Sean Welleck

About Me

Assistant Professor
Carnegie Mellon University
School of Computer Science
Language Technologies Institute
PI: L3 Lab

I received a PhD at New York University, advised by Kyunghyun Cho, and did a postdoc at the University of Washington, advised by Yejin Choi.
I host the Thesis Review Podcast.

Research

My research focuses on deep learning and generative models, including:

• Machine learning for code and mathematics
• Learning, inference, and evaluation algorithms
• Science of neural language models

Group (L3 Lab)

PhD

• Pranjal Aggarwal
• Weihua Du (co-advised with Yiming Yang)
• Andre He (co-advised with Daniel Fried)
• Seungone Kim (co-advised with Graham Neubig)

MS

• Riyaz Ahuja
• Jiewen Hu
• Qinyue Tan
• Thomas Zhu

Please see the CMU L3 Lab to learn more about our research.

Recent & Upcoming

• 01.2025: Invited talk at UCLA NLP Seminar
• 12.2024: NeurIPS tutorial on Inference-time Algorithms for Large Language Models
• 11.2024: Invited talk at Simons Institute [talk page]
• 09.2024: Invited talk on Reasoning with inference-time compute at CMU [slides]
• 09.2024: Invited talk on A few open problems in neural theorem proving at AITP [slides]
• 06.2024: Invited talk at Neurosymbolic Programming Summer School
• 05.2024: Invited talk at PNNL Mathematics for Artificial Reasoning in Science
• 04.2024: Version II of Neural Theorem Proving Tutorial [github][slides]
• 04.2024: LLMLean: LLMs+Lean on your laptop or in the cloud [github]
• 04.2024: Invited talk on Scaling laws of formal reasoning at SciFM [slides]
• 04.2024: Invited tutorial at MICDE Scientific Foundation Models Conference [slides][github]
• 01.2024: New Neural code generation course with Daniel Fried at CMU
• 12.2023: Co-organized Math+AI workshop at Neurips
• 12.2023: Panel on Machine Learning for Theorem Proving at Neurips
• 11.2023: Invited talk on Language models and formal mathematics at UW Mathematics [slides]
• 10.2023: Invited talk on Llemma at Harvard New Technologies in Mathematics [slides]
• 10.2023: Llemma: an open language model for mathematics [paper][code/data/models]
• 10.2023: llmstep: language model proofstep suggestions in Lean [paper][github]
• 08.2023: ntptutorial: interactive notebook tutorial on neural theorem proving [github]
• 08.2023: IJCAI tutorial on neural theorem proving [slides]
• 06.2023: Panel on Generative AI in Finance and Economics; CMU
• 06.2023: Invited talk at AI to Assist Mathematical Reasoning Workshop; NASEM [slides]
• 05.2023: Invited talk on Draft, Sketch, Prove at Hoskinson Center for Formal Mathematics. [slides]

