I'm an artificial intelligence and computational neuroscience researcher interested in closing the gap between deep learning and the human mind. Current deep neural networks excel at perceptual tasks, but have yet to break through in the kinds of problems that humans solve through conscious, effortful thought. In addition, they are static models that fail to continually learn in open-ended ways, like humans do. I aim to understand these more elusive modes of information processing so that we can mechanize them and build general and human-like artificial intelligence. My primary lens into these problems is algorithmic information theory (the mathematics of compression), which I use to define formal notions of structure, compositionality, and open-ended knowledge growth.
Currently, I'm a PhD student at Mila supervised by professors Guillaume Lajoie and Yoshua Bengio, where my research focuses on the development of AI models with inductive biases from consciousness and high-level human cognition, as well as compression-based perspectives on intelligence. I'm also a part-time student researcher at Google on the Paradigms of Intelligence team, where I work on the problem of open-ended knowledge acquisition using large sequence models capable of in-context learning.
Previously, I was a research assistant in the Department of Cognitive Science at Johns Hopkins University, where I worked with my advisor Mick Bonner to better comprehend human vision. Specifically, the group seeks to reverse engineer the representations and algorithms of human visual cognition through computational modeling, neuroimaging, and behavioral experiments. This work relies heavily on deep artificial neural networks as theoretical models of information processing in the human brain, and broadly makes use of large-scale computational methods to characterize visual perception.
Outside of research, I'm also very passionate about teaching. I worked as a Data Science Instructor at Lighthouse Labs, where I designed and taught my own lectures on various topics in machine learning to students enrolled in an intensive coding bootcamp. The students came from all types of academic and professional backgrounds, which made for a unique and exciting learning dynamic!
I am currently pursuing a PhD under the supervision of professors Guillaume Lajoie and Yoshua Bengio, where my research focuses on the development of AI models with inductive biases from consciousness and high-level human cognition, as well as compression-based perspectives on intelligence.
I did research into the computational basis of human vision with my advisor, Mick Bonner. Most of my work involved using deep neural networks (DNNs) as computational models, in addition to collected and analyzing fMRI/behavioral data. Projects of mine included:
My curriculum was broadly focused in computer software. I also took a number of machine learning courses in my final year after completing a 1-year computer vision internship at ModiFace.
I did my final year capstone project with Jonathan Rose. My team and I designed an application for human memory augmentation, where your phone could passively listen to all of the speech in your surroundings, transcribe it, and make it permanently available for future reference using Google Search-style queries. The idea was that the application would make it easy to retrieve the contents of any past conversation you've ever had. We were partially inspired by an episode of Black Mirror, but our hope was that the outcome would be a bit less dystopian.
I went to CEGEP in Quebec, which is a 2-year college program between high-school and university.
During my final project, I constructed a cloud chamber to observe cosmic solar radiation, partially replicating the experiments that first discovered antimatter and various subatomic particles.
Open-ended exploration using predictive sequence models and in-context learning, supervised by João Sacramento.
Research on information representation and algorithms in the visual system of the human brain with Professor Michael Bonner.
Lead groups of students through tutorial exercises and explain concepts in computational neuroscience.
Full-stack web development using AngularJS, Angular Material, ASP.NET MVC, Web API, and SQL Server in order to improve internal workflow efficiency for financial reporting.
For the most up-to-date list of publications, see my Google Scholar profile.
| 2026 | Can In-Context Learning Support Intrinsic Curiosity? Eric Elmoznino, Sangnie Bhardwaj, Johannes von Oswald, Rajai Nasser, João Sacramento, Rif A Saurous, Guillaume Lajoie Preprint |
| 2026 | Simplifying the Modeling of Arbitrary Conditionals in Natural Language. Yinhan Lu, Eric Elmoznino, Léo Gagnon, Sarthak Mittal, Tejas Kasetty, Guillaume Lajoie Preprint |
| 2026 | A Compression Perspective on Simplicity Bias. Tom Marty, Eric Elmoznino, Dhanya Sridhar Preprint |
| 2025 | A Complexity-Based Theory of Compositionality. Eric Elmoznino, Thomas Jiralerspong, Yoshua Bengio, Guillaume Lajoie ICML poster |
| 2025 | In-context learning and Occam's razor. Eric Elmoznino, Tom Marty, Tejas Kasetty, Léo Gagnon, Sarthak Mittal, Mahan Fathi, Dhanya Sridhar, Guillaume Lajoie ICML poster |
