Udit Arora

I am a masters student in Computer Science at New York University and a member of the ML2 Group at CILVR, where I am advised by Prof. He He.

Prior to this, I worked as a Research Engineer at Product Labs, IIIT Hyderabad, leading project teams applying research in the domain of object detection/tracking, document querying and speech processing to develop commercially viable products. I also worked as a Research Assistant at Laboratory for Computational Social Systems, IIIT Delhi, solving research problems in the domain of blackmarket-driven social media fraud.

I completed my bachelors in Computer Engineering from NSIT, University of Delhi, following which I worked at Microsoft as a Software Engineer in the Office Product Group, where I delivered impactful work for products like Excel, Kaizala and Skype for Business.

Email  /  CV  /  Google Scholar  /  Github  /  LinkedIn  /  Flashy Website

profile photo

I'm interested in natural language processing, multimodal learning and robust machine learning. My research interests are centered around developing robust and intelligent systems that can solve impactful problems.

Types of Out-of-distribution Texts and How to Detect Them
Udit Arora, William Huang, He He
Empirical Methods in Natural Language Processing (EMNLP), 2021

Categorization of different ID/OOD text dataset pairs based on the type of distribution shift, and evaluation of the OOD detection performance of the two most common detection methods - density estimation and calibration - on these datasets.

ABOME: A Multi-platform Data Repository of Artificially Boosted Online Media Entities
Hridoy Sankar Dutta, Udit Arora, Tanmoy Chakraborty
International AAAI Conference on Web and Social Media (ICWSM), 2021

A multi-platform data repository consisting of artificially boosted (also known as blackmarket-driven collusive entities) online media entities such as Twitter tweets/users and YouTube videos/channels, which are prevalent but often unnoticed in online media.

MAST: Multimodal abstractive summarization with trimodal hierarchical attention
Aman Khullar*, Udit Arora*
EMNLP Workshop on NLP Beyond Text, 2020

Multimodal summarization by utilizing information from all three modalities of a video and trimodal hierarchical attention.

(* = equal contribution)
Analyzing and Detecting Collusive Users Involved in Blackmarket Retweeting Activities
Udit Arora, Hridoy Sankar Dutta, Brihi Joshi, Aditya Chetan, Tanmoy Chakraborty
ACM Transactions on Intelligent Systems and Technology (TIST), 2020
PDF | Code and dataset

Detection of users involved in blackmarket-driven retweeting activities on Twitter using a multiview learning based approach to encapsulate different views of information about the users.

Multitask Learning for Blackmarket Tweet Detection
Udit Arora, William Scott Paka, Tanmoy Chakraborty
2019 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining (ASONAM ’19), 2019

Detection of tweets posted on blackmarket services to gain inorganic appraisals - using a multitask learning based framework.

OSAVA: An Android App for Teaching a Course on Operating Systems
Pinaki Chakraborty*, Udit Arora*, Namrata Mukhija, et. al.
Journal of Enginering Education Transformations (JEET), 2019

An Android app named Operating System Algorithms Visualization App (OSAVA) to visualize different types of algorithms used in operating systems. It was used to teach a course on operating systems.

Children aged 6-24 months like to watch YouTube videos but could not learn anything from them
Savita Yadav, Pinaki Chakraborty, Prabhat Mittal, Udit Arora
Acta Paediatrica, 2018

Parents sometimes show young children YouTube videos on their smartphones. We studied the interaction of 55 Indian children born between December 2014 and May 2015 who watched YouTube videos when they were 6–24 months old.