Roland Olsson

• Tran, Thi Thuy Chau; Nguyen, Hoa Thi; Magnusson, Lars Vidar & Olsson, Roland (2026). How Well Do Automatically Evolved Neurons Perform? A Comparative Study in Human Activity Recognition. ESREL 2026. doi: https:/cmsweb.com.sg/rps2prod/esrel2026/epro/pdf/esrel26-p26301.pdf. Full text in Research Archive Show summary

  Neuroevolution, a subfield of AI leveraging evolutionary algorithms, offers an effective method for automatically generating data-adaptive equations that optimize fit for specific datasets while minimizing human design bias. This study investigates the efficacy of the WISDM-ARN, a specialized Adaptive Recurrent Neuron (ARN) synthesized via the Automatic Design of Algorithms through Evolution (ADATE) system, for Human Activity Recognition (HAR). In safety-critical applications, such as healthcare monitoring and industrial safety, the reliability of locomotion classification is paramount to prevent system failures and ensure user protection. We benchmark the evolved WISDM-ARN against three established architectures: Simple RNN, Long Short-Term Memory (LSTM), and Transformers. Experimental results demonstrate that the WISDM-ARN successfully adapts to the unique temporal characteristics of gyroscope data, achieving a testing accuracy of 97\%, which is comparable to the baseline models with resulting 70\%, 95\%, and 92\%, respectively. These findings highlight the potential of neuroevolutionary synthesis as a dependable alternative to manually engineered architectures. By integrating neuroevolution with modern deep learning, the ADATE-generated neuron represents a significant step toward automated reliable neural modeling for human activity recognition tasks and safety-enhanced time-series analysis.

• Nguyen, Hoa Thi; Olsson, Roland & Haugen, Øystein (2024). Automatic Synthesis of Recurrent Neurons for Imitation Learning From CNC Machine Operators. IEEE Open Journal of the Industrial Electronics Society (OJ-IES). 5, p. 91–108. doi: 10.1109/OJIES.2024.3363500. Full text in Research Archive Show summary

  Analyzing time series data in industrial settings demands domain knowledge and computer science expertise to develop effective algorithms. AutoML approaches aim to automate this process, reducing human bias and improving accuracy and cost-effectiveness. This article applies an evolutionary algorithm to synthesize recurrent neurons optimized for specific datasets. This adds another layer to the AutoML framework, targeting the internal structure of neurons. We developed an imitation learning control system for an industry CNC machine to enhance operators' productivity. We specifically examine two recorded operator actions: adjusting the engagement rates for linear feed rate and spindle velocity. We compare the performance of our evolved neurons with support vector machine and four well-established neural network models commonly used for time series data: simple recurrent neural networks, long-short-term-memory, independently recurrent neural networks, and transformers. The results demonstrate that the neurons evolved via the evolutionary approach exhibit lower syntactic complexity than LSTMs and achieve lower error rates than other networks. They yield error rates 270% lower for the first operation action, while the error rates are 20% lower for the second action. We also show that our evolutionary algorithm is capable of creating skip-connections and gating mechanisms adapted to the specific characteristics of our dataset.

• Nguyen, Hoa Thi; Haugen, Øystein & Olsson, Roland (2023). Imitation Learning from Operator Experiences for a Real time CNC Machine Controller. IEEE Conference on Industrial Informatics. ISSN 1935-4576. doi: 10.1109/INDIN51400.2023.10218164. Full text in Research Archive Show summary