Papers

• From Decoding to Meta-Generation:
  Inference-time Algorithms for Large Language Models
  S. Welleck, A. Bertsch*, M. Finlayson*, H. Schoelkopf*, A. Xie, G. Neubig, I. Kulikov, Z. Harchaoui
  TMLR 2024
• Inference Scaling Laws: An Empirical Analysis of Compute-Optimal Inference for Problem-Solving with Language Models
  Y. Wu, Z. Sun, S. Li, S. Welleck, Y. Yang
  arXiv 2024
• ImProver: Agent-Based Automated Proof Optimization
  R. Ahuja, J. Avigad, P. Tetali, S. Welleck
  arXiv 2024
  [project page][code]
• Lean-STaR: Learning to Interleave Thinking and Proving
  H. Lin, Z. Sun, Y. Yang, S. Welleck
  arXiv 2024
  [project page][code]
• miniCTX: Neural Theorem Proving with (Long-)Contexts
  J. Hu, T. Zhu, S. Welleck
  arXiv 2024
  [project page][eval code][ntp-toolkit]
• miniCodeProps: a Minimal Benchmark for Proving Code Properties
  E. Lohn, S. Welleck
  arXiv 2024
  [code]
• The BiGGen Bench: A Principled Benchmark for Fine-grained Evaluation of Language Models with Language Models
  S. Kim, J. Suk, J. Cho, S. Longpre et al., …, S. Welleck, G. Neubig, M. Lee, K. Lee, M. Seo
  arXiv 2024
  [data][leaderboard]
• Easy-to-Hard Generalization: Scalable Alignment Beyond Human Supervision
  Z. Sun*, L. Yu*, Y. Shen, W. Liu, Y. Yang+, S. Welleck+, C. Gan+
  NeurIPS 2024
  [code]
• Prometheus 2: An Open Source Language Model Specialized in Evaluating Other Language Models
  S. Kim, J. Suk, S. Longpre, B. Lin, J. Shin, S. Welleck, G. Neubig, M. Lee, K. Lee, M. Seo
  EMNLP 2024
  [code]
• Llemma: An Open Language Model for Mathematics
  Z. Azerbayev, H. Schoelkopf, K. Paster, M. Dos Santos, S. McAleer, A. Jiang, J. Deng, S. Biderman, S. Welleck
  ICLR 2024
  [code][data][models][blog][sample explorer][poster]
• LLMstep: LLM proofstep suggestions in Lean
  S. Welleck, R. Saha
  NeurIPS Math-AI Workshop 2023
  [code][poster]
• Inference-Time Policy Adaptors
  X. Lu, F. Brahman, P. West, J. Jung, K. Chandu, A. Ravichander, L. Qin. P. Ammanabrolu, L. Jiang, S. Ramnath, N. Dziri, J. Fisher, B. Lin, S. Hallinan, X. Ren, S. Welleck, Y. Choi
  EMNLP 2023
• Self-refine: Iterative refinement with self-feedback
  A. Madaan, N. Tandon, P. Gupta, S. Hallinan, L. Gao, S. Wiegreffe, U. Alon, No. Dziri, S. Prabhumoye, Y. Yang, B. Prasad Majumder, S. Gupta, S. Welleck, A. Yazdanbakhsh, P. Clark
  NeurIPS 2023
• Limits of Transformers on Compositionality
  N. Dziri*, X. Lu*, M. Sclar*, X. Lorraine Li, L. Jiang, B. Lin, P. West, C. Bhagavatula, R. Le Bras, J. D Hwang, So. Sanyal, S. Welleck, X. Ren, A. Ettinger, Z. Harchaoui, Y. Choi
  NeurIPS 2023 (Spotlight)
• STEER: Unified Style Transfer with Expert Reinforcement
  S. Hallinan, F. Brahman, X. Lu, J. Jung, S. Welleck, Y. Choi
  EMNLP 2023
• A Survey of Deep Learning for Mathematical Reasoning
  P. Lu, L. Qiu, W. Yu, S. Welleck*, K. Chang*
  ACL 2023
• Generating Sequences by Learning to [Self-]Correct
  S. Welleck*, X. Lu*, P. West+, F. Brahman+, T. Shen, D. Khashabi, Y. Choi
  ICLR 2023.
  [poster]
• Draft, Sketch, and Prove: Guiding Formal Theorem Provers with Informal Proofs
  A. Jiang*, S. Welleck*, J. Zhou*, T. Lacroix, J. Liu, W. Li, M. Jamnik, G. Lample+, Y. Wu+
  ICLR 2023(Oral).
  [data]
• NaturalProver: Grounded Mathematical Proof Generation with Language Models
  S. Welleck, J. Liu, X. Lu, H. Hajishirzi, Y. Choi.
  NeurIPS 2022.
  [data][code][slides][poster]
• Quark: Controllable Text Generation with Reinforced [Un]learning
  X. Lu, S. Welleck, L. Jiang, J. Hessel, L. Qin, P. West, P. Ammanabrolu, Y. Choi.
  NeurIPS 2022 (Oral).
  [poster]
• COLD Decoding: Energy-based Constrained Text Generation with Langevin Dynamics
  L. Qin, S. Welleck, D. Khasabi, Y. Choi.
  NeurIPS 2022 (Oral).
  [code][poster]
• Maieutic Prompting: Logically Consistent Reasoning with Recursive Explanations
  J. Jung, L. Qin, S. Welleck, F. Brahman, C. Bhagavatula, R. Le Bras, Y. Choi.
  EMNLP 2022.
  [code][slides]
• Lila: A Unified Benchmark for Mathematical Reasoning
