Richa Rastogi

[Email]

Hi! I'am a Computer Science PhD Candidate at Cornell University and I'am fortunate to be advised by Professor Thorsten Joachims.
My research interests lie in the area of learning from feedback in interactive systems ranging from recommender systems to LLMs. My recent work introduces a policy learning framework to tractably align these systems with long-term objectives. I am also interested in the study of bias, fairness and uncertainty in these systems. Prior to starting PhD, I spent several years working as a controls engineer at Amazon Fullfillment Technologies and other companies. Prior to that, I completed a Masters in Industrial Engineering and Operations Research at ISyE, Georgia Tech and a Bachelors of Engineering at Delhi University, India.

Selected Works (complete works at Google Scholar)

MultiScale Contextual Bandits for Long-term Objectives
Richa Rastogi , Yuta Saito, Thorsten Joachims.
Preprint, 2025
MultiScale Policy Learning to contextually reconcile the disconnect in the timescales of short-term interventions (e.g., rankings, token feedback) and the long-term feedback (e.g., user retention, sentence feedback). Our method learns interventions and policies at multiple interdependent timescales in various settings ranging from recommender systems to LLMs.


Fairness in Ranking under Disparate Uncertainty
Richa Rastogi , Thorsten Joachims.
ACM Conference on on Equity and Access in Algorithms, Mechanisms, and Optimization (EAAMO'24), Spotlight (Oral) at UAI Workshop, 2023
We introduce Equal-Opportunity Ranking (EOR) as a new fairness criterion for ranking and a practical algorithm that provably reduces the group unfairness induced when the uncertainty of the underlying relevance model differs between groups of candidates.
code | slides | bibtex | poster | Media: Cornell News

Semi-Parametric Inducing Point Networks and Neural Processes
Richa Rastogi , Yair Schiff, Alon Hacohen, Zhaozhi Li, Ian Lee, Yuntian Deng, Mert R Sabuncu, Volodymyr Kuleshov.
International Conference on Learning Representations (ICLR'23)
We introduce Semi-Parametric Inducing Point Networks (SPIN), a general-purpose architecture that can query the training set at inference time in a compute-efficient manner.
code | slides | bibtex

System and Method for placing labels on moving objects
Daniel Muething, Richa Rastogi , Keith C Tate, Anand H Doshi, John K Robinson, Mahmoud Abugharbieh.
USPTO Patent on behalf of Amazon, 2021
Learning closed loop dynamics for accurately positioning labels on moving packages.