  Controlling complex industrial systems can be a challenging task as it requires extensive knowledge and skills that are usually acquired through years of experience. This makes it difficult to program such expertise into machine algorithms. In this paper, we present a use case that demonstrates how we built control algorithms for a CNC machine using historical logging of observations from experts. With the advent of digital technologies, machining parts are now controlled by computer programs that offer high precision and speed. However, unforeseen scenarios can still arise, which demand operators’ attention and intervention, even with finely crafted machine programs. For our experiment, we collected data from a 5-axis Mazak Integrex i500-series CNC machine over a month manufacturing multiple instances of the same part. We collected observational states, which are sensor data that match the information operators receive and output engagement feed rates following the operator’s trajectories. Using behavioral cloning, we built an initial control policy from this data, testing three families of machine learning models: regression models, ensemble methods, and deep neural networks. The results showed that ensemble methods outperformed the baseline model significantly, proving that they have learned useful control patterns. The policies also demonstrated that ML models could eliminate noisy behaviors from operators’ actions. We believe that with interactive demonstrations in the future, these models have the potential to fully mature. Overall, our study demonstrates the feasibility of building control algorithms for complex industrial systems using historical expert demonstrations and machine learning techniques.

• Magnusson, Lars Vidar; Olsson, Roland & Tran, Thi Thuy Chau (2023). Recurrent Neural Networks for Oil Well Event Prediction. IEEE Intelligent Systems. ISSN 1541-1672. 38(2), p. 73–80. doi: 10.1109/MIS.2023.3252446. Full text in Research Archive Show summary

  We have conducted a comparison between three types of recurrent neural networks and their ability to predict anomalies occurring in oil wells using a publicly available dataset. We have included two types of well-known state-of-the-art recurrent neural networks and a new type with neurons evolved specifically for the dataset using automatic programming. We show that the new type of recurrent neuron offers a massive improvement over the state of the art. The overall test accuracy of the new network type is 94.6%, which is an improvement by 18.3%, or 14.6 percentage points. We also show that a network with the new neuron performs better than any other solution proposed for the dataset.

• Geitle, Marius & Olsson, Roland (2019). A New Baseline for Automated Hyper-Parameter Optimization. In Nicosia, Giuseppe; Umeton, Renato; Sciacca, Vincenzo; Pardalos, Panos & Giuffrida, Giovanni (Ed.), Machine Learning,Optimization,and Data Science - 5th International Conference, LOD 2019. Springer Nature. ISSN 9783030375980. p. 521–530. doi: 10.1007/978-3-030-37599-7_43. Full text in Research Archive
• The, Hien Dang Ha; Bianchi, Filippo Maria & Olsson, Roland (2017). Local short term electricity load forecasting: Automatic approaches. Proceedings of the International Joint Conference on Neural Networks. ISSN 2161-4393. p. 4267–4274. doi: 10.1109/IJCNN.2017.7966396. Full text in Research Archive
• Brastein, Ole Magnus; Olsson, Roland; Skeie, Nils-Olav & Lindblad, Thomas (2017). Human Activity Recognition by machine learning methods. NIKT: Norsk IKT-konferanse for forskning og utdanning. ISSN 1892-0713. Full text in Research Archive Show summary

  Recognition of human activity from sensor data is a research field of great potential. Giving autonomous systems the ability to identify what a human subject is doing at a given time is highly useful in many industries, particularly in health care and security monitoring. Our results, using a public domain dataset, show that the state- of-the-art decision tree ensemble algorithm XGBoost gives an accuracy of 94.6% validated on an independent test set. Previously published results using support vector machines (SVM) gave an accuracy of 90.2%. As far as we know, our result is the new state of the art for this data set. Recognition of human activity carries potential privacy concerns, which to some degree constrain the choice of sensor technology. Therefore, systems such as ours which can identify activities from simple inertial sensors, e.g. accelerators and gyroscopes are of particular interest. Data from such inertial sensors are difficult to interpret using mechanistic models; hence the field of Machine Learning is particularly interesting for this application.

