Siddhant Garg

@sid7954

#NLProc Scientist at @AmazonScience @alexa99, CS Grad @UWMadison, CS undergrad @iitbombay, ex intern @epfl @tuBraunschweig

San Francisco, CA

sid7954.github.io

Replies Reposts Media

Pinned
Siddhant Garg
@sid7954
Dec 2, 2020
Self-Supervised Learning ➡️ Great Empirical Success Why❓➡️Still theoretically unclear Why only certain SSL tasks perform well❓ How many samples are needed for learning❓ Our #NeurIPS2020 paper: A theoretical framework for representation learning 📰: arxiv.org/abs/2008.02447
2
Siddhant Garg
@sid7954
Jul 7, 2023
New #ACL2023 paper: arxiv.org/abs/2305.15344 Core Idea: Refining answer generation LLMs with supervision from automatic QA evaluators for open domain abstractive QA. Intuitively analogous to RLHF: by replacing human-preference reward model by an automatic QA evaluator model
1
Siddhant Garg
@sid7954
Dec 5, 2022
I'm attending #EMNLP2022 at Abu Dhabhi to present two papers with two amazing interns who worked with our team in Alexa AI this summer @AmazonScience. Please reach out to chat about Question Answering research in industry, Language Generation, Pre-training of LLMs, etc. 😃
1
Siddhant Garg
@sid7954
Nov 11, 2022
Sharing super cool #EMNLP2022 paper led by our intern @MatteoGabburo: arxiv.org/abs/2210.12865 Knowledge transfer from discriminative answer ranking models to train answer generation models (without any labeled data) to write self-contained natural answer sentences
1
Siddhant Garg
@sid7954
Jul 8, 2022
Excited to share our #NAACL22 paper arxiv.org/abs/2205.01228 lead by our amazing intern Luca Di Liello We propose a new pretraining task for multi-sentence transformer encoders exploiting para/doc-level semantics. Gets significant improvements for multi-sentence inference tasks
3