reduced into even fewer unique rules, and how the concept becomes more relevant with larger neighbourhoods. For example, switching transformation alone reduces the number of unique rules in RBN with 4 neighbours from 65 536 to only 3 984 (6.1%) rules. Additionally, this paper examines the ME and transformations in substrates beyond Elementary CA (ECA), such as CA with additional spatial dimensions and number of states.

• Riccio, Piera; Galati, Francesco; Zuluaga, Maria; Martin, Juan Carlos De & Nichele, Stefano (2022). Translating Emotions from EEG to Visual Arts. In Martins, Tiago; Rodríguez-Fernández, Nereida & Rebelo, Sérgio M. (Ed.), Artificial Intelligence in Music, Sound, Art and Design. 11th International Conference, EvoMUSART 2022, held as part of EvoStar 2022, Madrid, Spain, April 20–22, 2022, Proceedings. Springer. ISSN 9783031037887. p. 243–258. doi: 10.1007/978-3-031-03789-4. Full text in Research Archive Show summary

  Exploring the potentialities of artificial intelligence (AI) in the world of arts is fundamental to understand and define how this technology is shaping our creativity. We propose a system that generates emotionally expressive paintings from EEG signals. The emotional information, encoded from the signals through a graph neural network, is inputted to a generative adversarial network (GAN), trained on a dataset of paintings. The design and experimental choices at the base of this work rely on the understanding that emotions are hard to define and formalize. Despite this, the proposed results witness an interaction between an AI system and a human, capable of producing an original and artistic re-interpretation of emotions. These results have a promising potential for AI technologies applied to visual arts.

• Filho, Sidney Pontes; Lind, Pedro & Nichele, Stefano (2022). Assessing the robustness of critical behavior in stochastic cellular automata. Physica D : Non-linear phenomena. ISSN 0167-2789. 441(441). doi: 10.1016/j.physd.2022.133507. Full text in Research Archive Show summary

  There is evidence that biological systems, such as the brain, work at a critical regime robust to noise, and are therefore able to remain in it under perturbations. In this work, we address the question of robustness of critical systems to noise. In particular, we investigate the robustness of stochastic cellular automata (CAs) at criticality. A stochastic CA is one of the simplest stochastic models showing criticality. The transition state of stochastic CA is defined through a set of probabilities. We systematically perturb the probabilities of an optimal stochastic CA known to produce critical behavior, and we report that such a CA is able to remain in a critical regime up to a certain degree of noise. We present the results using error metrics of the resulting power-law fitting, such as Kolmogorov–Smirnov statistic and Kullback–Leibler divergence. We discuss the implication of our results in regards to future realization of brain-inspired artificial intelligence systems.

• Bhandari, Shailendra; Overskott, Sebastian Testanière; Adamopoulos, Ioannis; Lind, Pedro; Denysov, Sergiy & Nichele, Stefano (2022). Evolving Quantum Circuits to Implement Stochastic and Deterministic Cellular Automata Rules. In Chopard, Bastien; Bandini, Stefania; Dennunzio, Alberto & Haddad, Mira Arabi (Ed.), Cellular Automata, 15th International Conference on Cellular Automata for Research and Industry, ACRI 2022, Geneva, Switzerland, September 12–15, 2022, Proceedings. Springer. ISSN 9783031149252. p. 119–129. doi: 10.1007/978-3-031-14926-9_11. Full text in Research Archive Show summary

  The aim of this work is to generate specific rules of deterministic and stochastic cellular automata (CA) using the set of five quantum gates, which is known to generate any quantum circuit. To build such quantum circuits, we use an evolutionary algorithm, based in mutations, which allows the optimization of quantum gate types and their connectivity. The fitness function of the evolutionary algorithm aims at minimizing the difference between the output of the quantum circuit and the CA rule. We also inspect the differences observed when changing the number of gates and the mutation rate. We benchmark our methods with stochastic as well as deterministic CA rules, and briefly discuss the possible extensions their quantum “cousins” may enable.

• Heiney, Kristine Anne; Ramstad, Ola Huse; Fiskum, Vegard; Sandvig, Axel; Sandvig, Ioanna & Nichele, Stefano (2022). Neuronal avalanche dynamics and functional connectivity elucidate information propagation in vitro. Frontiers in Neural Circuits. 16. doi: 10.3389/fncir.2022.980631. Full text in Research Archive Show summary

  Cascading activity is commonly observed in complex dynamical systems, including networks of biological neurons, and how these cascades spread through the system is reliant on how the elements of the system are connected and organized. In this work, we studied networks of neurons as they matured over 50 days in vitro and evaluated both their dynamics and their functional connectivity structures by observing their electrophysiological activity using microelectrode array recordings. Correlations were obtained between features of their activity propagation and functional connectivity characteristics to elucidate the interplay between dynamics and structure. The results indicate that in vitro networks maintain a slightly subcritical state by striking a balance between integration and segregation. Our work demonstrates the complementarity of these two approaches—functional connectivity and avalanche dynamics—in studying information propagation in neurons in vitro, which can in turn inform the design and optimization of engineered computational substrates.

• Dumo, Glare Eugenio; Lind, Pedro & Nichele, Stefano (2022). An HCI Experiment to Explore Interactive Artificial Life Art. In Stephanidis, S; Antona, M & Ntoa, S (Ed.), HCI International 2022 Posters. Springer. ISSN 9783031063909. p. 204–211. doi: 10.1007/978-3-031-06391-6_27. Full text in Research Archive
• Tapio, Hege; Nichele, Stefano; Bergaust, Kristin & Christensen-Scheel, Boel (2022). Artists Encoding New Meanings through BioArt and Unconventional Computing. In Adamatzky, Andrew (Eds.), WSPC Book Series in Unconventional Computing: Volume 2 Unconventional Computing, Arts, Philosophy. World Scientific. ISSN 9789811257148. p. 381–395. doi: 10.1142/9789811257155_0023. Full text in Research Archive Show summary

  The unique compendium re-assesses the value of future and emergent computing technologies via artistic and philosophical means. The book encourages scientists to adopt inspiring thinking of artists and philosophers to reuse scientific concepts in their works.

• Filho, Sidney Pontes; Olsen, Kristoffer; Yazidi, Anis; Riegler, Michael; Halvorsen, Pål & Nichele, Stefano (2022). Towards the Neuroevolution of Low-level artificial general intelligence. Frontiers in Robotics and AI. 9. doi: 10.3389/frobt.2022.1007547. Full text in Research Archive Show summary

  In this work, we argue that the search for Artificial General Intelligence should start from a much lower level than human-level intelligence. The circumstances of intelligent behavior in nature resulted from an organism interacting with its surrounding environment, which could change over time and exert pressure on the organism to allow for learning of new behaviors or environment models. Our hypothesis is that learning occurs through interpreting sensory feedback when an agent acts in an environment. For that to happen, a body and a reactive environment are needed. We evaluate a method to evolve a biologicallyinspired artificial neural network that learns from environment reactions named Neuroevolution of Artificial General Intelligence, a framework for low-level artificial general intelligence. This method allows the evolutionary complexification of a randomly-initialized spiking neural network with adaptive synapses, which controls agents instantiated in mutable environments. Such a configuration allows us to benchmark the adaptivity and generality of the controllers. The chosen tasks in the mutable environments are food foraging, emulation of logic gates, and cart-pole balancing. The three tasks are successfully solved with rather small network topologies and therefore it opens up the possibility of experimenting with more complex tasks and scenarios where curriculum learning is beneficial.

• Filho, Sidney Pontes; Walker, Kathryn; Najarro, Elias; Nichele, Stefano & Risi, Sebastian (2022). A single neural cellular automaton for body-brain co-evolution, GECCO '22: Proceedings of the Genetic and Evolutionary Computation Conference Companion. Association for Computing Machinery (ACM). ISSN 9781450392686. p. 148–151. doi: 10.1145/3520304.3529002. Full text in Research Archive Show summary

  A successful development of a complex multicellular organism took millions of years of evolution. The genome of such a multicellular organism guides the development of its body from a single cell, including its control system. Our goal is to imitate this natural pro- cess using a single neural cellular automaton (NCA) as a genome for modular robotic agents. In the introduced approach, called Neural Cellular Robot Substrate (NCRS), a single NCA guides the growth of a robot and the cellular activity which controls the robot during deployment. We also introduce three benchmark environments,which test the ability of the approach to grow different robot mor- phologies. We evolve the NCRS with covariance matrix adaptation evolution strategy (CMA-ES), and covariance matrix adaptation MAP-Elites (CMA-ME). While the NCRS is able to solve the easier tasks in the benchmark, the success rate reduces when the difficulty of the task increases. We discuss directions for future work that may facilitate the use of the NCRS approach for more complex domains.