• Dang-Ha, The-Hien; Roverso, Davide & Olsson, Roland (2017). Graph of virtual actors (GOVA): A Big Data analytics architecture for IoT. In IEEE, . (Eds.), 2017 IEEE International Conference on Big Data and Smart Computing (BigComp): Proceedings of a meeting held 13-16 February 2017, Jeju Island, South Korea. IEEE (Institute of Electrical and Electronics Engineers). ISSN 9781509030156. p. 162–169. doi: 10.1109/BIGCOMP.2017.7881733. Full text in Research Archive
• Geitle, Marius & Olsson, Roland (2017). Using automatic programming to design improved variants of differential evolution. In Bui, Lam Thu (Eds.), 2017 21st Asia Pacific Symposium on Intelligent and Evolutionary Systems (IES). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9781538607435. p. 13–18. doi: 10.1109/IESYS.2017.8233554. Full text in Research Archive
• Geitle, Marius & Olsson, Roland (2017). Improving competitive differential evolution using automatic programming. In Fei, Xiang (Eds.), 2017 4th International Conference on Systems and Informatics. IEEE (Institute of Electrical and Electronics Engineers). ISSN 9781538611074. p. 538–545. doi: 10.1109/ICSAI.2017.8248350. Full text in Research Archive
• Dang, Ha The Hien; Olsson, Roland & Wang, Hao (2016). Clustering Methods for Electricity Consumers: An Empirical Study in Hvaler - Norway. NIKT: Norsk IKT-konferanse for forskning og utdanning. ISSN 1892-0713. Full text in Research Archive
• Magnusson, Lars Vidar & Olsson, Roland (2016). Improving the Canny edge detector using automatic programming: Improving the filter, 2016 International Conference on Image, Vision and Computing (ICIVC 2016): August 3-5, 2016: Portsmouth, UK. IEEE (Institute of Electrical and Electronics Engineers). ISSN 9781509037551. p. 36–40. doi: 10.1109/ICIVC.2016.7571270. Full text in Research Archive
• Magnusson, Lars Vidar & Olsson, Roland (2016). Improving the Canny edge detector using automatic programming: Improving non-max suppression, GECCO '16: Proceedings of the Genetic and Evolutionary Computation Conference 2016. Association for Computing Machinery (ACM). ISSN 9781450342063. p. 461–468. doi: 10.1145/2908812.2908926. Full text in Research Archive
• Magnusson, Lars Vidar & Olsson, Roland (2016). Improving the Canny Edge Detector Using Automatic Programming: Improving Hysteresis Thresholding. NIKT: Norsk IKT-konferanse for forskning og utdanning. ISSN 1892-0713. doi: 10.1109/icivc.2016.7571270. Full text in Research Archive
• Dang, Ha The Hien; Olsson, Roland & Wang, Hao (2015). The Role of Big Data on Smart Grid Transition. In Liu, Xingang; Hsu, Robert C.; Wang, Peicheng; Xia, Feng; Wang, Yufeng; Dong, Mianxiong & Deng, Yuhui (Ed.), Proceedings of 2015 IEEE International Conference on Smart City (SmartCity 2015). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9781509018925. p. 33–39. doi: 10.1109/SmartCity.2015.43. Full text in Research Archive
• Larsen, Kristin; Magnusson, Lars Vidar & Olsson, Roland (2014). Edge Pixel Classification Using Automatic Programming. NIKT: Norsk IKT-konferanse for forskning og utdanning. ISSN 1892-0713. Full text in Research Archive
• Magnusson, Lars Vidar & Olsson, Roland (2014). Improving Graph-Based Image Segmentation Using Automatic Programming. In Esparcia-Alcázar, Anna I. & Mora, Antonio (Ed.), Applications of Evolutionary Computation. 17th European Conference, EvoApplications 2014, Granada, Spain, April 23-25, 2014. Springer. ISSN 9783662455234. p. 464–475. doi: 10.1007/978-3-662-45523-4_38. Full text in Research Archive
• Olsson, Roland & Løkketangen, Arne (2013). Using automatic programming to generate state-of-the-art algorithms for random 3-SAT. Journal of Heuristics. ISSN 1381-1231. 19(5), p. 819–844. doi: 10.1007/s10732-013-9226-x. Full text in Research Archive Show summary

  The paper focuses on automatic programming and how to synthesize stochastic local search algorithms using automatic design of algorithms through evolution (ADATE). The main goal is to provide support for the hypothesis that automatic programming can generate competitive heuristic algorithms. A well studied and highly optimized SAT solver, G2WSAT, is used as a case study. The results indicate that automatic programming is an effective design tool for heuristics and that there may be many well studied optimization problems that could benefit from the ADATE system and the specification methodology that we describe.

