Hi, my name is Ewan. I’m a senior Computer Science undergrad at the University Of Vermont, working toward a PhD in computational neuroscience. I’m interested in how in silico models can make the brain interpretable, and how much we can learn from sparse, everyday signals of cognition.
Modern deep networks are exceptionally good at predicting brain responses to stimuli, partly thanks to advances in Representations of text, images, and video. I’m curious about what these models have learned, and what probing them might teach us about the brain. Even models trained to predict classical single-neuron recordings (Lad et al., 2026) offer a new medium for interpreting circuits en masse. Highly recommend that paper.
These models can also be used to design exploratory, almost adversarial stimuli, using neuroimaging feedback to guide new experiments. Well-funded academic research (Henderson et al., 2026) has piqued my interest in pushing this further, guiding deep learning research toward real neural representations. The same approach can tell us about the content Representation models themselves (Prince et al., 2026): if two models predict responses to natural stimuli equally well, what new stimuli would distinguish them? On a much smaller scale, I’ve been doing experiments with more open tools like TRIBE v2, and impractical ones like Maximizing Activation through Differentiable Content Consumption.
As an undergrad with limited funding (my entire disposable income), I’ve taken these questions to sparse, everyday signals. It’s fun to think about how much of our internal brain state is already partially observable: what you say out loud or what video you’re watching might offer clues about what you’re thinking (Wu et al., n.d.). As fun as it is to apply evolutionary algorithms to TRIBE v2 inputs, it is a population-level model, and an fMRI setup is one of the few things my lab of a bedroom doesn’t have. Instead I start with two devices: a mediocre 4-channel BrainAccess HALO (after trying to 3d print my own) headset, and a Tobii ET 5, a gaming peripheral which will fight you on using it as a research eye tracker.
There’s also what we already have. With the amount of content we consume, I find interest in obsessively recording personal data as behavioral sources: computer activity (ActivityWatch), transcription/diarization of speech, and the writing timesink that is this Digital Garden vault (+git timeseries data). The same goes for Health and Longevity data; a large part of my work relies on wearable data for tracking physiology and the effects of interventions. Each stream is an incomplete view, but they are views of the same underlying state, aligned by timestamp. I want to know how much of a shared representation can be learned from several poor sensors together, and what each one adds.
Further out, I’m interested in using this model to test causal interventions against benchmarks (Anki, Chess): shaping content diet, and even extensions into other methods of Self Configuration (for me that is neuropharma, for most just stick to what you already consume).
At UVM, I work with speech data and transcription/diarization, post-training speaker representation models to better capture acoustic features that predict cognitive impairment. My room is always a lab and every kind of lab: homelab as the compute for Quantified Self data work and collection en masse, 3d Printing and ESP32s for everything, and a mini pharma shelf for pursuing the ever diminishing returns of health optimization.
Banner inspired by Simone Conradi’s Eigenfish.
Unless you’re looking for my class notes, navigate this garden through the tags rather than the folders. They’re described in How to use.
Thoughts Notes How to use Inspirations Some Fun Facts About Me
Bibliography
- Henderson, M. M., Luo, A. F., Park, S., Tarr, M. J., & Wehbe, L. (2026). Diffusion-Based Stimulus Optimization Reveals Functional Organization across Higher Visual Cortex (p. 2026.05.12.724119). bioRxiv. https://doi.org/10.64898/2026.05.12.724119
- Lad, V., Franke, K., Shaham, T. R., Ganguli, S., Tolias, A. S., Sanborn, S., & Karantzas, N. (2026). Letting the Neural Code Speak: Automated Characterization of Monkey Visual Neurons through Human Language. arXiv. https://doi.org/10.48550/arXiv.2605.12485
- Prince, J. S., Wang, B., Fel, T., Jagadeesh, A. V., Vaziri, P. A., Alvarez, G. A., Livingstone, M. S., & Konkle, T. (2026). Parametric Neural Control Differentiates Top Neural Network Models of Primate Visual Cortex (p. 2026.08.16.745063). bioRxiv. https://doi.org/10.64898/2026.08.16.745063
- Wu, Z., Schrater, P., & Pitkow, X. (n.d.). Inverse POMDP: Inferring What You Think from What You Do.