• Nadizar, Giorgia; Medvet, Eric; Ramstad, Ola Huse; Nichele, Stefano; Pellegrino, Felice Andrea & Zullich, Marco (2022). Merging pruning and neuroevolution: towards robust and efficient controllers for modular soft robots. Knowledge engineering review (Print). ISSN 0269-8889. 37(e3), p. 1–27. doi: 10.1017/S0269888921000151. Full text in Research Archive Show summary

  Artificial neural networks (ANNs) can be employed as controllers for robotic agents. Their structure is often complex, with many neurons and connections, especially when the robots have many sensors and actuators distributed across their bodies and/or when high expressive power is desirable. Pruning (removing neurons or connections) reduces the complexity of the ANN, thus increasing its energy efficiency, and has been reported to improve the generalization capability, in some cases. In addition, it is well-known that pruning in biological neural networks plays a fundamental role in the development of brains and their ability to learn. In this study, we consider the evolutionary optimization of neural controllers for the case study of Voxel-based soft robots, a kind of modular, bio-inspired soft robots, applying pruning during fitness evaluation. For a locomotion task, and for centralized as well as distributed controllers, we experimentally characterize the effect of different forms of pruning on after-pruning effectiveness, life-long effectiveness, adaptability to new terrains, and behavior. We find that incorporating some forms of pruning in neuroevolution leads to almost equally effective controllers as those evolved without pruning, with the benefit of higher robustness to pruning. We also observe occasional improvements in generalization ability.

• Glover, Tom Eivind; Lind, Pedro; Yazidi, Anis; Osipov, Evgeny & Nichele, Stefano (2021). The Dynamical Landscape of Reservoir Computing with Elementary Cellular Automata. In Cejkova, Jitka; Holler, Silvia; Soros, Lisa & Witkowski, Olaf (Ed.), ALIFE 2021: Proceedings of the Artificial Life Conference 2021. MIT Press. p. 292–300. doi: 10.1162/isal_a_00440. Full text in Research Archive Show summary

  Reservoir Computing with Cellular Automata (ReCA) is a promising concept by virtue of its potential for efficient hardware implementation and theoretical understanding of Cellular Auotmata (CA). However, ReCA has so far only been studied in exploratory studies. In this work, we take a more in depth view of the landscape of Elementary Cellular Automata for Reservoir Computing. In this paper, the ReCA is applied to the X-bit memory benchmark with a thorough exploration for key parameters including number of random mappings (R), number of bits (Nb) and size of the vector that the random mapping is mapped to (Ld). Our evidence shows that the parameter space, including the full panoply of CA rules, is much richer then what previous evidence indicates. This suggests that some CA rules would require careful consideration and custom parameters setup.

• Farner, Jørgen Jensen; Weydahl, Håkon; Jahren, Christian Ruben; Ramstad, Ola Huse; Nichele, Stefano & Heiney, Kristine Anne (2021). Evolving spiking neuron cellular automata and networks to emulate in vitro neuronal activity. In Haddow, Pauline (Eds.), IEEE Symposium Series on Computational Intelligence 2021. IEEE Press. ISSN 9781728190488. doi: 10.1109/SSCI50451.2021.9660185. Full text in Research Archive Show summary

  Neuro-inspired models and systems have great potential for applications in unconventional computing. Often, the mechanisms of biological neurons are modeled or mimicked in simulated or physical systems in an attempt to harness some of the computational power of the brain. However, the biological mechanisms at play in neural systems are complicated and challenging to capture and engineer; thus, it can be simpler to turn to a data-driven approach to transfer features of neural behavior to artificial substrates. In the present study, we used an evolutionary algorithm to produce spiking neural systems that emulate the patterns of behavior of biological neurons in vitro. The aim of this approach was to develop a method of producing models capable of exhibiting complex behavior that may be suitable for use as computational substrates. Our models were able to produce a level of network-wide synchrony and showed a range of behaviors depending on the target data used for their evolution, which was from a range of neuronal culture densities and maturities. The genomes of the top-performing models indicate the excitability and density of connections in the model play an important role in determining the complexity of the produced activity.

• Filho, Sidney Pontes; Dahl, Annelene; Nichele, Stefano & Mello, Gustavo (2021). A Deep Learning-Based Tool for Automatic Brain Extraction from Functional Magnetic Resonance Images of Rodents. In Arai, Kohei (Eds.), Intelligent Systems and Applications, Proceedings of the 2021 Intelligent Systems Conference (IntelliSys 2021) Volume 1. Springer Nature. ISSN 9783030821951. p. 549–558. doi: 10.1007/978-3-030-82199-9_36. Full text in Research Archive Show summary

  Removing skull artifacts from functional magnetic images (fMRI) is a well understood and frequently encountered problem. Because the fMRI field has grown mostly due to human studies, many new tools were developed to handle human data. Nonetheless, these tools are not equally useful to handle the data derived from animal studies, especially from rodents. This represents a major problem to the field because rodent studies generate larger datasets from larger populations, which implies that preprocessing these images manually to remove the skull becomes a bottleneck in the data analysis pipeline. In this study, we address this problem by implementing a neural network-based method that uses a U-Net architecture to segment the brain area into a mask and removing the skull and other tissues from the image. We demonstrate several strategies to speed up the process of generating the ground-truth of the dataset using watershedding, and several strategies for data augmentation that allowed to train robustly the U-Net to perform the segmentation. Finally, we deployed the trained network freely available.

• Nadizar, Giorgia; Medvet, Eric; Pellegrino, Felice Andrea; Zullich, Marco & Nichele, Stefano (2021). On the effects of pruning on evolved neural controllers for soft robots. In Chicano, Francisco (Eds.), GECCO '21: Proceedings of the Genetic and Evolutionary Computation Conference Companion. Association for Computing Machinery (ACM). ISSN 9781450383516. p. 1744–1752. Full text in Research Archive Show summary

  Artificial neural networks (ANNs) are commonly used for controlling robotic agents. For robots with many sensors and actuators, ANNs can be very complex, with many neurons and connections. Removal of neurons or connections, i.e., pruning, may be desirable because (a) it reduces the complexity of the ANN, making its operation more energy efficient, and (b) it might improve the generalization ability of the ANN. Whether these goals can actually be achieved in practice is however still not well known. On the other hand, it is widely recognized that pruning in biological neural networks plays a fundamental role in the development of brains and their ability to learn. In this work, we consider the case of Voxel-based Soft Robots, a kind of robots where sensors and actuators are distributed over the body and that can be controlled with ANNs optimized by means of neuroevolution. We experimentally characterize the effect of different forms of pruning on the effectiveness of neuroevolution, also in terms of generalization ability of the evolved ANNs. We find that, with some forms of pruning, a large portion of the connections can be pruned without strongly affecting robot capabilities. We also observe sporadic improvements in generalization ability.

• Valderhaug, Vibeke Devold; Heiney, Kristine; Ramstad, Ola Huse; Bråthen, Geir; Kuan, Wei-Li & Nichele, Stefano [Show all 8 contributors for this article] (2021). Early functional changes associated with alpha-synuclein proteinopathy in engineered human neural networks. American Journal of Physiology - Cell Physiology. 320(6), p. 1–12. doi: 10.1152/AJPCELL.00413.2020. Full text in Research Archive Show summary

  A patterned spread of proteinopathy represents a common characteristic of many neurodegenerative diseases. In Parkinson’s disease (PD), misfolded forms of α-synuclein proteins accumulate in hallmark pathological inclusions termed Lewy bodies and Lewy neurites. Such protein aggregates seem to affect selectively vulnerable neuronal populations in the substantia nigra and to propagate within interconnected neuronal networks. Research findings suggest that these proteinopathic inclusions are present at very early time points in disease development, even before clear behavioral symptoms of dysfunction arise. In this study, we investigate the early pathophysiology developing after induced formation of such PD-related α-synuclein inclusions in a physiologically relevant in vitro setup using engineered human neural networks. We monitor the neural network activity using multielectrode arrays (MEAs) for a period of 3 wk following proteinopathy induction to identify associated changes in network function, with a special emphasis on the measure of network criticality. Self-organized criticality represents the critical point between resilience against perturbation and adaptational flexibility, which appears to be a functional trait in self-organizing neural networks, both in vitro and in vivo. We show that although developing pathology at early onset is not clearly manifest in standard measurements of network function, it may be discerned by investigating differences in network criticality states.

