Daphne Demekas

Peer-reviewed Publications

Olsen, D., et al. (2025). NEAR: Neural Embeddings for Amino Acid Relationships. Bioinformatics.

Demekas, D., et al. (2023). An Analytical Model of Active Inference in the Iterated Prisoner's Dilemma. International Workshop on Active Inference (IWAI).

Heins, C., et al. (2023). Spin Glass Systems as Collective Active Inference. International Workshop on Active Inference (IWAI).

Albarracin., et al. (2022). Epistemic Communities Under Active Inference. Entropy.

Heins, C., Millidge, B., Demekas, D., et al. (2022). pymdp: A Python Library for Active Inference in Discrete State Spaces. Journal of Open Source Software.

Demekas, D., et al. (2019). An Investigation of the Free Energy Principle in Emotion Recognition. Frontiers in Computational Neuroscience.

Manuscripts in Preparation

Demekas, D. & Deane, G. Recursive self-models and minimal phenomenal experience

Awards

The 2025 Computational Phenomenology of Pure Awareness Prize

George Deane and I won the 2025 Computational Phenomenology of Pure Awareness Prize for our paper Recursive Self-Models and Minimal Phenomenal Experience, where we framed MPE as a limit case within a computational architecture, in which a policy model generating behavior is recursively coupled to a program model that explains that behavior and learns to model its own explanatory process.

Past work

Software Scientist at the Wheeler Lab, University of Arizona, 01/2023 - 08/2024

The Wheeler Lab is a computational biology research lab. I worked on ML projects including:

• NEAR: An CNN-based model for protein homology detection

• DIPLOMAT: an ML animal tracking and behaviour analysis tool

I also served a a mentor for Masters students, and participated in research groups and literature reviews.

Research Associate (ML) at Birkbeck, University of London, 05/2022 - 09/2022

At Birkbeck, I worked primarily on a project which was a collaboration between Birkbeck and the Victoria&Albert museum, in which I fine-tuned a diffusion model on the museum’s collection and developed a platform to demo at an in-person exhibition and workshop. People could use the model to generate images that combined different themes and styles across the collection, to visualize artifacts that would be an intersection between them - for example, a ceramic vase made by Versace from the 17th century.

Developer at Northeastern University, Network Science Insitute, 01/2022-01.

At Northeastern I worked as a researcher and developer on network and system science. I developed simulations of Erdős-Rényi and Watts-Strogatz networks for modeling propagation of beliefs within networks of active inference agents. Through this work, I contributed mathematical derivations to work showing that collectives of active inference agents are mathematically equivalent to spin glass systems - disordered systems from statistical physics used to model collective behavior under competing interactions (IWAI 2022).

I was also first author on an analytical model of agents playing the Iterated Prisoner's Dilemma (Demekas, Heins & Klein, 2023), showing that bounded‐rational Bayesian agents can recover optimal game-theoretic strategies without any explicit strategic reasoning, and proving that the critical time points when agents transition from defection to cooperation are functions of learning rate and reward structure.

Software Engineer at 9fin, 01/2022- 01/2023

9fin is a platform providing information about fixed icnome assets such as debt and bonds. My areas of work was mainly backend engineering, building a nd maintaining endpoints for for applicaiton, using AWS state machines and lambdas, as well as SQL and S3 for data management and boto3 for python client communication. I also worked with computer vision and NLP, helping the legal team optimize their workflow by building a recommendation engine that parsed PDF documents and recommended text-tags.

ML Engineer at Nested Minds, 01/2021 - 01-2022

Nested Minds was an active inference start-up coming out of Karl Friston’s theoretical neurobiology group at UCL. I helped with all aspects of engineering, (algorithm design, generative models and back-end development and infrastructure) as well as team leadership, teaching and onboarding. Noteable projects included Huxley: an AI diffusion algorithm used to create Duran Duran’s music video Invisible, and Disney Automatoy, a robot for Disney’s amusement park that engages in social interactions, detecting and replicating behaviours such as waving, facial expressions and clapping.