Research Fellow | Microsoft Research
Computer Science
IIT Bombay, India
Incorporated Counting-SAT to compute Bounded Reachability Probabilities in MDPs in one shot
Designed a succinct CNF representation for Markov Chains using the transition probabilities BDD to get a 10x faster algorithm
Improved the POMCP algorithm to create an online MEMDP algorithm and established its superiority over POMDP algorithms
Exploited the sparse transitions in MEMDPs to have faster belief updates [O(n) as opposed to O(n^2)] to get faster processing
Made POMDP policy iteration more efficient by improving the update of Directed Cyclic Graphs to get 20% higher rewards
Designed algorithms to handle multiple DCGs & improve arbitrary subsets of DCG-nodes while ensuring Policy Improvement