Inferring the Architecture of the Visual Pathway Using Deep Neural Networks


Engineering and Applied Sciences


Project Summary

This summer, I worked at the Computational Neuroscience Initiative Lab under the supervision of Professor Vijay Balasubramanian. While there, my lab partner, Chetan Parthiban, and I worked on using Deep Neural Networks to learn insights into the mechanisms by which the brain learns to process visual information.

 Deep Neural Networks (DNNs) are a recent development in machine learning and artificial intelligence. These networks hold great promise in allowing computers to learn patterns and trends in a variety of inputs, from images in a movie to stanzas in a poetry. Although DNNs have achieved incredible success in image processing, their applications on visual processing, starting with the output of the retina, have yet to be explored. Over the course of the summer, I learned how to code, apply, and interpret Deep Neural Networks on neural spikes from the retina. The goal of this project was to analyze the structure and nodes in our DNN to see if we could extrapolate how the visual cortex processes information.

This experience showed me the practical implications of DNNs within the context of computational neuroscience and beyond. I was able to explore an application of DNNs that had not previously been investigated- namely, the application of Deep Neural Networks on retinal inputs. Beyond this, I learned how to code in modern neural networking languages like TensorFlow and Keras. From a neuroscience perspective, I also gained insight into how the brain process visual information by interacting with real data gathered from Salamander retinas. I was also able to learn about the stochastic processes which lead to visual interpretation.

 From a skills perspective, the research I conducted over the summer contributed to my ability to conduct experiments, construct models, and analyze data. I was able to acquire skills through my work in coding and data analysis, which will be invaluable skills in my future as a CS major. More importantly, the research I conducted this summer was hugely influential in opening my eyes to the potential of neural networks and artificial intelligence in the modern age, even inspiring me to code my own neural network projects in my spare time. Ultimately, this experience was so much more than just the research I conducted; it was an opportunity for me to explore a field of computer science I had never been exposed to previously. Out of all the neural networks I trained to learn over the course of the summer, it was I who learned the most.

Inferring the Architecture of the Visual Pathway Using Deep Neural Networks