Tengyu Ma
where else to find them
no profile links on file
what they talk about
where they keep going back
no series more than once — all 7 in the filters
on a stage with
never shared a talk
who talks about the same things
nobody close enough yet
| title | conference | date |
|---|---|---|
| RAG in 2025: State of the Art and the Road Forward | AI Engineer World's Fair 2025 | 2025-06-03 |
Argues that RAG beats fine-tuning and naive long-context for enterprise use, cites an average retrieval accuracy of ~80% across ~100 datasets with wide… | ||
| Toward Understanding the Extrapolation of Nonlinear Models to Unseen Domains or the Whole Domain | Domain Adaptation and Related Areas 2024 | 2024-11-12 |
Shows a negative result that wide neural nets and RKHS contain arbitrarily spiky functions, so multiplicative extrapolation can be infinite unless the model… | ||
| Toward Understanding In-context Learning | Special Year on Large Language Models and Transformers, Part 1 Boot Camp | 2024-09-04 |
Argues that in‑context learning comes from pretraining on a mixture of HMMs so next‑token Bayes inference implicitly infers a hidden model; proves… | ||
| Understanding Generalization from Pre-training Loss to Downstream Tasks | Modern Paradigms in Generalization Boot Camp 2024 | 2024-08-27 |
Contrastive pretraining provably produces embeddings where contrastive loss → spectral clustering and Euclidean distance ≈ manifold distance on the population… | ||
| Beyond NTK: A Mean-Field Analysis of Neural Networks with Polynomial Width, Samples, and Time | Optimization and Algorithm Design 2023 | 2023-11-27 |
Claim: gradient-flow mean-field analysis proves unmodified SGD learns single-index targets with quartic activations using polynomial width, iterations, and… | ||
| Practical Model-based Algorithms for Reinforcement Learning and Imitation Learning, with... | Frontiers of Deep Learning 2019 | 2019-07-15 |
Proposes a model-based RL recipe that learns neural-network dynamics with a reward-aware loss (no squared error), plugs that model into a planner solved by… | ||
| Learning One-hidden-layer Neural Networks with Landscape Design | Optimization, Statistics and Uncertainty | 2017-11-28 |
Designs a new population objective g that uses Hermite-polynomial coefficients to match Hermite coefficients between training activation gamma and data… | ||
| nothing matches | ||