• Heiney, Kristine Anne; Valderhaug, Vibeke Devold; Ramstad, Ola Huse; Sandvig, Ioanna; Sandvig, Axel & Nichele, Stefano (2021). Hallmarks of Criticality in Neuronal Networks Depend on Cell Type and the Temporal Resolution of Neuronal Avalanches. International Journal of Unconventional Computing. ISSN 1548-7199. 16(4), p. 303–326. Full text in Research Archive Show summary

  The human brain has a remarkable capacity for computation, and it has been theorized that this capacity arises from the brain self-organizing into the critical state, a dynamical state poised between ordered and dis- ordered behavior and widely considered to be well-suited for computation. Criticality is commonly identified in in vitro neuronal networks using an analytical approach based on the size distribution of cascades of activity called neuronal avalanches. In this study, criticality analysis was applied to different in vitro neuronal networks with two areas of focus: evaluating the effect of the size of the time bins used for neuronal avalanche detection and observation of the development of networks of neurons derived from human induced pluripotent stem cells. This pre- preliminary study is expected to aid in the construction of models capable of emulating neuronal behaviors identified as well-suited for computation and ultimately inform the development of brain-inspired computing substrates that are better able to keep pace with increased demand for data storage and processing power.

• Heiney, Kristine; Ramstad, Ola Huse; Fiskum, Vegard; Christiansen, Nicholas; Sandvig, Axel & Nichele, Stefano [Show all 7 contributors for this article] (2021). Criticality, connectivity, and neural disorder: A multifaceted approach to neural computation. Frontiers in Computational Neuroscience. 15:611183, p. 1–19. doi: 10.3389/fncom.2021.611183. Full text in Research Archive Show summary

  It has been hypothesized that the brain optimizes its capacity for computation by self-organizing to a critical point. The dynamical state of criticality is achieved by striking a balance such that activity can effectively spread through the network without overwhelming it and is commonly identified in neuronal networks by observing the behavior of cascades of network activity termed “neuronal avalanches.” The dynamic activity that occurs in neuronal networks is closely intertwined with how the elements of the network are connected and how they influence each other’s functional activity. In this review, we highlight how studying criticality with a broad perspective that integrates concepts from physics, experimental and theoretical neuroscience, and computer science can provide a greater understanding of the mechanisms that drive networks to criticality and how their disruption may manifest in different disorders. First, integrating graph theory into experimental studies on criticality, as is becoming more common in theoretical and modeling studies, would provide insight into the kinds of network structures that support criticality in networks of biological neurons. Furthermore, plasticity mechanisms play a crucial role in shaping these neural structures, both in terms of homeostatic maintenance and learning. Both network structures and plasticity have been studied fairly extensively in theoretical models, but much work remains to bridge the gap between theoretical and experimental findings. Finally, information theoretical approaches can tie in more concrete evidence of a network’s computational capabilities. Approaching neural dynamics with all these facets in mind has the potential to provide a greater understanding of what goes wrong in neural disorders. Criticality analysis therefore holds potential to identify disruptions to healthy dynamics, granted that robust methods and approaches are considered.

• Talamini, Jacopo; Medvet, Eric & Nichele, Stefano (2021). Criticality-Driven Evolution of Adaptable Morphologies of Voxel-Based Soft-Robots. Frontiers in Robotics and AI. 8. doi: 10.3389/frobt.2021.673156. Full text in Research Archive Show summary

  The paradigm of voxel-based soft robots has allowed to shift the complexity from the control algorithm to the robot morphology itself. The bodies of voxel-based soft robots are extremely versatile and more adaptable than the one of traditional robots, since they consist of many simple components that can be freely assembled. Nonetheless, it is still not clear which are the factors responsible for the adaptability of the morphology, which we define as the ability to cope with tasks requiring different skills. In this work, we propose a task-agnostic approach for automatically designing adaptable soft robotic morphologies in simulation, based on the concept of criticality. Criticality is a property belonging to dynamical systems close to a phase transition between the ordered and the chaotic regime. Our hypotheses are that 1) morphologies can be optimized for exhibiting critical dynamics and 2) robots with those morphologies are not worse, on a set of different tasks, than robots with handcrafted morphologies. We introduce a measure of criticality in the context of voxel-based soft robots which is based on the concept of avalanche analysis, often used to assess criticality in biological and artificial neural networks. We let the robot morphologies evolve toward criticality by measuring how close is their avalanche distribution to a power law distribution. We then validate the impact of this approach on the actual adaptability by measuring the resulting robots performance on three different tasks designed to require different skills. The validation results confirm that criticality is indeed a good indicator for the adaptability of a soft robotic morphology, and therefore a promising approach for guiding the design of more adaptive voxel-based soft robots.

• Zhang, Jianhua; Li, Jianrong & Nichele, Stefano (2020). Instantaneous Mental Workload Recognition Using Wavelet-Packet Decomposition and Semi-Supervised Learning. In Huang, Tingwen (Eds.), Proceedings of IEEE Symposium Series on Computational Intelligence (SSCI2019). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9781728124858. p. 410–416. doi: 10.1109/SSCI44817.2019.9002997. Full text in Research Archive Show summary

  The real-time monitoring of human operator's mental workload (MWL) is crucial for development of adaptive/intelligent human-machine cooperative systems in various safety/mission-critical application fields. Although datadriven approach has shown promise in MWL recognition, its major difficulty lies in how to acquire sufficient labeled data to train the model. This paper applies semi-supervised extreme learning machine (SS-ELM) to the problem of MWL classification based only on a small number of labeled data. The experimental data analysis results have shown that the proposed SS-ELM paradigm can effectively improve the accuracy and efficiency of MWL classification. The proposed semi-supervised learning paradigm may provide an alternative data-driven machine learning approach to effectively utilize a large number of unlabeled data, which can be readily collected under naturalistic (operational) task environments in many real-world applications.

• Zhang, Jianhua; Chen, Peng; Nichele, Stefano & Yazidi, Anis (2020). Emotion Recognition Using Time-frequency Analysis of EEG Signals and Machine Learning. In Huang, Tingwen (Eds.), Proceedings of IEEE Symposium Series on Computational Intelligence (SSCI2019). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9781728124858. p. 403–408. doi: 10.1109/SSCI44817.2019.9003057. Full text in Research Archive
• Pontes-Filho, Sidney; Lind, Pedro; Yazidi, Anis; Zhang, Jianhua; Hammer, Hugo Lewi & Mello, Gustavo [Show all 9 contributors for this article] (2020). EvoDynamic: A Framework for the Evolution of Generally Represented Dynamical Systems and Its Application to Criticality. In Castillo, Pedro A.; Laredo, Juan Luis Jiménez & Vega, Francisco Fernández de (Ed.), Applications of Evolutionary Computation. Springer. ISSN 9783030437220. p. 133–148. doi: 10.1007/978-3-030-43722-0_9. Full text in Research Archive Show summary

  Dynamical systems possess a computational capacity that may be exploited in a reservoir computing paradigm. This paper presents a general representation of dynamical systems which is based on matrix multiplication. That is similar to how an artificial neural network (ANN) is represented in a deep learning library and its computation can be faster because of the optimized matrix operations that such type of libraries have. Initially, we implement the simplest dynamical system, a cellular automaton. The mathematical fundamentals behind an ANN are maintained, but the weights of the connections and the activation function are adjusted to work as an update rule in the context of cellular automata. The advantages of such implementation are its usage on specialized and optimized deep learning libraries, the capabilities to generalize it to other types of networks and the possibility to evolve cellular automata and other dynamical systems in terms of connectivity, update and learning rules. Our implementation of cellular automata constitutes an initial step towards a more general framework for dynamical systems. Our objective is to evolve such systems to optimize their usage in reservoir computing and to model physical computing substrates. Furthermore, we present promising preliminary results toward the evolution of complex behavior and criticality using genetic algorithm in stochastic elementary cellular automata.

