Arna Ghosh is a third-year PhD student working with Blake Richards.
What inspired you to pursue your current degree?
I wanted to work on the intersection of artificial intelligence and neuroscience in order to build brain-like intelligent systems. Following my MSc degree in Neuroscience, where I worked on deep learning pipelines for analyzing neuroimaging data, I wanted to focus on improving deep learning models by incorporating learning principles from the brain. Furthermore, I believe mimicking the functionality of the brain in artificial systems allows us to actively investigate the typical functioning of the brain.
My research entails building brain-inspired AI systems, specifically self-supervised learning systems. I am currently working on biologically plausible self-supervised learning for training deep networks that can better mimic the brain鈥檚 visual processing pathway. Additionally, my work aims to explore biologically-plausible algorithms in deep neural networks with the hope of demystifying the computational framework of learning in the brain.
What about neuroscience and your research area excites you?
As an engineer by training, understanding (and trying to build parts of) the brain as an active dynamical system excites me the most. My research area deals with decoding the learning principles underlying the functioning of the brain. These include the cost functions that different parts of the brain might be trying to optimize or plasticity rules that different constituent units might employ in order to achieve incremental performance on day-to-day tasks. Understanding these components requires approaching the problem from an interdisciplinary viewpoint, coming up with possible hypotheses and systematically testing them in a reduced artificial (digital) system.
In what way does working with your particular model benefit your research?
Having an artificial abstraction of the brain, namely the artificial neural network model, allows us to manipulate and observe several aspects of the system that would otherwise be incredibly difficult to perform experimentally. Furthermore, it is possible for us to investigate and characterize the performance using different metrics on large-scale naturalistic datasets as well as derive mathematical bounds for the model on artificial datasets. In doing so, we can understand the consequences of the learning principles being studied, compare and contrast them to behavioral studies in biological systems and accordingly design better abstractions of the brain.
What are some of your favourite activities outside the lab?
Prior to the pandemic, I would play in the McGill intramural cricket league. These days, I like going on a run outdoor during the summer. Other than that, I love going up Mont Royal to watch the sunrise. However, being a late riser, this often involves staying up all night.
What is one important thing you have learned during the pandemic?
The pandemic has taught me to be more emotionally perceptive to people around me. With people's lifestyles changing significantly since the outbreak of the pandemic, everybody has had to deal with their own mental health challenges. Talking to friends and relatives allowed me to understand what they were going through and how flawed the standards we set for others could very often be. I hope to maintain this level of understanding once we start going back to our pre-pandemic lives. I believe that doing so could help us all contribute to a healthier and happier workplace and build stronger relationships in our personal and social lives.
听