• Olsson, Roland & Vu, Huyen Phuong (2012). Automatic improvement of graph based image segmentation. In Bebis, George; Boyle, Richard; Parvin, Bahram; Koracin, Darko; Fowlkes, Charless; Wang, Sen; Choi, Min-Hyung; Mantler, Stephan; Schulze, Jürgen; Acevedo, Daniel; Mueller, Klaus & Papka, Michael (Ed.), Advances in Visual Computing; 8th International Symposium, ISVC 2012, Rethymnon, Crete, Greece, July 16-18, 2012, Revised Selected Papers, Part I. Springer. ISSN 9783642331787. p. 578–587. doi: 10.1007/978-3-642-33191-6_57. Full text in Research Archive
• Olsson, Roland & Løkketangen, Arne (2011). Improving State-of-the-Art 3-SAT solvers using Automatic Design of Algorithms through Evolution. In Hao, Jin-Kao; Legrand, Pierrick; Collet, Pierre; Monmarché, Nicolas; Lutton, Evelyne & Schoenauer, Marc (Ed.), Artificial Evolution 2011 (Evolution Artificielle 2011): 10th Biennal International Conference on Artificial Evolution: proceedings. Université d’Angers. ISSN 9782953926712. p. 192–197. Full text in Research Archive Show summary

  Automatic generation of code is becoming an increasingly important field. This also includes code for combinatorial optimization. We show how such an automated code generation system, ADATE, is able to improve significantly on a state-of-the-art SAT-solver, G2WSAT, by automatically evolving and improving the original program in a controlled fashion. We have chosen SAT - the Satisfiability Problem - as our target problem, as this is a central problem in many real-world applications, and because improving SAT-solvers might have significant economic impact. Our computational testing clearly shows the code improvement, as well as showing that the resulting code can be generalized and used outside its original learning domain.

• Løkketangen, Arne & Olsson, Roland (2010). Generating meta-heuristic optimization code using ADATE. Journal of Heuristics. ISSN 1381-1231. 16(6), p. 911–930. doi: 10.1007/s10732-009-9119-1. Full text in Research Archive Show summary

  Local Search based meta-heuristic methods for finding good solutions to hard combinatorial optimization problems have attained a lot of success, and a plethora of methods exist, each with its own successes, and also with its own parameter settings and other method-specific details. At the same time, experience is needed to implement highly competitive code, and some of the experience applied is not easy to quantify. ADATE is a system to automatically generate code based on a set of input-output specifications, and can work in vastly different domains. It generates code in a subset of the programming language ML and works by searching for transformations of purely functional ML programs. Code automatically generated by the ADATE system compares with state-of-the-art handcrafted meta-heuristic optimization code. In particular, the programs generated by ADATE target the move selection part of BOOP - Boolean Optimization Problems. The baseline is a highly successful Tabu Search implementation. Comparisons are made for versions running for a limited number of iterations, being suitable for applications needing a short response time. The computational results show that the ADATE system is able to generate highly competitive code that produces more optimal solutions to hard BOOP instances within given iteration limits than the previously published Tabu Search implementation. The automatically generated code also gives new insights into the general design of meta-heuristic mechanisms, and contains novel search mechanisms.