• Zhang, Jianhua; Yin, Zhong; Chen, Peng & Nichele, Stefano (2020). Emotion recognition using multi-modal data and machine learning techniques: A tutorial and review. Information Fusion. ISSN 1566-2535. 59, p. 103–126. doi: 10.1016/j.inffus.2020.01.011. Full text in Research Archive Show summary

  In recent years, the rapid advances in machine learning (ML) and information fusion has made it possible to endow machines/computers with the ability of emotion understanding, recognition, and analysis. Emotion recognition has attracted increasingly intense interest from researchers from diverse fields. Human emotions can be recognized from facial expressions, speech, behavior (gesture/posture) or physiological signals. However, the first three methods can be ineffective since humans may involuntarily or deliberately conceal their real emotions (so-called social masking). The use of physiological signals can lead to more objective and reliable emotion recognition. Compared with peripheral neurophysiological signals, electroencephalogram (EEG) signals respond to fluctuations of affective states more sensitively and in real time and thus can provide useful features of emotional states. Therefore, various EEG-based emotion recognition techniques have been developed recently. In this paper, the emotion recognition methods based on multi-channel EEG signals as well as multi-modal physiological signals are reviewed. According to the standard pipeline for emotion recognition, we review different feature extraction (e.g., wavelet transform and nonlinear dynamics), feature reduction, and ML classifier design methods (e.g., k-nearest neighbor (KNN), naive Bayesian (NB), support vector machine (SVM) and random forest (RF)). Furthermore, the EEG rhythms that are highly correlated with emotions are analyzed and the correlation between different brain areas and emotions is discussed. Finally, we compare different ML and deep learning algorithms for emotion recognition and suggest several open problems and future research directions in this exciting and fast-growing area of AI.

• Pontes-Filho, Sidney; Lind, Pedro; Yazidi, Anis; Zhang, Jianhua; Hammer, Hugo Lewi & Mello, Gustavo [Show all 9 contributors for this article] (2020). A neuro-inspired general framework for the evolution of stochastic dynamical systems: Cellular automata, random Boolean networks and echo state networks towards criticality. Cognitive Neurodynamics. ISSN 1871-4080. doi: 10.1007/s11571-020-09600-x. Full text in Research Archive Show summary

  Although deep learning has recently increased in popularity, it suffers from various problems including high computational complexity, energy greedy computation, and lack of scalability, to mention a few. In this paper, we investigate an alternative brain-inspired method for data analysis that circumvents the deep learning drawbacks by taking the actual dynamical behavior of biological neural networks into account. For this purpose, we develop a general framework for dynamical systems that can evolve and model a variety of substrates that possess computational capacity. Therefore, dynamical systems can be exploited in the reservoir computing paradigm, i.e., an untrained recurrent nonlinear network with a trained linear readout layer. Moreover, our general framework, called EvoDynamic, is based on an optimized deep neural network library. Hence, generalization and performance can be balanced. The EvoDynamic framework contains three kinds of dynamical systems already implemented, namely cellular automata, random Boolean networks, and echo state networks. The evolution of such systems towards a dynamical behavior, called criticality, is investigated because systems with such behavior may be better suited to do useful computation. The implemented dynamical systems are stochastic and their evolution with genetic algorithm mutates their update rules or network initialization. The obtained results are promising and demonstrate that criticality is achieved. In addition to the presented results, our framework can also be utilized to evolve the dynamical systems connectivity, update and learning rules to improve the quality of the reservoir used for solving computational tasks and physical substrate modeling.

• Mello, Gustavo; Pontes-Filho, Sidney; Sandvig, Ioanna; Valderhaug, Vibeke Devold; Zouganeli, Evi & Ramstad, Ola Huse [Show all 8 contributors for this article] (2020). Method to Obtain Neuromorphic Reservoir Networks from Images of in Vitro Cortical Networks. In Huang, Tingwen (Eds.), Proceedings of IEEE Symposium Series on Computational Intelligence (SSCI2019). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9781728124858. p. 2360–2366. doi: 10.1109/SSCI44817.2019.9002741. Full text in Research Archive Show summary

  In the brain, the structure of a network of neurons defines how these neurons implement the computations that underlie the mind and the behavior of animals and humans. Provided that we can describe the network of neurons as a graph. We can employ methods from graph theory to investigate its structure or use cellular automata to mathematically assess its function. Additionally, these graphs can provide biologically plausible designs for networks, which can be integrated as reservoirs to support computing. Although, software for the analysis of graphs and cellular automata are widely available. Graph extraction from the image of networks of brain cells remains difficult. Nervous tissue is heterogeneous, and differences in anatomy may reflect relevant differences in function. Here we introduce a deep learning based toolbox to extracts graphs from images of brain tissue. This toolbox provides an easy- to-use framework allowing system neuroscientists to generate graphs based on images of brain tissue by combining methods from image processing, deep learning, and graph theory. The goals are to simplify the training and usage of deep learning methods for computer vision and facilitate its integration into graph extraction pipelines. In this way, the toolbox provides an alternative to the required laborious manual process of tracing, sorting and classifying. We expect to democratize the machine learning methods to a wider community of users beyond the computer vision experts and improve the time-efficiency of graph extraction from large brain image datasets, which may lead to further understanding of the human mind.

• Valderhaug, Vibeke Devold; Ramstad, Ola Huse; Wijdeven, Rosanne Francisca van de; Heiney, Kristine; Nichele, Stefano & Sandvig, Axel [Show all 7 contributors for this article] (2020). Structural and functional alterations associated with the LRRK2 G2019S mutation revealed in structured human neural networks. BioRxiv. ISSN 0362-4331. doi: 10.1101/2020.05.02.073726. Full text in Research Archive Show summary

  Mutations in the LRRK2 gene have been widely linked to Parkinson’s disease. The G2019S variant has been shown to contribute uniquely to both familial and sporadic forms of the disease. LRRK2-related mutations have been extensively studied, yet the wide variety of cellular and network events directly or indirectly related to these mutations remain poorly understood. In this study, we structured multi-nodal human neural networks carrying the G2019S mutation using custom-designed microfluidic chips coupled to microelectrode-arrays. By applying live imaging approaches, immunocytochemistry and computational modelling, we have revealed alterations in both the structure and function of the resulting neural networks when compared to controls. We provide first evidence of increased neuritic density associated with the G2019S LRRK2 mutation, while previous studies have found either a strong decrease, or no change, compared to controls. Additionally, we corroborate previous findings regarding increased baseline network activity compared to control neural networks. Furthermore, we can reveal additional network alterations attributable to the specific mutation by selectively inducing transient overexcitation to confined parts of the structured multi-nodal networks. These alterations, which we were able to capture both at the micro- and mesoscale manifested as differences in relative network activity and correlation, as well as in mitochondria activation, neuritic remodelling, and synaptic alterations. Our study thus provides important new insights into early signs of neural network pathology significantly expanding upon the current knowledge relating to the G2019S Parkinson’s disease mutation.

• Valderhaug, Vibeke Devold; Heiney, Kristine; Ramstad, Ola Huse; Bråthen, Geir; Kuan, Wei-Li & Nichele, Stefano [Show all 8 contributors for this article] (2020). Criticality as a measure of developing proteinopathy in engineered human neural networks. BioRxiv. ISSN 0362-4331. doi: 10.1101/2020.05.03.074666. Full text in Research Archive Show summary

  A patterned spread of proteinopathy represents a common characteristic of many neurodegenerative diseases. In Parkinson’s disease (PD), misfolded forms of alpha-synuclein proteins aggregate and accumulate in hallmark pathological inclusions termed Lewy bodies and Lewy neurites, which seems to affect selectively vulnerable neuronal populations and propagate within interconnected neuronal networks. Research findings suggest that these proteinopathic inclusions are present at very early timepoints in disease development, even before strong behavioural symptoms of dysfunction arise, but that these underlying pathologies might be masked by homeostatic processes working to maintain the function of the degenerating neural circuits. This study investigates whether inducing the PD-related alpha-synuclein pathology in engineered human neural networks can be associated with changes in network function, and particularly with network criticality states. Self-organised criticality represents the critical point between resilience against perturbation and adaptational flexibility, which appears to be a functional trait in self-organising neural networks, both in vitro and in vivo. By monitoring the developing neural network activity through the use of multielectrode arrays (MEAs) for a period of three weeks following proteinopathy induction, we show that although this developing pathology is not clearly manifest in standard measurements of network function, it may be discerned by differences in network criticality states.

