Sachin Kumar (@shocheen) / X

Sachin Kumar

588 posts

Sachin Kumar

@shocheen

Assistant Professor at @OhioStateCSE. Previous: @allen_ai, @UWNLP, @LTICMU. He/Him 🏳️‍🌈

Columbus, OH

Joined March 2011

Pinned
Sachin Kumar
@shocheen
Oct 28, 2025
🚨Check out our new preprint and an accompanying demo led by amazing @patrickqdasilva and @HananeNMoussa! We have developed an evaluation system for research ideas that provides detailed, actionable, and grounded feedback on the ideas' contribution and soundness
Hanane Nour Moussa
@HananeNMoussa
Oct 28, 2025
📢 As AI becomes increasingly explored for research idea generation, how can we rigorously evaluate the ideas it generates before committing time and resources to them? We introduce ScholarEval, a literature grounded framework for research idea evaluation across disciplines 👇!
00:00
2.4K
Sachin Kumar
@shocheen
Jul 8, 2023
📢 I am excited to share that I will be starting as an Assistant Professor @OhioState @OhioStateCSE in Fall 2024 🥳. Before that, I will spend a year as a postdoc in the @ai2_allennlp team @allen_ai!
37K
Sachin Kumar
@shocheen
Dec 3, 2023
Thanks for the shoutout Yu! I'm hiring multiple PhD students this cycle. If you are interested in building NLP solutions for real use cases that work for all people, those that can uniformly support diverse languages, domains, populations, and individuals, please apply.
Yu Su
@ysu_nlp
Dec 1, 2023
Replying to @kaiwei_chang @jieyuzhao11 and 13 others
Adding @shocheen who is joining @osunlp and looking for students! OSU is growing quickly in NLP
43K
Sachin Kumar
@shocheen
Dec 21, 2018
Check out our paper on "Von Mises-Fisher Loss for Training Sequence to Sequence Models with Continuous Outputs" (now accepted at #iclr2019) with my amazing advisor Prof. Yulia Tsvetkov:
openreview.net
Von Mises-Fisher Loss for Training Sequence to Sequence Models with...
Language generation using seq2seq models which produce word embeddings instead of a softmax based distribution over the vocabulary at each step enabling much faster training while maintaining...
Sachin Kumar
@shocheen
Dec 8, 2023
Come find me at 9 am today in the East Foyer!
Oreva Ahia
@orevaahia
Dec 8, 2023
Unfortunately, I can't attend EMNLP in person. @shocheen will be presenting our work in poster session 5 at 9am SGT on Saturday. Come talk to Sachin about the flaws of tokenization in multilingual LMs, tokenizer-free LMs and so much more about tokenization in general!
12K
Sachin Kumar
@shocheen
Dec 8, 2022
Super excited to introduce our #EMNLP2022 paper! MuCoLa: Gradient-based Constrained Sampling from Language Models. With @biswajitsc and Yulia Tsvetkov @tsvetshop arxiv.org/abs/2205.12558 (1/7)
Sachin Kumar
@shocheen
Jul 8, 2023
Replying to @shocheen
I will be recruiting Ph.D. students in the next academic cycle. If you are interested in working on all things NLP, please consider applying. Typing this as my flight to Toronto is about to take off, I will be at #ACL2023NLP starting tomorrow and happy to chat more there!
7.5K
Sachin Kumar
@shocheen
Aug 13, 2020
Couldn't agree more!
Anjalie Field
@anjalie_f
Aug 13, 2020
Dear advisors, please send your students a giant box of snacks, because my advisor did this and it’s the best thing ever
Sachin Kumar
@shocheen
Dec 3, 2023
Replying to @shocheen
Apropros to that, I'll be at EMNLP starting Dec 6 where I'll co-present a tutorial on mitigating societal harms in LLMs: llm-harm-mitigation.github.io with @anas_ant, @vidhisha_b and Lucille Njoo. Come check us out!
2K
Sachin Kumar
@shocheen
May 23, 2023
Please check out Oreva's amazing work on analyzing cost disparities in LLM APIs across languages.
Oreva Ahia
@orevaahia
May 23, 2023
Do you use LM APIs like ChatGPT in non-English languages? You might be overpaying and your in-context learning performance may also be negatively impacted. tinyurl.com/y6hakyc8 🧵⬇️
2K
Sachin Kumar
@shocheen
Dec 5, 2023
Replying to @mtschannen
Hi Michael, great paper! We also attempted something similar by predicting continuous vectors for text generation: arxiv.org/abs/1812.04616. This work was based on LSTMs but we have follow-up works such as this on the same idea using transformers as well: aclanthology.org/2021.acl-short…🙂
arxiv.org
Von Mises-Fisher Loss for Training Sequence to Sequence Models...
The Softmax function is used in the final layer of nearly all existing sequence-to-sequence models for language generation. However, it is usually the slowest layer to compute which limits the...
1.3K
Sachin Kumar
@shocheen
Jul 8, 2023
Replying to @shocheen
Extremely grateful to my incredible advisor Yulia Tsvetkov, and all my collaborators and friends @tsvetshop, @LTIatCMU, @uwnlp for all your help and support.
1.9K
Sachin Kumar
@shocheen
Feb 28, 2023
Check out our new survey paper on mitigating harm from language generation models!
Vidhisha Balachandran
@vidhisha_b
Feb 28, 2023
We’re super excited to share our #eacl2023 paper - Language Generation Models Can Cause Harm: So What Can We Do About It? An Actionable Survey Paper: arxiv.org/abs/2210.07700 w/ @shocheen, Lucille Njoo, @anas_ant, Yulia Tsvetkov from @tsvetshop lab. 1/7
2.6K
Sachin Kumar
@shocheen
Nov 5, 2020
Heisenberg's Election Principle. Source: 9gag