S. Zayd Enam Sciences
Neurally Inspired Self-Organizing Maps for Image Coding
In this project we plan on using parallelized computation to build realistic sparse coding models for neurons in the primary visual cortex (V1). Sparse coding is a stimulus encoding technique used by V1 neurons that aims to minimize the number active neurons required in encoding any input image. Due to computational constraints, previous sparse coding models have been limited in their ability to match the biology of lateral geniculate nucleus projections to V1. Our models will allow us to better describe recorded biological data and provide further evidence that V1 relies upon sparse coding of input images.
This has been a fantastic summer. There is no other way I could describe a summer spent working on projects that excite me. As an added bonus I got to work with lab mates and researchers who are just as excited about the same stuff.
At the beginning of the summer I began with some project ideas I wanted to explore. Over the course of the summer I got the opportunity to take a stab at these projects and got feedback from some smart people. Alongside this, I went back and read some of the early literature in my field. I would not have read as much of the early papers if I was not setting my own research agenda. Feedback from colleagues and reading as much as I could opened up a lot of interesting avenues to explore. It has become clear to me that there still are a lot of questions in my field with only hand-wavy answers.
Regardless, my summer and my projects continued along. I faced some interesting hurdles and learned some valuable lessons. One of the most valuable lessons was learning to manage time between discussing ideas and implementing ideas. There's a lot of value in getting feedback and collaborating on ideas but at some point you need to just build your model.
This program has helped me focus on the sort of problems I want to tackle in my career. Over the summer I have become increasingly excited about the challenge of attempting to reduce complex systems - like vision - to a set of fundamental principles.