Nilaksh

The development of autonomous agents that can learn and adapt continually throughout their lifetime is a central goal of artificial intelligence. Current machine learning models often struggle with catastrophic forgetting when faced with new tasks and fail to adapt efficiently to dynamic, real-world environments. During my PhD I aim to address these limitations by developing novel interactive and adaptive reinforcement learning (RL) algorithms. A key objective is to create embodiment-agnostic agents that not only generalize to unseen tasks but can also adapt quickly to new domains and data modalities, a critical challenge for developing truly general intelligent systems. I also aim to explore how transformer-based architectures can serve as the foundation for such learning systems, aiming for sample-efficiency and robustness in complex applications. Ultimately, during my PhD I seek to contribute to the foundations of adaptable and more general intelligent systems.