• Heiney, Kristine; Ramstad, Ola Huse; Sandvig, Ioanna; Sandvig, Axel & Nichele, Stefano (2020). Assessment and manipulation of the computational capacity of in vitro neuronal networks through criticality in neuronal avalanches. In Huang, Tingwen (Eds.), Proceedings of IEEE Symposium Series on Computational Intelligence (SSCI2019). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9781728124858. p. 246–253. doi: 10.1109/SSCI44817.2019.9002693. Full text in Research Archive Show summary

  In this work, we report the preliminary analysis of the electrophysiological behavior of in vitro neuronal networks to identify when the networks are in a critical state based on the size distribution of network-wide avalanches of activity. The results presented here demonstrate the importance of selecting appropriate parameters in the evaluation of the size distribution and indicate that it is possible to perturb networks showing highly synchronized—or supercritical—behavior into the critical state by increasing the level of inhibition in the network. The classification of critical versus non-critical networks is valuable in identifying networks that can be expected to perform well on computational tasks, as criticality is widely considered to be the state in which a system is best suited for computation. In addition to enabling the identification of networks that are well-suited for computation, this analysis is expected to aid in the classification of networks as perturbed or healthy. This study is part of a larger research project, the overarching aim of which is to develop computational models that are able to reproduce target behaviors observed in in vitro neuronal networks. These models will ultimately be used to aid in the realization of these behaviors in nanomagnet arrays to be used in novel computing hardwares.

• Fladby, Torgeir; Haugerud, Hårek; Nichele, Stefano; Begnum, Kyrre & Yazidi, Anis (2020). Evading a Machine Learning-based Intrusion Detection System through Adversarial Perturbations. In NN, NN (Eds.), RACS '20: Proceedings of the International Conference on Research in Adaptive and Convergent Systems. Association for Computing Machinery (ACM). ISSN 9781450380256. p. 161–166. doi: 10.1145/3400286.3418252. Full text in Research Archive Show summary

  Machine-learning based Intrusion Detection and Prevention Systems provide significant value to organizations because they can efficiently detect previously unseen variations of known threats, new threats related to known malware or even zero-day malware, unrelated to any other known threats. However, while such systems prove invaluable to security personnel, researchers have observed that data subject to inspection by behavioral analysis can be perturbed in order to evade detection. We investigated the use of adversarial techniques for adapting the communication patterns between botnet malware and control unit in order to evaluate the robustness of an existing Network Behavioral Analysis solution. We implemented a packet parser that let us extract and edit certain properties of network flows and automated an approach for conducting a grey-box testing scheme of Stratosphere Linux IPS. As part of our implementation, we provided several techniques for providing perturbation to network flow parameters, including a Simultaneous Perturbation Stochastic Approximation method, which was able to produce sufficiently perturbed network flow patterns while adhering to an underlying objective function. Our results showed that network flow parameters could indeed be perturbed to ultimately enable evasion of intrusion detection based on the detection models that were used with the Intrusion Detection System. Additionally, we demonstrated that it was possible to combine evading detection with techniques for optimization problems that aimed to minimize the magnitude of perturbation to network flows, effectively enabling adaptive network flow behavior.

View all works in NVA

• Nehaniv, Chrystopher L.; Lewis, Peter R.; Nichele, Stefano; Čejková, Jitka; Salge, Christoph & Khan, Imran Qayym [Show all 7 contributors for this article] (2026). ALIFE 2026: Proceedings of the 2026 Artificial Life Conference. MIT Press. doi: https:/direct.mit.edu/isal/isal2026/volume/38. Full text in Research Archive
• Nichele, Stefano (2023). IEEE Conference Proceedings: 2023 3rd International Conference on Applied Artificial Intelligence (ICAPAI). IEEE (Institute of Electrical and Electronics Engineers). ISBN 9798350328929. 100 p. Full text in Research Archive
• Nichele, Stefano; Aamodt, Jonas Moræus; Misra, Sanjay & Mölder, Anicka (2023). 2023 3rd International Conference on Applied Artificial Intelligence (ICAPAI). IEEE (Institute of Electrical and Electronics Engineers). ISBN 9798350328929. Full text in Research Archive

View all works in NVA

• Guichard, Etienne; Reimers, Felix; Kvalsund, Mia-Katrin; Lepperød, Mikkel & Nichele, Stefano (2025). Engramnca: a neural cellular automaton model of memory transfer. doi: 10.1162/ISAL.a.831. Full text in Research Archive
• Halnes, Geir; Javaid, Anam; Solvang, Michael; Nichele, Stefano; Riegler, Michael & Singstad, Bjørn-Jostein [Show all 82 contributors for this article] (2025). Proceedings of NORA’s annual conference 2025. Nordic Machine Intelligence (NMI). 5(3), p. 1–14. doi: 10.5617/nmi.12551. Full text in Research Archive
• Hudcová, Barbora; Mikolov, Tomas & Nichele, Stefano (2025). Studying Encoder–Decoder Relation Between Cellular Automata to Uncover Their Computational Structure. In Adamatzky, Andrew; Sirakoulis, Georgios & Martinez, Genaro (Ed.), Advances in Cellular Automata Volume 2: Computation and Applications. Springer. ISSN 9783031810978. p. 1–32. doi: 10.1007/978-3-031-81097-8_1. Full text in Research Archive
• Valderhaug, Vibeke Devold; Ramstad, Ola Huse; Wijdeven, Rosanne Francisca van de; Heiney, Kristine; Nichele, Stefano & Sandvig, Ioanna [Show all 7 contributors for this article] (2024). MICRO- AND MESOSCALE ASPECTS OF NEURODEGENERATION IN MULT-NODAL HUMAN NEURAL NETWORKS CARRYING THE LRRK2 G2019S MUTATION. Full text in Research Archive
• Lindell, Trym; Ramstad, Ola Huse; Sandvig, Axel; sandvig, Ionna & Nichele, Stefano (2024). Stability of Input Representations in Biological Neural Network Reservoirs. Full text in Research Archive
• Ramstad, Ola Huse; Wijdeven, Rosanne Francisca van de; Heiney, Kristine; Nichele, Stefano; Sandvig, Axel & Sandvig, Ioanna (2024). MICRO- AND MESOSCALE ASPECTS OF NEURODEGENERATION IN MULT-NODAL HUMAN NEURAL NETWORKS CARRYING THE LRRK2 G2019S MUTATION. Full text in Research Archive
• Nichele, Stefano (2024). In-vitro models for NeuroAI. Full text in Research Archive Show summary

  In-vitro neural network have been shown to recapitulate important mechanisms also seen in brains. The advancement of micro-electrode array (MEA) technologies allows to (1) investigate biological neural networks as artificial intelligent systems (either by training the neural substrate directly, or by relying on its dynamics with a reservoir computing paradigm) therefore contributing to advances in Neuro-inspired AI, and (2) test hypotheses in artificial neural network models and then validate them in the dish, therefore contributing to the understanding of biological neural network processes and the emergence of intelligence. In this talk, I will review recent work in in-vitro AI systems, present ongoing work, and discuss possible implications of this approach.

• Ramstad, Ola Huse; Wijdeven, Rosanne Francisca van de; Heiney, Kristine; Nichele, Stefano; Sandvig, Axel & Sandvig, Ioanna (2024). MICRO- AND MESOSCALE ASPECTS OF NEURODEGENERATION IN MULTI-NODAL HUMAN NEURAL NETWORKS CARRYING THE LRRK2 G2019S MUTATION. Full text in Research Archive
• Reimers, Felix Simon; Ramstad, Ola Huse; Sæbø, Solve; Sandvig, Axel; Sandvig, Ioanna & Nichele, Stefano (2024). Examination of Computational Performance of Biological Neural Networks with Reservoir Computing. Full text in Research Archive
• Nichele, Stefano (2024). What’s a Machine? From In-Silico to In-Vitro AI. Full text in Research Archive
• Reimers, Felix Simon; Ramstad, Ola Huse; Sæbø, Solve; Sandvig, Axel; Sandvig, Ioanna & Nichele, Stefano (2024). Performance of C.Elegans Connectomes as Computational Reservoirs. Full text in Research Archive
• Heiney, Kristine Anne; Jozsa, Monika; Rule, Michael E.; Nichele, Stefano & O'Leary, Timothy (2023). Rate of representational drift correlates with information theoretic measures of neural and behavioural coupling. Full text in Research Archive
• Pontes-Filho, Sidney; Nichele, Stefano & Lepperød, Mikkel (2023). Critical Neural Cellular Automata. Full text in Research Archive
• Reimers, Felix Simon; Jain, Sanyam; Aarati, Shrestha & Nichele, Stefano (2023). Pathfinding Neural Cellular Automata with Local Self-Attention. Full text in Research Archive
• Farner, Jørgen Jensen; Ramstad, Ola Huse; Nichele, Stefano & Heiney, Kristine (2023). Beyond weight plasticity: Local learning with propagation delays in spiking neural networks. Full text in Research Archive
• Jain, Sanyam; Shrestha, Aarati & Nichele, Stefano (2023). Capturing Emerging Complexity in Lenia. Full text in Research Archive
• Reimers, Felix Simon; Jain, Sanyam; Shrestha, Aarati & Nichele, Stefano (2023). Pathfinding Neural Cellular Automata with Local Self-Attention. Full text in Research Archive
• Heiney, Kristine Anne; Józsa, Mónika; Rule, Michael; Nichele, Stefano & O'Leary, Timothy (2023). Encoding of behaviour by pairwise neuronal interactions correlates with representational drift. Full text in Research Archive
• Heiney, Kristine Anne; Józsa, Mónika; Rule, Michael; Nichele, Stefano & O'Leary, Timothy (2023). Neural correlations, information, and representational drift. Full text in Research Archive
• Heiney, Kristine Anne; Józsa, Mónika; Rule, Michael; Nichele, Stefano & O'Leary, Timothy (2022). Encoding of behaviour by pairwise neuronal interactions correlates with representational drift. Full text in Research Archive
• Nadizar, Giorgia; Medvet, Eric; Ramstad, Ola Huse; Nichele, Stefano; Pellegrino, Felice Andrea & Zullich, Marco (2022). Erratum: Merging pruning and neuroevolution: Towards robust and efficient controllers for modular soft robots (The Knowledge Engineering Review DOI: 10.1017/S0269888921000151). Knowledge engineering review (Print). ISSN 0269-8889. 37. doi: 10.1017/S0269888922000017. Full text in Research Archive
• Heiney, Kristine Anne; Rule, Michael; Józsa, Mónika; Nichele, Stefano & O'Leary, Timothy (2022). Changing to stay the same. Full text in Research Archive
• Filho, Sidney Pontes; Walker, Kathryn; Najarro, Elias; Nichele, Stefano & Risi, Sebastian (2022). A Unified Substrate for Body-Brain Co-evolution. Full text in Research Archive Show summary

