Learning Neuroscience and Biology by re-implementing AI-related papers

I’m currently also working towards implementing the two papers below :partying_face: :laughing:. I will document everything I learn along the way in this post, including all the implementation details :blush: :blush:

:brain: :computer: Paper 1: Decoding neural activity into text [link paper] [my implementation]

:dna: :microbe: Paper 2: Generating new proteins using language model [link paper] [my implementation]

EDIT 2: typo


The last few days i learned: how neurons communicate, how to identify toxic molecules in the human body (deep dive soon)

1 Like

TIL: an overview of how to implant a chip into the brain, represent molecules that can be understood by a computer

the last two days i learned: the common types of brain responses in brain-computer interfaces, an overview of generating new proteins using language models (ProGen paper), how to go from DNA to proteins

[ROUND 1] the last two days i learned: how to integrate external sensory input into the brain, how the brain can control prosthetic arms, how to extract neural activity, completed a language model base for generating new proteins (will add more) + some biology


the last four days i learned: 1/6 the decoder model for neural activity, how to handle neural representation change over time (will share notes after finishing it), more on brain responses uses to control a BCI, and some biology


the last three days i learned: wrote dataset classes for character and sentence neural activity, techniques for stimulating neurons , what can be achieved with bidirectional BCI (I’m looking forward to building one :wink: ), how to create synthetic neural data (next, I will implement it), some biology


the last three days: 2/5 how to decode neural activity into text


the last three days i learned: wrote a vanilla generator for synthetic neural data (will improve), wrote a protein dataset class


the last two days i learned: fixed some bugs in the decoder model for neural activity and added some improvements to the generate protein model, some biology

Update: I’m considering allocating 75% of my time to AI and 25% to non-AI (science, including this thread). So you will see my progress on this thread slowing down. This isn’t fixed, I am still figuring out how I should allocate my time…

Hey @xariusdrake!! Interesting work here. My research in grad school will likely be focused on these topics so I’m curious to see what I can learn on here. Would you be open to have short calls on Discord (like an informal study group), so that we can discuss these papers together? (I’m in the GMT+5:30 timezone btw).

Hey Jame. Thank you for reaching out! Please drop me a message neuralink#7014 on Discord

1 Like

HI @xariusdrake,
you might find this work interesting, precisely at the intersection between Biological and Artificial Neural Networks: Brain connectivity meets reservoir computing
Here’s the scikit-learn compatible Python package used for the experiments as well: GitHub - fabridamicelli/echoes: Machine Learning with Echo State Networks, a scikit-learn compatible package.

I’d be happy to hear any feedback :slight_smile:

1 Like

Hey. Thanks for the recommendation. I’d check it out

the last two days i learned: how biological neurons recognize faces (not like CLIP’s neurons, biological neurons can react to both visual and conceptual stimuli), experimental design

the last six days i learned: navigation capabilities in animals, brain development in infants

1 Like

Update: I’m currently spending 75% of my time on AI and 25% on science, so you’ll see my progress here slow down

TIL: how biological neural network solve catastrophic forgeting

1 Like