Effective Abstraction in Relational RL

Providing state abstraction for better task transfer in taskable reinforcement learning environments using D-FOCI statements. Kokel et al. ICAPS 2021.

Relational Q Learning

Implemented Q-learning for relational domains like blocksworld, where the states are represented using first-order logic predicates. [ code ]

Communicating with Computers

This is a DARPA funded project to build intelligent minecraft agent that can communicate and collaborate with humans through chat to build structures. [ video ]

Knowledge Intensive Gradient Boosting

Leveraging qualitative domain knowledge while learning tree based gradient boosting models to improve predictions in regions where data is noisy or absent. Kokel et al. AAAI 2020.

Learning Sparse Graph for GNN

Used meta-learning techniques to optimize the graph structure for obtaining sparse graph. Wan and Kokel, DLG 2021.

Causal inference from Protein Expression Data

Discovering causal molecular relationships from the evaluation of observational data using do-calculus. More details.

Ja-Walk-ER

Developed an interface that allows users with basic understanding of ER Diagrams to provide search bias for Inductive Logic Programming based. As described in Hayes et al. 2017. [ code ]

Expression Detection

A small project to detect wink and shush expression using OpenCV. [ code ]

RL for Healthcare

Learning polices for management of children on ECMO using batch reinforcement learning techniques. More details.

SRL model for credit default

Learnt and evaluated a statistical relational model for Kaggle Home credit default risk dataset and compared it with propositional models.

Optimization

Implementation of various first-order and second-order gradient methods for optimization. Including Barzilai-Borwein Gradient Descent, Conjugate Gradient Descent, Limited Memory BFGS and Armijo Line-search. [ code ]

ML/AI basket

My basket of Machine Learning and AI algorithms implemented over time. [ code ]