This session aims to explore cutting-edge advancements in the integration of machine learning and edge intelligence with the theory and practice of neuroscience of movement, under the principles of active inference. Active inference is a theoretical framework in computational neuroscience that proposes that organisms actively minimize their free energy, or surprise, by continuously updating their internal models of the environment based on sensory input. This framework is often used in the study of perception, decision-making, and learning in biological systems (Parr, Pezzulo, and Friston 2022) More recently, active inference is also used to develop agent-based algorithms and models that simulate how organisms interact with their environment to make predictions and decisions (Vries 2023), and to optimize their performance on an Edge infrastructure (Taheri et al. 2023).

Motor impairments due to neurological conditions such as stroke, spinal cord injuries, and Parkinson’s disease present significant challenges for rehabilitation. Traditional therapeutic methods often lack the precision and personalization needed to optimize patient outcomes. By incorporating AI-driven predictive models and real-time data analysis, offering highly tailored therapeutic interventions and improving the efficacy of treatment protocols may be possible, even on consumer devices using the principles of Active inference.

In this session, following the work of de Vries (2023), we will first identify the necessary features for a software toolbox that supports a competent non-expert engineer to develop working active inference agents. We will then conduct a hands-on session to introduce software examples to design active inference agents in a similar way as TensorFlow propelled applications of deep learning technology. Attendees will gain insights into how Active Inference can be harnessed to solve complex dynamic challenges, the interdisciplinary collaboration necessary for such innovations, and the potential impacts of these advancements.

Program

Introduction and overview (5 minutes)

• Introduction to the workshop and its objectives.
• Overview of the session agenda and activities
  Presented by the organizers.

Keynote presentation: Learning under uncertainty (15 minutes)
Detailed presentation by the invited guest expert: Chris Mathys, Aarhus University, Interacting Minds Centre

Interactive workshop: Tools for Developing AI-Driven Therapeutic Models (45 minutes)

Hands-on session where participants will engage in developing
simple AI models for therapeutic interventions using exemplar toolboxes.
Facilitated by session organizers.

Wrap-up and panel discussion: Future Directions and Interdisciplinary Collaboration (25 minutes)

• Panelists: Dr. Carmelo Bellardita, Assoc. Prof. Melih Kandemir, Assoc. Prof. Cumhur Erkut, and the invited guest expert: Chris Mathys
• Discussion on the future of AI in neurorehabilitation, interdisciplinary collaboration, and potential for scaling and commercialization
• Audience Q&A.

Target audience

This session proposal provides a practical and engaging exploration of active inference principles, ensuring relevance and value to the D3A 2.0 community. The session is designed for researchers, clinicians, AI and machine learning experts, and healthcare technology developers who are familiar with Python and Matlab. We aim to attract participants interested in the practical applications of Active Inference.

Organizers

• Assoc. Prof. Cumhur Erkut, Aalborg University, cer@create.aau.dk (Main organizer)
• Dr. Carmelo Bellardita, University of Copenhagen, cbellardita@sund.ku.dk
• Assoc. Prof. Melih Kandemir, U. Southern Denmark, kandemir@imada.sdu.dk

Diversity and Inclusion, and balance between Junior and Senior Organizers

Gender diversity and inclusion, as well as age balance will be maintained by actively involving our post-docs (female) and researchers of Copenhagen Hearing and Balance Center (mostly female).

References

Parr, Thomas, Giovanni Pezzulo, and Karl J Friston. 2022. Active Inference. The MIT Press.

Taheri, Javid, Schahram Dustdar, Albert Zomaya, and Shuiguang Deng. 2023. Edge Intelligence, From Theory to Practice. Springer

Vries, Bert de. 2023. Toward Design of Synthetic Active Inference Agents by Mere Mortals. Communications in Computer and Information Science, 173–85