• Berg, Henrik; Olsson, Roland; Rusås, Per-Olav & Jakobsen, Morgan (2009). Automated synthesis of control algorithms from first principles, Proceedings IEEE International Conference on Intelligent Robots and Systems. IEEE (Institute of Electrical and Electronics Engineers). ISSN 9781424438037. p. 2958–2965. Full text in Research Archive
• Berg, Henrik; Rusås, Per-Olav; Olsson, Roland & Jakobsen, Morgan (2009). Synthesis of control algorithms for autonomous vehicles through automatic programming. In Wang, Haiying; Low, Kai Soon; Wei, Kexin & Sun, Junqing (Ed.), Fifth International Conference on Natural Computation, ICNC 2009. IEEE (Institute of Electrical and Electronics Engineers). ISSN 9780769537368. p. 445–453. Full text in Research Archive
• Chilo, José; Horvath, György; Lindblad, Thomas & Olsson, Roland (2009). Electronic nose ovarian carcinoma diagnosis based on machine learning algorithms, Advances in data mining : applications and theoretical aspects. Springer. ISSN 9783642030666. p. 13–23. Full text in Research Archive
• Chilo, José; Horvath, György; Lindblad, Thomas; Olsson, Roland; Redeby, Johan & Roeraade, Johan (2009). A flexible electronic nose for odor discrimination using different methods of classification, 16th IEEE-NPSS Real Time Conference, 2009 : RT '09. IEEE (Institute of Electrical and Electronics Engineers). ISSN 9781424444540. p. 317–320. Full text in Research Archive
• Olsson, Roland; Berg, Henrik; Lindblad, Thomas & Chilo, Jose (2008). Automatic design of pulse coupled neurons for image segmentation. Neurocomputing. ISSN 0925-2312. 71(10.des), p. 1980–1993. Full text in Research Archive Show summary

  Automatic Design of Algorithms through Evolution (ADATE) is a program synthesis system that creates recursive programs in a functional language with automatic invention of recursive help functions and self-adaptive optimization of numerical values. We implement a neuron in a pulse coupled neural network (PCNN) as a recursive function in the ADATE language and then use ADATE to automatically evolve better PCNN neurons for image segmentation. Our technique is generally applicable for automatic improvement of most image processing algorithms and neural computing methods. It may be used either to generally improve a given implementation or to tailor that implementation to a specific problem, which with respect to image segmentation for example can be road following for autonomous vehicles or infrared image segmentation for heat seeking missiles that are to distinguish the heat source of the target from flares.

• Olsson, Roland (2007). Automatic design of algorithms through evolution, Proceedings of the Workshop on Approaches and Applications of Inductive programming (AAIP '07). Faculty of Information Systems and Applied Computer Science. Full text in Research Archive
• Olsson, Roland; Hansen, Stig-Erland; Chilo, Jose & Lindblad, Thomas (2007). Classification of infrasound events with various machine learning techniques, Proceedings from the 4th International Conference on Cybernetics and Information Technologies, Systems and Applications : CITSA 2007. International Institute of Informatics and Systemics. Full text in Research Archive
• Olsson, Roland & Hansen, Stig-Erland (2007). Improving decision tree pruning through automatic programming, NIK 2007 : Norsk informatikkonferanse. Tapir Akademisk Forlag. p. 31–40. Full text in Research Archive
• Olsson, Roland; Hansen, Stig-Erland; Lindblad, Thomas & Chilo, Jose (2007). Comparison of three feature extraction techniques to distinguish between different infrasound signals, Progress in pattern recognition. Springer. p. 75–82. Full text in Research Archive
• Berg, Henrik & Olsson, Roland (2005). Improving the Scalability of Automatic Programming. Lecture Notes in Computer Science (LNCS). ISSN 0302-9743. Full text in Research Archive Show summary

  When automatically constructing large programs using program transformations, the number of possible transformations grows very fast. In this paper, we introduce and test a new way of combining several program transformations into one transformation, inspired by the combinatorial concept of Covering Arrays (CA). We have equipped the ADATE automatic programming system with this new CA transformation and conducted a series of 18 experiments which show that the CA transformation is a highly useful supplement to the existing ADATE transformations.

