Tanmay Parekh

258 posts

Tanmay Parekh

Tanmay Parekh

@tparekh97

PhD student @UCLA | Fellowship @ Amazon, Bloomberg | Intern @Bloomberg @AIatMeta @AmazonLab126 | MLT @LTIatCMU | Applied Scientist @amazonIN | BTech @iitbombay

Katılım Nisan 2020
509 Takip Edilen865 Takipçiler
Sabitlenmiş Tweet
Tanmay Parekh
Tanmay Parekh@tparekh97·
📢 Excited to share that our paper introducing Divergent-Convergent Reasoning (DiCoRe) has been accepted at EMNLP Main 2025! 🎉🎊 Our paper introduces a three-component pipeline to dream big and openly, ground the thoughts to the task, and verify to improve precision! #EMNLP2025
Tanmay Parekh tweet media
Tanmay Parekh@tparekh97

🚨 New work: LLMs still struggle at Event Detection due to poor long-context reasoning and inability to follow task constraints, causing precision and recall errors. We introduce DiCoRe — a lightweight 3-stage Divergent-Convergent reasoning framework to fix this.🧵📷 (1/N)

English
0
6
42
5.3K
Tanmay Parekh
Tanmay Parekh@tparekh97·
Proposing a new “Parallel Exploration” Agentic framework PExA for complex code generation, specifically text-to-SQL. We achieve one of the best numbers on the Spider 2.0 leaderboard. Paper incoming on ArXiv soon! #agenticAI #text2sql #llm
Tech At Bloomberg@TechAtBloomberg

A team of researchers from Bloomberg’s #AI Engineering group introduced PExA, an #agenticAI framework that achieved 70.2% execution accuracy on the Spider 2.0 leaderboard, one of the most demanding benchmarks for #text2sql generation bloom.bg/4af28yB #CodeGeneration (1/2)

English
2
1
9
1.1K
Tanmay Parekh
Tanmay Parekh@tparekh97·
This is a very relevant and challenging problem to solve in the near future! No wonder such wide interests!
uclanlp@uclanlp

.@kaiwei_chang is getting a full house for his talk on “mathematical reasoning in visual context” at the Towards Comprehensive Reasoning in Vision-Language Models tutorial at #ICCV2025. Still time to come and engage in room 318A!

English
0
0
5
667
Tanmay Parekh
Tanmay Parekh@tparekh97·
It's my great pleasure to host my batchmate and an RL expert @aviral_kumar2 for giving a talk at the UCLA NLP Seminar Series. Please sign up if you want to attend the talk virtually and check out our website for future talks - uclanlp.github.io/nlp-seminar/
uclanlp@uclanlp

The UCLA NLP Seminar is back! 📢 This Friday, we are thrilled to host Aviral Kumar @aviral_kumar2 from CMU @CarnegieMellon to give a talk on test-time scaling! When: 2–3 PM (PDT), Friday, October 17th Registration: ucla.zoom.us/meeting/regist…

English
0
3
58
11.7K
Tanmay Parekh
Tanmay Parekh@tparekh97·
MoE LLMs are omni-present. Here, @LucasBandarkar presents a fantastic in-depth study of MoE routing for multilingual texts and how can interventions help with better steering. A good read for multilingual folks!
Lucas Bandarkar@LucasBandarkar

Multilingual Routing in Mixture-of-Experts LLMs We present (1) an in-depth analysis of how MoE LLMs route multilingual texts, with very clear patterns + (2) a router intervention (steering) method that leads to consistent multilingual improvements! 🧵1/4 arxiv.org/pdf/2510.04694

English
0
0
8
556
Tanmay Parekh retweetledi
Jia-Chen Gu
Jia-Chen Gu@Jiachen_Gu·
🚨Model editing in practice often collapses with catastrophic forgetting! Meet SPHERE🌐: an energy-regularized method that keeps weights uniformly distributed on hyperspheres, making sequential editing stable. Paper: arxiv.org/abs/2510.01172 Code: github.com/PlusLabNLP/SPH…
Jia-Chen Gu tweet media
English
1
8
16
2.8K
Tanmay Parekh retweetledi
Mohsen Fayyaz
Mohsen Fayyaz@mohsen_fayyaz·
🚨 You can bypass ALL safety guardrails of GPT-OSS-120B 🚨❗🤯 How? By detecting behavior-associated experts and switching them on/off. 📄 Steering MoE LLMs via Expert (De)Activation 🔗 arxiv.org/abs/2509.09660 🧵👇
English
5
22
131
36.7K
Tanmay Parekh
Tanmay Parekh@tparekh97·
In conclusion, our work introduces a new domain-aware task-specific data generation method using LLMs. Our work has been accepted as a main conference paper at #EMNLP2025. Paper link: arxiv.org/pdf/2502.17394
English
1
0
1
108
Tanmay Parekh
Tanmay Parekh@tparekh97·
🤔 Creating task-specific AI technologies for specialized domains is tough due to the lack of task-specific supervised data. 🚨 New work: We introduce SNaRe 🥁, a domain-aware task-specific data generation LLM pipeline. Accepted as main paper at #EMNLP2025 (1/N)
Tanmay Parekh tweet media
English
1
6
16
1.6K

Keşfet

@hbXNov @ChrisGPotts @yu_bryan_zhou @armancohan @aviral_kumar2 @LucasBandarkar @rohanwadhawan7 @IHung_Hsu