Research Projects Predictive RL

Optimizing Task Performance with Predictive RL

This project examines how reinforcement learning (RL) utilizes predictive maps to enhance task performance. It explores how RL leverages semi-supervised predictive models to create robust representations (model based RL). These algorithms are tested on neuroscientific tasks and compared with neuroscience data to deepen our understanding of how the brain integrates predictive and RL-based models to solve tasks.

Predicting Neural Dynamics with AI Models

The project aims to apply modern time series forecasting techniques, specifically transformers and mixture of linear systems, to predict neural activity. The goal is to create predictive embeddings of neural activity, providing a compact, yet rich, representation of the brain’s dynamics. By integrating both transformer-based and linear system models, we aim to capture the complex temporal dependencies in neural data and generate accurate forecasts. This project offers a focused exploration of the intersection between machine learning and neuroscience, with an emphasis on time series forecasting. 

Other related projects are ongoing. For example simulating RL agents for multiple neuroscientific tasks. Please reach out to know more.