• Olsson, Roland (2003). Machine learning of human language through automatic programming. ?. p. 18–22. Full text in Research Archive
• Olsson, Roland & Powers, David (2003). Learning human language semantics through inductive functional programming, Eighth Scandinavian Conference on Artificial Intelligence : SCAI'03. IOS Press. p. 131–140. Full text in Research Archive
• Olsson, Roland (2000). Neural Programming. HØit : tidsskrift for Avdeling for informasjonsteknologi, Høgskolen i Østfold. ISSN 0805-6692. Full text in Research Archive
• Olsson, Roland & Montelius, Gunnar (1998). Låt datorn Skriva Programmen! ?. p. 22–23. Full text in Research Archive Show summary

  R. Olsson har utviklet et system for automatiske programmering, ADATE, som betyr Automatic Design of Algorithms through Evolution. Artikkelen gir n eksempler på hvordan Novisen, Proffset og Forskaren kan andvende ADATE.

View all works in NVA

• Nguyen, Hoa Thi; Haugen, Øystein & Olsson, Roland (2023). Imitation Learning from Operator Experiences for a Real time CNC Machine Controller. Full text in Research Archive
• Olsson, Roland & Acharya, Shubodha (2022). Using Automatic Programming to Improve Gradient Boosting for Classification. Full text in Research Archive
• Geitle, Marius & Olsson, Roland (2019). A New Baseline for Automated Hyper-Parameter Optimization. Full text in Research Archive
• Geitle, Marius & Olsson, Roland (2017). Improving competitive differential evolution using automatic programming. Full text in Research Archive
• Geitle, Marius & Olsson, Roland (2017). Using automatic programming to design improved variants of differential evolution. Full text in Research Archive
• Brastein, Ole Magnus; Olsson, Roland; Skeie, Nils-Olav & Lindblad, Thomas (2017). Human Activity Recognition by machine learning methods. Full text in Research Archive Show summary

  Recognition of human activity from sensor data is a research field of great potential. Giving autonomous systems the ability to identify what a human subject is doing at a given time is highly useful in many industries, particularly in health care and security monitoring. Our results, using a public domain dataset, show that the state- of-the-art decision tree ensemble algorithm XGBoost gives an accuracy of 94.6% validated on an independent test set. Previously published results using support vector machines (SVM) gave an accuracy of 90.2%. As far as we know, our result is the new state of the art for this data set. Recognition of human activity carries potential privacy concerns, which to some degree constrain the choice of sensor technology. Therefore, systems such as ours which can identify activities from simple inertial sensors, e.g. accelerators and gyroscopes are of particular interest. Data from such inertial sensors are difficult to interpret using mechanistic models; hence the field of Machine Learning is particularly interesting for this application.

