Yifei Wang (on the job market!)
Postdoc at MIT CSAIL
I am a machine learning researcher at MIT CSAIL, advised by Stefanie Jegelka. My research has contributed to theoretical principles of foundation models (generative and representation models) and efficient algorithms for model capabilities and safety:
- Mathematical Principles of Foundation Models. I developed theoretical foundations for a broad spectrum of self-supervised learning (SSL) methods used in pretraining, encompassing contrastive [1, 2], non-contrastive [3], reconstructive [4], autoregressive [5], and predictive [6, 7] approaches. I also analyzed the feature dynamics in deep neural networks [8, 9] and pioneered a rigorous understanding [10] of LLMs’ self-correction ability, a critical component for test-time reasoning.
- Improving Model Capabilities. I leveraged these principles to “debug” foundation models. I addressed the rank collapse issue in neural networks [11, 12], generalized self-supervised learning with unsupervised world models [13]; alleviated data shortage by adaptively utilizing AI-generated data [14], and enhanced LLM long-context understanding with novel perplexity measures [15].
- Trustworthy Foundation Models. I contributed to theory-inspired algorithms to build trustworthy foundation models with respect to adversarial robustness [16, 17, 18, 11], feature interpretability [19, 20], and domain generalization [21, 22, 23]. In particular, I contributed to the use of LLMs’ own emergent abilities (such as in-context learning [24] and self-correction [10]) for jailbreaking and defending LLMs (featured and scaled up by Anthropic).
My first-author papers received the Best ML Paper Award at ECML-PKDD 2021, the Silver Best Paper Award at ICML 2021 AdvML workshop, and the Best Paper Award at ICML 2024 ICL workshop. My thesis won the CAAI Outstanding Ph.D. Dissertation Runner-Up Award. I published 33 papers at NeurIPS, ICLR, and ICML, and I am a (co)-first author on 22 of them.
I served as an area chair for ICLR 2024 and 2025 and as a regular reviewer for main AI/ML conferences (NeurIPS, ICML, ECML, AISTATS, LoG, CVPR, ACL). I co-organized the NeurIPS 2024 Workshop on Red Teaming GenAI and the MIT ML Tea Seminar.
I obtained my PhD in Applied Mathematics from Peking University in 2023, advised by Yisen Wang, Zhouchen Lin, Jiansheng Yang. Prior to that, I did my undergraduate at School of Mathematical Sciences, Peking University.
news
December, 2024 | I gave a talk on Principles of Foundations Models at Johns Hopkins University. |
---|---|
November, 2024 | I gave a guest lecture on Towards Test-time Self-supervised Learning (slides) at Boston College. |
October, 2024 | 3 new preprints are out, exploring 1) how existing long-context training of LLMs is problematic and how to address it (paper), 2) how sparse autoencoders can significantly improve robustness at noisy and few-shot scenarios (paper), and 3) whether ICL can truly extrapolate to OOD scenarios (paper). |
October, 2024 | 6 papers were accepted to NeurIPS 2024. We inverstigated how LLMs perform self-correction at test time (paper), how to build dynamic world models through joint embedding methods (paper), how Transformers avoid feature collapse with LayerNorm and attention masks (paper), and why equivariant prediction of data corruptions helps learn good representations (paper). |
September, 2024 | I gave a talk at NYU Tandon on Building Safe Foundation Models from Principled Understanding. |
August, 2024 | I gave a talk at Princeton University on Reimagining Self-Supervised Learning with Context. |
selected publications
- arXiv
LLM training & eval What is Wrong with Perplexity for Long-context Language Modeling?arXiv preprint arXiv:2410.23771, 2024We proposed a long-context perplexity measure that emphasizes long-context relevant tokens at training and evaluation, improving benchmark scores on LongBench, LongEval, and RULER by up to 22%. - NeurIPS
Best Paper Award
at ICML-W’24A Theoretical Understanding of Self-Correction through In-context AlignmentIn NeurIPS, 2024Best Paper Award at ICML 2024 ICL Workshop
We proposed the first theoretical explanation of how LLM self-correction works (as in OpenAI o1) and showed its effectiveness against social bias and jailbreak attacks. - NeurIPS
Oral at NeurIPS-W’24 In-Context Symmetries: Self-Supervised Learning through Contextual World ModelsIn NeurIPS, 2024Oral Presentation (top 4) at NeurIPS 2024 SSL Workshop
We introduced unsupervised test-time adaptation ability to self-supervised learning through a contextual world model designed for joint embedding (JEPA) models. - arXiv
Featured by Anthropic Jailbreak and guard aligned language models with only few in-context demonstrationsarXiv preprint arXiv:2310.06387, 2023Cited over 140 times. Featured and scaled up in Anthropic’s research blog, where it successfully demonstrated jailbreaking prominent LLMs, including GPT and Claude. - ICLRChaos is a Ladder: A New Theoretical Understanding of Contrastive Learning via Augmentation OverlapIn ICLR, 2022Cited over 120 times. We derived tight generalization bounds for contrastive learning with a new realistic theoretical framework. It derived unsupervised evaluation metrics with 97% correlation to downstream performance.
- ICLR
Silver Best Paper
at ICML-W’21A Unified Contrastive Energy-based Model for Understanding the Generative Ability of Adversarial TrainingIn ICLR, 2022Silver Best Paper Award at ICML 2021 AdvML workshop
From an energy-based perspective, we formulated contrastive learning as a generative model, and established the connection between adversarial training and maximum likelihood, thus briding generative and discriminative models together. - ECML-PKDD
Best ML Paper Award Reparameterized Sampling for Generative Adversarial NetworksIn ECML-PKDD, 2021Best ML Paper Award (1/685), invited to Machine Learning
We explored using GAN discriminator (as a good reward model) to bootstrap sample quality through an efficient MCMC algorithm, which not only guarantees theoretical convergence but also improves sample efficiency and quality in practice.