|
My name is Vishal Sharma. In my work, I build AI technologies like LLMs, rankers, retrievers, data-pipelines for (extremely) large scale data. During my PhD at Indian Institute of Technology Delhi, I did research in "Generalized AI in planning, RL and Probabilistic Graphical Models" under the guidance of Prof. Parag Singla and Prof. Mausam. Prior to joining PhD, I worked at Oracle India for two years as a software engineer.
Email | Google Scholar | LinkedIn | Github I am in the job market this year PhD WorkIn my PhD, I tried to answer a simple question -- Can we teach machines to transfer solutions from a small-size problem(s) to larger unseen (but related) problems in structured worlds? (Here, size implies the number of objects in a problem). The work primarily focuses on learning generalized object-centric neural policies for Reinforcement Learning (RL) and Relational Planning -- policies that can be applied to an unseen variation of a given environment. This is unlike the typical RL (and planning) setting, where the policy is learned for a single environment. This work is done under the joint supervision of Prof. Parag Singla and Prof. Mausam, IIT Delhi. During initial years of my PhD, I have also worked on developing transferable lifted inference approaches for Probabilistic Graphical Models in collaboration with Vibhav Gogate, UT Dallas. I have also worked on learning object-centric video prediction models while collaborating with Guy Van den Broeck, UCLA. |
|
Conference Publications
|
Workshop Publications
|