• Olsson, Roland & Vu, Huyen (2012). Automatic Improvement of Graph Based Image Segmentation. Full text in Research Archive
• Løkketangen, Arne & Olsson, Roland (2010). Generating Meta-Heuristic Optimization Code using ADATE. Full text in Research Archive
• Løkketangen, Arne & Olsson, Roland (2009). Generating meta-heuristic optimization code using ADATE. Full text in Research Archive
• Chilo, José; Horvath, György; Lindblad, Thomas & Olsson, Roland (2009). Electronic nose ovarian carcinoma diagnosis based on machine learning algorithms. Full text in Research Archive
• Berg, Henrik; Olsson, Roland; Rusås, Per-Olav & Jakobsen, Morgan (2009). Synthesis of control algorithms for autonomous vehicles through automatic programming. Full text in Research Archive
• Berg, Henrik; Olsson, Roland; Rusås, Per-Olav & Jakobsen, Morgan (2009). Automated synthesis of control algorithms from first principles. Full text in Research Archive
• Chilo, José; Horvath, György; Lindblad, Thomas; Olsson, Roland; Redeby, Johan & Roeraade, Johan (2009). A flexible electronic nose for odor discrimination using different methods of classification. Full text in Research Archive
• Løkketangen, Arne & Olsson, Roland (2009). Generating metaheuristic optimization code using ADATE. Full text in Research Archive
• Olsson, Roland; Hansen, Stig-Erland; Lindblad, Thomas & Chilo, Jose (2007). Comparison of three feature extraction techniques to distinguish between different infrasound signals. Full text in Research Archive
• Olsson, Roland & Hansen, Stig-Erland (2007). Improving decision tree pruning through automatic programming. Full text in Research Archive
• Olsson, Roland; Hansen, Stig-Erland; Chilo, Jose & Lindblad, Thomas (2007). Classification of infrasound events with various machine learning techniques. Full text in Research Archive
• Olsson, Roland (2007). Automatic design of algorithms through evolution (ADATE). Full text in Research Archive
• Olsson, Roland & Jakobsen, Morgan (2006). A look at generation of user interface behaviour using automatic programming. WSEAS Transactions on Information Science and Applications. ISSN 1790-0832. 3(10), p. 1909–1915. Full text in Research Archive Show summary

  With the rising popularity of mobile devices, applications for these devices must meet the challenges of a system with keypad input only. Navigating between randomly placed items using four directional keys can be frustrating, if the navigation does not meet the users intuitive expectations. Automatic programming with the ADATE system is used to generate a navigation algorithm based on subjective inputs from a user. This means that the user's mental model of navigation automatically is translated into a program.

• Olsson, Roland & Jakobsen, Morgan (2006). Generation of a four-way item-to-item navigation algorithm using automatic programming. ?. p. 9–13. Full text in Research Archive
• Berg, Henrik & Olsson, Roland (2005). Improving the Scalability of Inductive Functional Programming. WSEAS Transactions on Computers. ISSN 1109-2750. 4(8), p. 837–848. Full text in Research Archive Show summary

  When automatically synthesizing big programs, it is important to consider the combinatorial properties of the program transformations that are employed. In this paper, we show how to effectively generate transformation sequences representing incremental changes to a given program. In order to cover combinations of transformations, we use so-called covering arrays in which each line is a transformation sequence to be translated into a program. Covering and orthogonal arrays are standard tools from combinatorial design theory to guarantee coverage of all pair-wise parameter combinations, for example. We equipped the ADATE automatic programming system with our new covering array (CA) transformation and conducted a series of 18 experiments which show that the CA transformation is a highly useful supplement to the old ADATE transformations.

• Olsson, Roland & Powers, David (2003). Machine Learning of Human Language through Automatic Programming. Full text in Research Archive
• Olsson, Roland (2002). Self-Improving Genetic Programming. Full text in Research Archive
• Olsson, Roland (2001). Neural programming : automatic synthesis of symbolic programs using numerical optimization. Full text in Research Archive Show summary

  This paper shows how symbolic programs can be extracted from a special kind of recurrent neural networks, called neural progr ms. The technique is based on transforming between ordinary discrete data structures and ``continuous'' data structures representing probability distributions. The linear combinations of real values in ordinary neural networks are replaced with mi ing of continuous data structures. Node activation functions, such as the sigmoid and tanh functions, are replaced with continuous equivalents of normal functions in symbolic programs, for example sorting a list. The resulting nets can use very ma y hidden layers since these new activation functions turn out to be smoother than for example the sigmoid.

• Olsson, Roland & Ricardo, Poli (2000). How to Invent Functions. Full text in Research Archive
• Olsson, Roland (1999). Population management for Automatic Design of Algorithms Through Evolution. Full text in Research Archive
• Olsson, Roland (1999). The Art of Writing Specifications for the ADATE Automatic Programming System. Full text in Research Archive

View all works in NVA