The three master students Jørgen Jensen Farner, Ruben Jahren, and Håkon Weydahl, Kristine Heiney in the front, then postdoc/Assoc. Prof. Ola Huse Ramstad, and Prof. Stefano Nichele.

NordSTAR paper in ICES

A student project from one of our master courses, Evolutionary AI and Robotics, in the Applied Computer and Information Technology programme is set to be presented and published as a conference paper [1] this December and has received two nominations for Best Paper Award.

The project is part of the area on biologically inspired artificial intelligence in NordSTAR and was conducted as part of the NFR-funded SOCRATES (ntnu.edu) and DeepCA (nichele.eu) projects.

– This is a fantastic achievement for our students, says Kristine Heiney (krisheiney.eu), a PhD Fellow in the Department of Computer Science and one of the senior authors on the paper. 

– They’ve done really spectacular work elevating their original project, which was already quite impressive! It’s been a real treat seeing them work as a team and working with them to turn the great work they’ve done into a complete paper.

Bringing the team together

This paper grew out of one of the course’s final projects, the concept of which was developed by Kristine in cooperation with the course coordinator and her supervisor, Professor Stefano Nichele (nichele.eu).

In this project, the students had to develop models that produced behavior similar to activity recorded from living neurons in a dish. 

– It is of great importance for us to teach the students cutting edge AI methods, but at the same time allow them to apply the acquired knowledge in ongoing research projects and make them part of our research environment. In this way, they learn even more than what is planned in the course learning outcomes, says Stefano. 

Kristine and Stefano were impressed by the work done by a group of three students — Håkon Weydahl, Ruben Jahren, and Jørgen Jensen Farner — and approached them after the course had concluded to see if they would be interested in developing it into a paper, together with Håkon’s supervisor Ola Huse Ramstad, a postdoctoral fellow and associate professor also involved in the DeepCA project.

Emulating the activity of brain cells

In this project, the students started with a selection from a large dataset published by the Potter Lab in 2006 [2]. The data were collected by recording electrical signals from biological neurons plated in a glass dish embedded with 60 electrodes.

Our brain uses these electrical signals to communicate, and observing these small populations of neurons in a dish as they mature can give us insight into how our brains encode and process information.

The goal of the project was to take these recorded signals and use them as a target for an artificial intelligence algorithm called an evolutionary algorithm. This algorithm is inspired by the natural process of evolution, and it works by passing on beneficial traits from generation to generation of individuals, representative of solutions to the problem at hand.

In this particular case, each individual was a neural network model, and those that produced activity closest to the target data were used to pass on traits to the next generation of models. This was continued generation after generation, until individuals with activity patterns close to the data were obtained.

The figure illustrates how bio-inspired models are created from data collected from living neurons using an evolutionary approach.

The figure illustrates how bio-inspired models are created from data collected from living neurons using an evolutionary approach.

Bio-inspiration for next-generation computing

– The human brain is extremely efficient and powerful when it comes to information processing, says Kristine. 

– If we can produce models that mimic the behavior of living neural networks, but with simple components that could be replicated in hardware, this can open the door to a new paradigm of computation—greater computational power, lower energy consumption, and more flexibility.

– I believe we’re at the edge of a big shift in how computers work, and what’s going to push us over that edge is learning from nature.

The students were able to successfully capture some of the interesting patterns of activity recorded in the experimental data: network-wide bursts of activity happening at varying intervals of time, representing information propagating from neuron to neuron throughout the network. 

With these models on hand, one possible next step would be to plug them into a framework where they can be used to perform concrete computational tasks, and their performance and efficiency could be compared against conventional methods. Ultimately, they could be used as a template for computational hardware.

“The best aspect of being an AI student at OsloMet”

The students really enjoyed doing this project as well.

– The project was fun from the get-go, challenging us to conceptualise biological neurons as a complex system to be modelled, says Ruben.

– While our process was thorough, and results certainly interesting, Kristine's understanding of the subject is what really took our project to the next level. For me personally, being exposed to, and even a part of, biologically inspired artificial intelligence projects like this is the best aspect of being an AI student at OsloMet.

The students all echoed the same sentiment, saying they learned a lot about how different their results seemed when grounded in the theory their supervisors brought to the table, and about how important it is to present data and results in a way that clearly communicates a study’s findings.

– It was one of those projects where you get so into it that you lose track of time and end up spending your evenings working on it, says Jørgen of the process.

– It is a great feeling that our hard work paid off and that it got accepted to the conference, but it would not have been possible without the help of our supervisor Kristine, who turned our heap of code into sensible theory.

The venue: IEEE SSCI 2021

The paper, entitled “Evolving spiking neuron cellular automata and networks to emulate in vitro neuronal activity” [2], has been accepted at the International Conference on Evolvable Systems (ICES), which is one of the symposia of IEEE SSCI 2021 (Symposium Series on Computational Intelligence).

ICES has been held since 1995 and has become the leading conference for showcasing techniques and applications of evolvable systems. SSCI is the flagship annual international series of symposia on computational intelligence sponsored by the IEEE Computational Intelligence Society (CIS).

With resoundingly positive reviews from two of the three editors, the paper also received two nominations for the Best Paper Award at the symposium series.

References: 

[1] J Jensen Farner, H Weydahl, CR Jahren, O Huse Ramstad, S Nichele, and K Heiney. "Evolving spiking neuron cellular automata and networks to emulate in vitro neuronal activity," International Conference on Evolvable Systems (IEEE Symposium Series on Computational Intelligence 2021), 2021.

[2] DA Wagenaar, J Pine, and SM Potter, "An extremely rich repertoire of bursting patterns during the development of cortical cultures," BMC Neuroscience, 7(1):11, 2006.

Loading ...

The top photo shows the three master students Jørgen Jensen Farner, Ruben Jahren, and Håkon Weydahl, Kristine Heiney in the front, then postdoc/Assoc. Prof. Ola Huse Ramstad, and Prof. Stefano Nichele.