  S. Mishra, M. Finlayson, P. Lu, L. Tang, S. Welleck, C. Baral, T. Rajpurohit, O. Tafjord, A. Sabharwal, P. Clark, A. Kalyan
  EMNLP 2022.
  [data][project page]
• Rainier: Reinforced Knowledge Introspector for Commonsense Question Answering
  J. Liu, S. Hallinan, X. Lu, P. He, S. Welleck, H. Hajishirzi, Y. Choi.
  EMNLP 2022.
• NeuroLogic A*esque Decoding: Constrained Text Generation with Lookahead Heuristics
  X. Lu, S. Welleck, P. West, L. Jiang, J. Kasai, D. Khasabi, R. Le Bras, L. Qin, Y. Yu, R. Zellers, N. Smith, Y. Choi.
  NAACL 2022.
  [code][slides]
• Symbolic Knowledge Distillation: from General Language Models to Commonsense Models
  P. West, C. Bhagavatula, J. Hessel, J. Hwang, L. Jiang, R. Le Bras, X. Lu, S. Welleck, Y. Choi.
  NAACL 2022.
  [code]
• Prompt Waywardness: The Curious Case of Discretized Interpretation of Continuous Prompts
  D. Khasabi, S. Lyu, S. Min, L. Qin, K. Richardson, S. Singh, S. Welleck, H. Hajishirzi, T. Khot, A. Sabharway, Y. Choi.
  NAACL 2022.
• Generated Knowledge Prompting for Commonsense Reasoning
  J. Liu, A. Liu, X. Lu, S. Welleck, P. West, R. Le Bras, Y. Choi, H. Hajishirzi.
  ACL 2022.
  [code]
• Symbolic Brittleness in Sequence Models: on Systematic Generalization in Symbolic Mathematics
  S. Welleck, P. West, J. Cao, Y. Choi.
  AAAI 2022.
  [code][slides][talk]
• Towards Grounded Natural Language Proof Generation
  S. Welleck, J. Liu, J. Han, Y. Choi.
  MathAI4Ed Workshop at NeurIPS 2021 (Contributed Talk).
  [poster][slides]
• NaturalProofs: Mathematical Theorem Proving in Natural Language
  S. Welleck, J. Liu, R. Le Bras, H. Hajishirzi, Y. Choi, K. Cho.
  NeurIPS 2021 Datasets and Benchmarks (Oral (Top 1%)).
  [data/code][talk][related data]
• MAUVE: Measuring the Gap Between Neural Text and Human Text using Divergence Frontiers
  K. Pillutla, S. Swayamdipta, R. Zellers, J. Thickstun, S. Welleck, Y. Choi, Z. Harchaoui.
  NeurIPS 2021 (Oral, Outstanding Paper Award (Top 0.1%)).
  [code][press]
• Divergence Frontiers for Generative Models: Sample Complexity, Quantization Level, and Frontier Integral
  L. Liu, K. Pillutla, S. Welleck, S. Oh, Y. Choi, Z. Harchaoui.
  NeurIPS 2021.
• Mode recovery in neural autoregressive sequence modeling
  I. Kulikov, S. Welleck, K. Cho.
  SPNLP 2021.
  [code]
• Order and Learning in Sequential Neural Structured Prediction
  S. Welleck.
  PhD Thesis, New York University.
  [slides]
• MLE-guided parameter search for task loss minimization in neural sequence modeling
  S. Welleck, K. Cho.
  AAAI 2021.
  [code][poster][talk]
• Consistency of a Recurrent Language Model With Respect to Incomplete Decoding
  S. Welleck, I. Kulikov, J. Kim, R. Pang, K. Cho.
  EMNLP 2020.
  [code][talk]
• A Generalized Framework of Sequence Generation with Application to Undirected Sequence Models
  E. Mansimov, A. Wang, S. Welleck, K. Cho.
  arXiv pre-print 2020.
  [code]
• Making Inconsistent Dialogue Unlikely with Unlikelihood Training
  M. Li, S. Roller, I. Kulikov, S. Welleck, Y.L. Boureau, K. Cho, J. Weston.
  ACL 2020.
• Neural Text Generation with Unlikelihood Training
  S. Welleck, I. Kulikov, S. Roller, E. Dinan, K. Cho, J. Weston.
  ICLR 2020.
  [code]
• Non-Monotonic Sequential Text Generation
  S. Welleck, K. Brantley, H. Daume III, K. Cho.
  ICML 2019.
  [code] [slides] [poster]
• Sequential Graph Dependency Parser
  S. Welleck, K. Cho.
  RANLP 2019.
  [slides]
• Dialogue Natural Language Inference
  S. Welleck, J. Weston, A. Szlam, K. Cho.
  ACL 2019.
  [dataset][poster][press]
• Loss Functions for Multiset Prediction
  S. Welleck, Z. Yao, Y. Gai, J. Mao, Z. Zhang, K. Cho.
  NeurIPS 2018.
  NVIDIA AI Labs Pioneering Research Award 2018.
  [poster]
• Saliency-based Sequential Image Attention with Multiset Prediction
  S. Welleck, J. Mao, K. Cho, Z. Zhang.
  NeurIPS 2017.
  NVIDIA AI Labs Pioneering Research Award 2017.
  [poster][press]
• Efficient AUC Optimization for Information Ranking Applications
  S. Welleck.
  ECIR 2016.