  A successful development of a complex multicellular organism took millions of years of evolution. The genome of such a multicellular organism guides the development of its body from a single cell, including its control system. Our goal is to imitate this natural process using a single neural cellular automaton (NCA) as a genome for modular robotic agents. In the introduced approach, called Neural Cellular Robot Substrate (NCRS), a single NCA guides the growth of a robot and the cellular activity which controls the robot during deployment. We also introduce three benchmark environments, which test the ability of the approach to grow different robot morphologies. We evolve the NCRS with covariance matrix adaptation evolution strategy (CMA-ES), and covariance matrix adaptation MAP-Elites (CMA-ME) for quality diversity and observe that CMA-ME generates more diverse robot morphologies with higher fitness scores. While the NCRS is able to solve the easier tasks in the benchmark, the success rate reduces when the difficulty of the task increases. We discuss directions for future work that may facilitate the use of the NCRS approach for more complex domains.

• Heiney, Kristine Anne; Ramstad, Ola Huse; Fiskum, Vegard; Sandvig, Axel; Sandvig, Ioanna & Nichele, Stefano (2022). Neuronal avalanche dynamics and functional connectivity as indicators of the computational capacity of in vitro neuronal networks. Full text in Research Archive Show summary

  The brain has long been a source of inspiration for the development of novel computational methods and architectures. Most notably, modern machine learning (ML) and deep learning (DL) algorithms in the field of artificial intelligence (AI) are based on an artificial neuron model, and such approaches have enabled us to process data on a previously unimaginable scale. However, current ML and DL approaches are computationally expensive and task-specific, suggesting we have much yet to learn from the efficiency and learning capabilities of the brain. In this work, we evaluated the dynamics and functional connectivity of networks of cortical neurons as they matured over approximately two months in vitro. The electrophysiological activity of the networks was captured daily using microelectrode arrays (MEAs). To assess the dynamics, we identified patterns of activity termed ‘neuronal avalanches’ and used this activity to compute the branching ratio and complexity of the observed activity. The branching ratio describes how activity is propagated through the network and allows the activity to be classified into different dynamical regimes, while the complexity provides a measure that is maximized when activity indicates a balance between integration and segregation in the network. Graphs representing the functional connectivity of the network before were extracted from their electrophysiological behavior, and these graphs were used to evaluate the relationship between the avalanche dynamics and the connectivity of the network. This work serves as a starting point for the data-driven development of biologically plausible models to inform AI algorithms. Future work will involve information theoretical analysis of the network behavior as well as investigations into the role of plasticity in determining the dynamical state. By evaluating the interplay between computation and connectivity, we hope to drive the development of novel powerful AI models.

• Nichele, Stefano; Sæbo, Solve; Pettersen, Klas Henning & Lepperød, Mikkel Elle (2022). NeuroAI - A strategic opportunity for Norway and Europe. Nordic Machine Intelligence (NMI). Full text in Research Archive
• Heiney, Kristine Anne; Józsa, Mónika; Rule, Michael; Nichele, Stefano & O'Leary, Timothy (2022). Encoding of behaviour by pairwise neuronal interactions correlates with representational drift. Full text in Research Archive Show summary

  In some brain areas, populations of neurons can change their behavioural tuning gradually over days. At the single cell level, the rate of this drift varies considerably, raising the question: Does the amount of information contained in interactions between cells predict how much they drift? We reanalysed data from Driscoll et al. (2017), in which the activity of mouse posterior parietal cortex (PPC) neurons was tracked over weeks as mice performed a learned navigation task. We quantified the informativeness of pairwise interactions by measuring the mutual information between behaviour and pairwise neuronal responses, relative to a conditionally shuffled baseline. This conditional shuffling preserves the tuning of single cells while destroying noise correlations or higher-order dependencies in a pair. The amount of pairwise information relative to this baseline can be synergistic (positive) or redundant (negative). After thresholding these measures, we found that neurons with high average pairwise redundancy typically also showed greater average pairwise synergy. Furthermore, the average redundancy was positively correlated with tuning stability, more strongly than was single-cell informativeness. This suggests that redundant subpopulations of neurons form a more stable ‘backbone’ that is less susceptible to drift and more informative about behaviour on average. Thus, redundant coupling is more predictive of stable tuning than the information conveyed about behaviour at the single cell level.

• Farner, Jørgen Jensen; Weydahl, Håkon; Jahren, Ruben; Ramstad, Ola Huse; Nichele, Stefano & Heiney, Kristine Anne (2021). Evolving spiking neuron cellular automata and networks to emulate in vitro neuronal activity. Full text in Research Archive Show summary

  Neuro-inspired models and systems have great potential for applications in unconventional computing. Often, the mechanisms of biological neurons are modeled or mimicked in simulated or physical systems in an attempt to harness some of the computational power of the brain. However, the biological mechanisms at play in neural systems are complicated and challenging to capture and engineer; thus, it can be simpler to turn to a data-driven approach to transfer features of neural behavior to artificial substrates. In the present study, we used an evolutionary algorithm to produce spiking neural systems that emulate the patterns of behavior of biological neurons in vitro. The aim of this approach was to develop a method of producing models capable of exhibiting complex behavior that may be suitable for use as computational substrates. Our models were able to produce a level of network-wide synchrony and showed a range of behaviors depending on the target data used for their evolution, which was from a range of neuronal culture densities and maturities. The genomes of the top-performing models indicate the excitability and density of connections in the model play an important role in determining the complexity of the produced activity.

• Nichele, Stefano (2021). Towards a less artificial Artificial Intelligence. Full text in Research Archive
• Heiney, Kristine; Ramstad, Ola Huse; Fiskum, Vegard; Sandvig, Axel; Sandvig, Ioanna & Nichele, Stefano (2021). From in vitro to computational models of the brain: What can criticality tell us about neural computation? Full text in Research Archive Show summary

  It has been hypothesized that some biological systems, including parts of the brain, may operate near the critical point of a phase transition, delicately balanced between ordered and disordered behavior. Criticality maximizes a number of properties that are favorable for computation, such as dynamic range, information transmission, and the number of input–output mappings; put more simply, systems at criticality are well-suited to take inputs for some computational problem, perform transformations on the inputs as they propagate through the system, and give meaningful outputs. In this talk, I will present our ongoing work on evaluating the closeness of in vitro neuronal networks to criticality and discuss the implications of our findings for understanding neural computation and the development of bioinspired computing methods and hardware. To determine the closeness of networks to criticality, the spatiotemporal scaling behavior of neuronal avalanches is observed, and we compare this behavior across networks with different seeding densities as they mature. The functional connectivity is also obtained for these networks and the small-worldness assessed using the small-world propensity measure. Additionally, the excitation-to-inhibition ratio was chemically manipulated to evaluate its effect on the network dynamics. Although none of our networks tended to mature toward the critical state, we observed consistent differences across the two considered densities in the avalanche scaling and branching ratio indicative of a greater dominance of network bursts in the higher-density networks. Additionally, the lower-density networks tended to show small-world organization in their functional connectivity. Chemical perturbation to increase inhibition brought the networks closer to criticality, and the higher-density networks showed greater disruption to both their functional activity and avalanche behavior under chemical perturbation. From these findings, we aim to find connectivity and activity patterns correlated with closeness to criticality so that we may emulate these features in models for bioinspired computation. Future work will also involve evaluating measures of information processing in these networks as indicators of the suitability of these networks for computation.

