Decoding the Brain
I'm Molin, a second-year biomedical engineering student at UC Davis. I'm just getting started in research, with independent projects on language brain-computer interfaces and on interpretability.
I also build small tools, like a Mac app that lets you scroll using the Doppler effect.
Selected work
All work →- A causal location code in a vision-language model After watching a walk through a room, can a model tell where a new view was taken, and is there anything inside it that carries the location?
- Can a brain-to-text decoder skip the phoneme stage? Decoding attempted speech with a diffusion language model, connected to the neural encoder without a phoneme stage.
- Does a decoder trained on listening decode inner speech? Reproducing the isolated-word analysis of Kunz et al. 2025 on the public data, then training on one behavior and testing on another.
- Self-supervised pretraining on fMRI, tested Does pretraining on naturalistic fMRI help decode what a listener hears? Short answer: no.
Writing
All writing →- A diffusion language model as a brain-to-text decoder Training the language model in a speech decoder by block diffusion, not left to right, cut word error rate by about ten points. The gain disappeared once the encoder had been trained on phonemes.
- Aligning a neural encoder to a language model, without phonemes Contrastive alignment taught a speech encoder to pick the right sentence out of 704 a third of the time, with no phoneme labels. It did not make the encoder a better starting point for decoding.
- Can a projector turn inner speech into attempted speech? Inner speech looks like attempted speech, shifted and quieter. Reversing that with a linear map made the sentence easier to identify and did nothing for the synthesized voice.
- Place cells are easy to find. That's the problem. Units that pass the rodent place-cell test turn up in raw pixels and in random networks. To find out whether a vision-language model knows where it is, I had to change its activations and watch the answer move.
Toys
All toys →- Place fields Be the rat. Walk a pointer around a room and watch simulated place units fire, with a live spatial-information score.
- Caffeine Curve Log your coffees and see how much caffeine is still in you at bedtime. Works for anything with a half-life.
- A five-digit AmazingHand A fork of Pollen Robotics' open-source robot hand, with a pinky, three-segment fingers and a powered wrist.