Selected Talks

• Beyond Decoding: Meta-Generation Algorithms for Large Language Models [talk page]
  Simons Institute for the Theory of Computing
  11.2024
• Reasoning with inference-time compute [slides]
  CMU
  09.2024
• A few open problems in neural theorem proving [slides]
  AITP 2024 Invited Talk
  09.2024
• Scaling laws of formal reasoning [slides]
  MICDE Scientific Foundation Models Conference Invited Talk
  04.2024
• A tutorial on neural theorem proving VII [slides][github]
  MICDE Scientific Foundation Models Conference Invited Tutorial
  04.2024
• Language models and formal mathematics [slides]
  UW Mathematics
  11.2023
• Llemma [slides]
  Harvard New Technologies in Mathematics
  10.2023
• A tutorial on neural theorem proving [slides]
  OIST
  09.2023
  IJCAI
  08.2023
• Data for AI + mathematics [slides]
  AI to Assist Mathematical Reasoning Workshop
  06.2023
• Draft, Sketch, Prove [slides]
  Hoskinson Center for Formal Mathematics
  05.2023
• Bridging Informal and Formal Mathematical Reasoning [slides]
  Microsoft Research
  11.2022
  NSF Expeditions: Understanding the World Through Code seminar
  10.2022
• Maeutic Inference [slides]
  Google
  09.2022
• Constrained text generation [slides]
  IFDS
  05.2022
  Cohere.ai
  05.2022

Teaching

• Neural Code Generation
  Carnegie Mellon University
  Spring 2024.
• Guest Lecture: Neural sequence generation (DATA 598) [slides]
  University of Washington
  March 2023.
• Guest Lecture: Reliable text generation through graph search (CSE 373) [slides]
  University of Washington
  November 2022.
• Guest Lecture: Neural sequence generation (DATA 598) [slides]
  University of Washington
  March 2022.
• Deep Learning (DS-GA 1008)
  New York University
  Fall 2020
• Deep Learning for NLP
  African Master’s Program in Machine Intelligence
  March 2020
• Introduction to Machine Learning (CSCI-UA 0473)
  New York University
  Spring 2020
• NLP with Representation Learning (DS-GA 1011)
  New York University
  Fall 2019

Tutorials

• Neural theorem proving II [slides][github]
  SciFM 2024.
• Neural theorem proving [slides][github]
  In Deep Learning in Mathematical Reasoning [tutorial site]
  IJCAI 2023.
• Neurosymbolic NLP: Modularity & Constraints for Neural Language Models [slides][tutorial site]
  COLING 2022.
• Denoising Diffusion Models [slides]
  July 2022.
• Generative Modeling with (W)GAN [slides]
  NYU Shanghai
  April 2018.

Workshops

• Math-AI: 3rd Workshop on Mathematical Reasoning and AI
  NeurIPS 2023
• Math-AI: Toward Human-Level Mathematical Reasoning
  NeurIPS 2022
• MathNLP: 1st Workshop on Mathematical Natural Language Processing
  EMNLP 2022
• Math AI for Education: Bridging the Gap Between Research and Smart Education
  NeurIPS 2021

Past

• NYU (PhD), Sep. 2016 - Jan. 2021
• Facebook, AI Research Team (FAIR), May. 2019 - Sep. 2019
• Facebook, AI Research Team (FAIR), May. 2018 - Jan. 2019
• Primer AI, Feb. 2016 - Aug. 2016
• IBM, Sep. 2014 - Feb. 2016
• University of Pennsylvania (Computer Science, MSE), May. 2013 - May. 2014
• University of Pennsylvania (Computer Science, BSE), Sep. 2009 - Feb. 2013