MIT EECS | Advanced Micro Devices Undergraduate Research and Innovation Scholar
Manipulating Symbols with Recurrent Neural Networks
- Brain and Cognitive Science
The assembly hypothesis states that there is an intermediate level of brain computation, called assembly operations, that can be understood in terms of the synchronized firing of large, interconnected populations of neurons and may be implicated in memory and higher cognitive functions. The full set of assembly operations provides a natural system for encoding symbols in the brain and performing symbolic manipulations. The goal of this project is to train RNN models of these operations and use them as motifs to be repeated in a deep neural network, similar to convolution layers. We will then construct novel network architectures that can solve problems requiring symbolic reasoning in the domains of Visual Intelligence and Natural Language Processing, as well as more abstract tasks.
“I am participating in SuperUROP because I am excited to develop the relationship between the science of how we think and the computational tools we use to solve problems. Through this process, I look forward to learning the necessary skills to work on a research project and applying knowledge from past classes in new and creative ways.”