Jan 2022 – Present · 4 yrs 4 mos · Mountain View, California, United States

Jun 2021 – Sep 2021 · 3 mos · Sunnyvale, California, United States

• Host/Mentors: Dr. Spandana Gella, Prof. Mohit Bansal, Prof. Jesse Thomason
• Manager: Dr. Dilek Hakkani-Tur
• Team: Amazon Alexa AI
• Project Description: I worked on identifying biases in Vision-Language-Navigation models for ALFRED benchmark. Specifically, we show that ALFRED models such as Episodic Transformer (ET) are heavily biased towards vision inputs and are language agnostic. We improve generalization performance of ET model through syntax based pre-training of language encoder, curriculum learning, and multi-task learning.

Jul 2020 – Sep 2020 · 2 mos · Los Angeles, California, United States

• Host/Mentors: Dr. Soravit Changpinyo, Piyush Sharma, Dr. Boqing Gong
• Manager: Dr. Radu Soricut
• Team: GARCON Vision and Language Grounding
• Project Description: I proposed a visual question generation (VQG) module that facilitate in systematically evaluating cross-dataset adaptation capabilities of VQA models. Specifically, using our proposed VQG module, we generate out-of-domain test sets for source and target datasets by controlling and disentangling distribution shifts in vision and language features.
• [Published in EMNLP 2021]

Jul 2019 – Sep 2019 · 2 mos · Palo Alto, California, United States

• Hosts/Mentors: Dr. Spandana Gella, Prof. Siva Reddy
• Manager: Dr. Yaser Al-Onaizan
• Project Description: I worked on visual referring expression (RER) comprehension problem where the goal is to ground natural language expressions in images. We examine SOTA models (such as ViLBERT and MattNet) and show that they fail to perform reasoning on the linguistic structure.
• [Published this work in ACL 2020.]

Mar 2014 – Sep 2016 · 2 yrs 6 mos · India

• Team: Cognitive Research and Language Technologies
• Manager: Dr. Gargi B Dasgupta
• I worked on a wide array of natural language processing and machine learning projects at IBM Research.
• Following is a brief overview of my research work at IBM:
• 1. Implemented a prototype to answer natural language queries in IT Services Domain using IBM
• Watson Question Answering System.
• 2. Designed and developed a web based reporting tool to measure adoption, utilization and business benefits of dynamic automations in IT services delivery research. Many business units of IBM are now using this tool for auto-remediation of IT service tickets.
• 3. Proposed a new set of Discourse relations among IT service tickets to improve classification accuracy of tickets. Filed a US patent application on this.
• 4. Designed and developed a novel classification algorithm to classify problem tickets in IT Incident Management Process. Published a paper on this in ICSOC 2014, a top-tier conference in IT services domain.