• Heiney, Kristine; Tufte, Gunnar & Nichele, Stefano (2021). On Artificial Life and Emergent Computation in Physical Substrates. Full text in Research Archive Show summary

  In living systems, we often see the emergence of the ingredients necessary for computation---the capacity for information transmission, storage, and modification---begging the question of how we may exploit or imitate such biological systems in unconventional computing applications. What can we gain from artificial life in the advancement of computing technology? Artificial life provides us with powerful tools for understanding the dynamic behavior of biological systems and capturing this behavior in manmade substrates. With this approach, we can move towards a new computing paradigm concerned with harnessing emergent computation in physical substrates not governed by the constraints of Moore's law and ultimately realize massively parallel and distributed computing technology. In this paper, we argue that the lens of artificial life offers valuable perspectives for the advancement of high-performance computing technology. We first present a brief foundational background on artificial life and some relevant tools that may be applicable to unconventional computing. Two specific substrates are then discussed in detail: biological neurons and ensembles of nanomagnets. These substrates are the focus of the authors' ongoing work, and they are illustrative of the two sides of the approach outlined here---the close study of living systems and the construction of artificial systems to produce life-like behaviors. We conclude with a philosophical discussion on what we can learn from approaching computation with the curiosity inherent to the study of artificial life. The main contribution of this paper is to present the great potential of using artificial life methodologies to uncover and harness the inherent computational power of physical substrates toward applications in unconventional high-performance computing.

• Øye, Olav-Johan; Nichele, Stefano & Normann, Maria Storvig (2021). Slik kan biologien inspirere kunstig intelligens. [Internet]. forskning.no. Full text in Research Archive Show summary

  Mennesker løser kompliserte oppgaver med svært lite energiforbruk. Kunstig intelligens bruker mye strøm.

• Øye, Olav-Johan; Normann, Maria Storvig & Nichele, Stefano (2021). Slik kan biologien inspirere kunstig intelligens. [Internet]. oslomet.no. Full text in Research Archive Show summary

  Mennesker løser kompliserte oppgaver med svært lite energiforbruk, mens kunstig intelligens bruker mye strøm. Hvordan kan det bli hvis kunstig intelligens lærer av biologien?

• Filho, Sidney Pontes; Dahl, Annelene Gulden; Nichele, Stefano & Mello, Gustavo (2021). A Deep Learning-Based Tool for Automatic Brain Extraction from Functional Magnetic Resonance Images of Rodents. Full text in Research Archive Show summary

  Removing skull artifacts from functional magnetic images (fMRI) is a well understood and frequently encountered problem. Because the fMRI field has grown mostly due to human studies, many new tools were developed to handle human data. Nonetheless, these tools are not equally useful to handle the data derived from animal studies, especially from rodents. This represents a major problem to the field because rodent studies generate larger datasets from larger populations, which implies that preprocessing these images manually to remove the skull becomes a bottleneck in the data analysis pipeline. In this study, we address this problem by implementing a neural network-based method that uses a U-Net architecture to segment the brain area into a mask and removing the skull and other tissues from the image. We demonstrate several strategies to speed up the process of generating the ground-truth of the dataset using watershedding, and several strategies for data augmentation that allowed to train robustly the U-Net to perform the segmentation. Finally, we deployed the trained network freely available.

• Filho, Sidney Pontes & Nichele, Stefano (2021). A Conceptual Bio-Inspired Framework for the Evolution of Artificial General Intelligence. Full text in Research Archive Show summary

  In this work, a conceptual bio-inspired parallel and distributed learning framework for the emergence of general intelligence is proposed, where agents evolve through environmental rewards and learn throughout their lifetime without supervision, i.e., self-learning through embodiment. The chosen control mechanism for agents is a biologically plausible neuron model based on spiking neural networks. Network topologies become more complex through evolution, i.e., the topology is not fixed, while the synaptic weights of the networks cannot be inherited, i.e., newborn brains are not trained and have no innate knowledge of the environment. What is subject to the evolutionary process is the network topology, the type of neurons, and the type of learning. This process ensures that controllers that are passed through the generations have the intrinsic ability to learn and adapt during their lifetime in mutable environments. We envision that the described approach may lead to the emergence of the simplest form of artificial general intelligence.

• Nichele, Stefano (2021). Tim Taylor and Alan Dorin: Rise of the self-replicators—early visions of machines, AI and robots that can reproduce and evolve. Genetic Programming and Evolvable Machines. ISSN 1389-2576. doi: 10.1007/s10710-021-09398-5. Full text in Research Archive
• Heiney, Kristine; Ramstad, Ola Huse; Pontes-Filho, Sidney; Glover, Tom; Lindell, Trym & Farner, Jørgen Jensen [Show all 8 contributors for this article] (2021). Bridging the computational gap: From biological to artificial substrates. Full text in Research Archive
• Vie, Knut Jørgen; Thorstensen, Erik; Gjefsen, Mads Dahl; Nichele, Stefano & Tufte, Gunnar (2020). Ethical Tissues. Full text in Research Archive
• Tapio, Hege; Bergaust, Kristin; Nichele, Stefano; Christensen-Scheel, Boel & Wettre, Mikkel (2020). FeLT - Futures of Living Technologies, Project Launch Symposium. Full text in Research Archive Show summary

  Launch of FeLT, an interdisciplinary artistic research project started in October 2020. Presentations live and pre-recorded by participants and collaborators: Baltan Laboratories (NL), Olga Mink, Bioartsociety (FI)Erich Berger, Cesar & Lois (BR/US) Lucy HG Solomon Coalesce Center for Biological Art(US) Paul Vanouse, MetaMorf(NO), Espen Gangvik, SEADS collective(INT), Angelo Vermeulen, Symbiotica(AU) RIXC(LV), Rasa Smite Maria Castellanos (ES), Mikkel Wettre(NO), Hege Tapio(NO), Haakon H Roen(NO) Trym A. Eidsvik(NO)

• Pontes-Filho, Sidney; Lind, Pedro; Yazidi, Anis; Zhang, Jianhua; Hammer, Hugo Lewi & Mello, Gustavo [Show all 9 contributors for this article] (2020). EvoDynamic: A Framework for the Evolution of Generally Represented Dynamical Systems and Its Application to Criticality. Full text in Research Archive
• Heiney, Kristine; Ramstad, Ola Huse; Sandvig, Axel; Sandvig, Ioanna & Nichele, Stefano (2020). Manipulating the dynamic state of in vitro neuronal networks. Full text in Research Archive Show summary

  The dynamic state of neuronal networks is influenced by a wide range of functional and structural factors, including network connectivity, synaptic function, and the balance of excitation and inhibition in the network. Additionally, it has been shown that only some dissociated networks self-organize into the critical state as they are left to mature in vitro, with many ultimately showing the tightly synchronized behavior indicative of a supercritical state. To explore the possibility of manipulating dissociated networks into the critical state by chemical intervention, we increased inhibition in the network by adding -aminobutyric acid (GABA) at day in vitro (DIV) 51. Prior to perturbation with GABA, the networks showed large network-wide bursts with close synchrony, and the size distribution of neuronal avalanches was bimodal. After perturbation, the synchrony was broken, and the avalanche distribution followed a power law, one of the hallmarks of criticality. This finding indicates that it is possible to use chemical perturbation in dissociated networks to manipulate the dynamic state. This will allow for future comparative experimental studies on the characteristics and computational capabilities of networks in different dynamical regimes, shedding light on the mechanisms and benefits of critical and near-critical behavior.