| 2025 | Does learning the right latent variables necessarily improve in-context learning? Eric Elmoznino, Sarthak Mittal, Léo Gagnon, Sangnie Bhardwaj, Tom Marty, Dhanya Sridhar, Guillaume Lajoie ICML poster |
| 2025 | Illusions of AI consciousness Yoshua Bengio & Eric Elmoznino Science |
| 2025 | Multi-agent cooperation through learning-aware policy gradients. Alexander Meulemans, Seijin Kobayashi, Johannes von Oswald, Nino Scherrer, Eric Elmoznino, Blake Richards, Guillaume Lajoie, Blaise Aguera y Arcas, João Sacramento ICLR poster |
| 2025 | Convolutional architectures are cortex-aligned de novo Atlas Kazemian, Eric Elmoznino, Michael F. Bonner Nature Machine Intelligence |
| 2024 | Amortizing intractable inference in large language models. Edward J. Hu, Moksh Jain, Eric Elmoznino, Younesse Kaddar, Guillaume Lajoie, Yoshua Bengio, Nikolay Malkin ICLR talk — best paper honorable mention |
| 2024 | Sources of Richness and Ineffability for Phenomenally Conscious States. Eric Elmoznino, Xu Ji, George Deane, Axel Constant, Guillaume Dumas, Guillaume Lajoie, Jonathan Simon, Yoshua Bengio Neuroscience of Consciousness |
| 2024 | High-performing neural network models of visual cortex benefit from high latent dimensionality. Eric Elmoznino, & Michael F. Bonner PLOS Computational Biology |
| 2023 | Discrete, compositional, and symbolic representations through attractor dynamics. Andrew Nam, Eric Elmoznino, Nikolay Malkin, Chen Sun, Yoshua Bengio, Guillaume Lajoie NeurIPS Workshop talk |
| 2023 | Consciousness in Artificial Intelligence: Insights from the Science of Consciousness. Patrick Butlin, Robert Long, Eric Elmoznino, Yoshua Bengio, Jonathan Birch, Axel Constant, George Deane, Stephen M. Fleming, Chris Frith, Xu Ji, Ryota Kanai, Colin Klein, Grace Lindsay, Matthias Michel, Liad Mudrik, Megan A. K. Peters, Eric Schwitzgebel, Jonathan Simon, Rufin VanRullen Preprint, condensed 2025 version published in Trends in Cognitive Sciences |
| 2023 | Scene context is predictive of unconstrained object similarity judgments. Caterina Magri, Eric Elmoznino, Michael F. Bonner Cognition |
| 2023 | Learning Macro Variables with Auto-encoders. Maitreyi Swaroop, Eric Elmoznino, Dhanya Sridhar NeurIPS Workshop poster |
| 2020 | Visual representations derived from multiplicative interactions. Eric Elmoznino, & Michael F. Bonner NeurIPS Workshop poster |
| 2019 | A new procedure, free from human assessment that automatically grades some facial skin structural signs. Comparison with assessments by experts, using referential atlases of skin ageing. Jiang R., Kezele I., Levinshtein A., Flament F., Zhang J., Elmoznino E., Ma J., Ma J., Coquide J., Arcin V., Omoyuri E., Aarabi P. International Journal of Cosmetic Science |
| 2026 | Compression & intelligence — Center for Neuroscience Imaging Research, South Korea |
| 2026 | Compositional languages — Cognition theory journal club, NYU |
| 2025 | Compositional languages — Analogy group |
| 2025 | Compositional languages — Jenniffer Hu Lab, Johns Hopkins University |
| 2025 | Good visual representations are high-dimensional — Martin Hebart Lab, Max Planck Institute |
| 2024 | Why can't we describe our conscious experiences? An information theoretic attractor dynamics perspective of ineffability — Models of Consciousness conference |
| 2024 | Consciousness, ineffability, and AI safety — AI Safety Reading Group, Mila |
| 2023 | Sampling discrete objects through continuous attractor dynamics — GFlowNet Reading Group, Mila |
| 2023 | Why can't we describe our conscious experiences? An information theoretic attractor dynamics perspective of ineffability — Computational Phenomenology Group |
| 2023 | Why can't we describe our conscious experiences? An attractor dynamics perspective of the ineffability of qualia — University of Toronto guest lecture |
| 2020 | How does the brain work? Cognitive science research — SABES |
| 2020 | Introduction to Programming with Python — UofTHacks |
| 2025 | Could AIs be Conscious? — New Horizons podcast |
| 2023 | Why can't we describe our conscious experiences? An information theoretic attractor dynamics perspective of ineffability — Active Inference Institute podcast |
| 2022 | System and method for image processing using deep neural networks. Levinshtein A., Chang C., Phung E., Kezele I., Guo W., Elmoznino E., Jiang R., Aarabi P. U.S. Patent No. 11216988 |
| 2021 | Image-to-image translation using unpaired data for supervised learning. Elmoznino E., Kezele I., Aarabi P. U.S. Patent Application No. 17096774 |
| 2020 | System and method for augmented reality using conditional cycle-consistent generative image-to-image translation models. Elmoznino E., Ma H., Kezele I., Phung E., Levinshtein A., Aarabi P. U.S. Patent Application No. 16683398 |
| 2020 | Machine image colour extraction and machine image construction using an extracted colour. Elmoznino E., Aarabi P., Zhang Y. U.S. Patent Application No. 16854975 |
| 2020 | Automatic image-based skin diagnostics using deep learning. Jiang R., Ma J., Ma H., Elmoznino E., Kezele I., Levinshtein A., Charbit J., Despois J., Perrot M., Antoinin F., Flament R.S., Parham A. U.S. Patent Application No. 16702895 |
Mila - Quebec AI Institute
6666 St-Urbain, #200
Montreal, QC, H2S 3H1