Balanced Session based Recommendation Systems for robustness to changing environments, TCS Research Labs
Popularity bias in Recommendation Systems pose serious problems when less popular/new items are present. This hurts even more in scenarios having covariate shift in terms of user distribution. A robust and balanced Deep Learning based Recommendation System was proposed which showed significant gains wrt click/buy rate even when user distribution changes.
Mitigating the effect of popularity bias in Session-based Recommendations, TCS Research Labs
Mitigating popularity bias is crucial to ensure that less popular but relevant items are part of the recommendation list shown to the user. This phenomenon of popularity bias in session-based Recommendation Systems was studied at data generation stage as well as model training stage and a comprehensive causal inference framework was proposed which identifies and mitigates the effect at both stages.
Effect of lockdown interventions to control the COVID-19 epidemic in India, TCS Research Labs
Different lockdown intervention strategies/agents to control the spread of COVID-19 virus were studied and their impact was compared in the states of Maharashtra and Tamil Nadu. A simulator was built using detailed demographic data and several agents were compared one of which was a Reinforcement Learning based Deep Q Learning agent.
Ad/Offer effect estimation in high dimensions, TCS Research Labs
A Deep Learning based solution was proposed which was designed to estimate the incremental effect of an ad/offer in scenarios where covariates are high dimensional and number of ad/offers is large. SOTA results were obtained on benchmark datasets.
Meta-Learning for ad/offer effect estimation, TCS Research Labs
Observational data accrued from multiple homogeneous subgroups of a heterogeneous population can be used to meta-learn a model which would generalize well in cases when training set is small. MetaCI was proposed as a Deep Learning based solution in ad/offer effect estimation use case. Significant gains were reported as compared to other methods.
M.Tech thesis: Classification of proteins on the basis of thermal stability using supervised learning, IIIT-Delhi
Thesis Advisors: Dr. Debajyoti Bera and Dr. Ganesh Bagler
Analysis of feature-model relationship over self-curated and statistically validated dataset and use of a combination of standard and newly proposed features to address the problem of protein classification.
Community Identification in Egocentric social networks, IIIT-Delhi
Advisor: Dr. Saket Anand
Use of a series of supervised machine learning classifiers over egonets to identify communities in social networks and performing inferential analysis.
Structural Motif Identification using algorithmic approach, IIIT-Delhi
Advisors: Dr. Debarka Sengupta and Dr. Ganesh Bagler
Use of various algorithmic approaches to solve the hard problem of network motif identification.
Monte Carlo simulation of Apoptosome formation, IIIT-Delhi
Advisor: Dr. Subhadip Raychaudhuri
Probability based Monte Carlo simulation of the process of apotosome formation and analysis of the results.
Bayesian Network learning from multi parameter Single-Cell Data, IIIT-Delhi
Advisor: Dr. Chetan Arora
Use of concepts of Probabilistic Graphical Models to learn a directed network from the data and to infer causal relationships from it.
Inferring spice-disease associations, IIIT-Delhi
Advisor: Dr. Ganesh Bagler
Application of association rule mining algorithms in order to discover spice-disease associations.
Smart Health Portal, IIIT-Delhi
Advisor: Dr. Tanusri Bhattacharya
Health portal with social networking facility using OOPD.