• Weir, Janelle Shari; Christiansen, Nicholas; Nichele, Stefano; Lind, Pedro; Sandvig, Axel & Sandvig, Ioanna (2020). From topology to functionality: Does perturbation in in-vitro neural networks drive hub adaptation towards computational efficiency? Full text in Research Archive
• Nichele, Stefano (2020). Novel models and substrates for the emergence of intelligence. Full text in Research Archive
• Pontes-Filho, Sidney; Olsen, Kristoffer; Yazidi, Anis; Halvorsen, Pål; Riegler, Michael & Nichele, Stefano (2020). Towards the evolution of spiking neural networks for self-supervision with neuroplasticity. Full text in Research Archive
• Heiney, Kristine; Ramstad, Ola Huse; Fiskum, Vegard; Sandvig, Axel; Sandvig, Ioanna & Nichele, Stefano (2020). Manipulation of the scaling behavior of neuronal avalanches toward achieving criticality. Full text in Research Archive
• Heiney, Kristine; Ramstad, Ola Huse; Fiskum, Vegard; Sandvig, Ioanna; Sandvig, Axel & Nichele, Stefano (2020). Computational behavior of biological neural networks through the lens of criticality. Full text in Research Archive Show summary

  The human brain is a powerful computational machine, and harnessing some of this power in manmade computing systems would enable advances in terms of computational speed, power consumption, and fault tolerance. One interesting phenomenon theorized to occur in the brain is self-organized criticality. The critical state is a dynamical regime poised between ordered behavior and chaotic behavior and recognized as good for computation. Systems in the critical state have been shown to optimize performance by many measures, including dynamic range, complexity, and information transmission. It has been proposed that the brain self-organizes into this state by striking a balance between excitation and inhibition to optimize its computational capacity. Networks of neurons in vitro can be classified critical if the size of network-wide cascades of activity, termed “neuronal avalanches,” follows a power-law distribution. This method of classification provides a path to identifying networks that are likely suited for computation. Our lab has conducted experiments to assess the dynamical state of networks of dissociated cortical neurons as they mature in vitro. It was found that the networks progressed from a period of low activity to a subcritical state before settling into a supercritical state, characterized by tight network-wide synchronization. It has been reported that not all dissociated networks self-organize into the critical state, begging the question: what can be done to induce criticality in these cases? To adjust the excitation-to-inhibition ratio toward the balance expected at criticality, small volumes of an inhibitory neurotransmitter (GABA) were incrementally added to the networks. After perturbation with GABA, it was observed that the networks showed avalanche activity with power-law scaling, manifested in more complex patterns of activity in the network, indicating that it is possible to manipulate networks into the critical state. The ability to identify the dynamical state of networks and manipulate them into the critical state may enable the comparison of the architectures and features of networks in different states via different information and network theoretical measures. Identifying features correlated with networks at criticality will provide a foundation for the behaviors we wish to replicate as we develop novel neuroscience-inspired artificial neural network models and biologically plausible learning algorithms.

• Christiansen, Nicholas; Nichele, Stefano; Tufte, Gunnar; Sandvig, Ioanna & Sandvig, Axel (2019). Using computer models to infer disease pathology in in vitro models of amyotrophic lateral sclerosis. Full text in Research Archive
• Pontes-Filho, Sidney; Yazidi, Anis; Zhang, Jianhua; Hammer, Hugo Lewi; Mello, Gustavo & Sandvig, I. [Show all 8 contributors for this article] (2019). A general representation of dynamical systems for reservoir computing. Full text in Research Archive
• Heiney, Kristine Anne & Nichele, Stefano (2023). Deciphering and emulating neuronal communication: Population-level computation from elemental interactions. Norges teknisk-naturvitenskapelige universitet. ISSN 9788232671397. Full text in Research Archive Show summary

  This thesis explores computation in neural systems and how we might draw inspiration from it in designing computational systems. The research conducted for this thesis encompasses both neural data analysis and computational modeling. In both of these research components, the overarching goal of this thesis is to elucidate how representations at the system level emerge from the communication of the system components, both to aid in expanding our understanding of neural computation and to support the development of novel computing substrates and algorithms. The first half of the thesis considers how to decipher the language of neurons. We focus on the level of populations of neurons—tens to thousands of interconnected cells, both in vivo and in vitro and evaluate how the interactions of the individual cells lead to computation at the population level. In our in vivo study, we evaluated how a neuron’s interactions with the population relate to the stability of its tuning to a task. In our in vitro work, we probed the dynamics and network organization of dissociated neuronal populations and assessed how these might be tuned to optimize their capacity of the populations to represent and process information. The second half of the thesis examines how the lessons we learned from studying neural systems can inform the development of engineered computational systems. We pursue this using both a bottom-up approach, where the system-level behavior emerges from the designed elements and their interactions, and a top-down approach, where the system behavior is explicitly targeted and the components are iteratively refined until the target is reached. In our bottom-up approach, we de-signed a novel learning rule to locally adjust the speed at which activity propagates in a spiking neural network and successfully used it to train networks for a classification task. In our top-down approach, we used evolutionary algorithms to design models targeting two different fitness goals: the first, to match activity patterns observed in neuronal populations in vitro, and the second, to maximize the separability of output waveforms in a hydrodynamic computational reservoir. Our findings highlight the relevance of understanding the nature of elemental interactions to make sense of computation at the system level and demonstrate the myriad lessons we can learn from neural computation in designing computational systems.

• Filho, Sidney Pontes; Nichele, Stefano; Yazidi, Anis; Zhang, Jianhua; Hammer, Hugo Lewi & Sandvig, Ioanna [Show all 7 contributors for this article] (2023). Optimization of dynamical systems towards criticality and intelligent behavior. Norges teknisk-naturvitenskapelige universitet. ISSN 9788232669530. Full text in Research Archive Show summary

  With the progress of computing power, artificial intelligence (AI) systems are able to achieve outstanding results that surpass human-level performance on some tasks, such as image recognition. Mainstream AI systems are only able to learn one or more specific tasks, and they commonly fail to go beyond what they were trained for. To have a real breakthrough in the field of AI, those systems require adaptability and the capacity to learn general tasks and data distributions. Such characteristics are among the objectives of artificial general intelligence (AGI) research. The inspiration from biology, neuroscience, and complex systems may guide the development of AGI because AI systems are still rigid instead of self-organizing. Therefore, the research reported in this thesis aims at intelligent dynamical systems that resemble the brain, such as the brain’s critical behavior that may allow the cortex to self-organize to criticality in order to increase computational capacity. The plan to accomplishsuch systems encompasses the usage of optimization methods to find adequate interactions of the components in a complex system, how they are connected with each other, and how they adapt those interactions and connections overtime. The dynamical systems investigated in this research are cellular automata, Boolean networks, and recurrent neural networks with abstract neuron models or with more biologically plausible ones (spiking neurons). Themain optimization method applied in these systems is evolutionary computa-tion or artificial evolution. There are three parts in the presented research. Thefirst uses a deep neural network library to evolve dynamical systems towardscriticality. The second part works on the complexification of spiking neuralnetworks with adaptive synapses for solving tasks in mutable environments.The last one involves the application of neural cellular automata for control-ling robots and even developing their morphology with artificial embryogeny.The results of this research attempt to address some unanswered questionsrelated to the practicality, methodology, and benefits of applying dynamicalsystems in AI.

• Heiney, Kristine; Nichele, Stefano; Sandvig, Ioanna & Tufte, Gunnar (2023). Deciphering and emulating neuronal communication: Population-level computation from elemental interactions. NTNU. Full text in Research Archive
• Variengien, Alexandre; Filho, Sidney Pontes; Glover, Tom Eivind & Nichele, Stefano (2021). Towards Self-organized Control: Using Neural Cellular Automata to Robustly Control a Cart-pole Agent. Innovations in Machine Intelligence, Crosslabs, Vol. 1/2021. doi: 10.54854/imi2021.01. Full text in Research Archive Show summary

  Neural cellular automata (Neural CA) are a recent framework used to model biological phenomena emerging from multicellular organisms. In these systems, artificial neural networks are used as update rules for cellular automata. Neural CA are end-to-end differentiable systems where the parameters of the neural network can be learned to achieve a particular task. In this work, we used neural CA to control a cart-pole agent. The observations of the environment are transmitted in input cells while the values of output cells are used as a readout of the system. We trained the model using deep-Q learning where the states of the output cells were used as the Q-value estimates to be optimized. We found that the computing abilities of the cellular automata were maintained over several hundreds of thousands of iterations, producing an emergent stable behavior in the environment it controls for thousands of steps. Moreover, the system demonstrated life-like phenomena such as a developmental phase, regeneration after damage, stability despite a noisy environment, and robustness to unseen disruption such as input deletion.

View all works in NVA