Lune

学者档案2021-2026

Issei Sato

Department of Computer Science The University of Tokyo

16论文数
16末位作者
10合作者人数
0一作
0独著

一作与末位作者按论文署名顺序统计,部分顶会采用姓氏字母顺序署名。

近期论文16

  1. A Formal Comparison Between Chain of Thought and Latent Thought第 2 位,共 2 位作者ICML 2026 · 12 次引用
  2. Explaining Grokking and Information Bottleneck through Neural Collapse Emergence第 2 位,共 2 位作者ICLR 2026 · 5 次引用
  3. Understanding Generalization in Physics Informed Models through Affine Variety Dimensions第 2 位,共 2 位作者NeurIPS 2025 · 3 次引用
  4. Multiplicative Logit Adjustment Approximates Neural-Collapse-Aware Decision Boundary Adjustment第 2 位,共 2 位作者ICLR 2025
  5. On the Optimal Memorization Capacity of Transformers第 2 位,共 2 位作者ICLR 2025
  6. Theoretical Analysis of Hierarchical Language Recognition and Generation by Transformers without Positional Encoding第 2 位,共 2 位作者ACL 2025
  7. Are Transformers with One Layer Self-Attention Using Low-Rank Weight Matrices Universal Approximators?第 2 位,共 2 位作者ICLR 2024 · 31 次引用
  8. Understanding Linear Probing then Fine-tuning Language Models from NTK Perspective第 2 位,共 2 位作者NeurIPS 2024 · 30 次引用
  9. Understanding the Expressivity and Trainability of Fourier Neural Operator: A Mean-Field Perspective第 4 位,共 4 位作者NeurIPS 2024 · 13 次引用
  10. Exploring Weight Balancing on Long-Tailed Recognition Problem第 2 位,共 2 位作者ICLR 2024 · 10 次引用
  11. Top-Down Bayesian Posterior Sampling for Sum-Product Networks第 2 位,共 2 位作者KDD 2024
  12. Neural Lagrangian Schrödinger Bridge: Diffusion Modeling for Population Dynamics第 2 位,共 2 位作者ICLR 2023 · 1 次引用
  13. A Closer Look at Prototype Classifier for Few-shot Image Classification第 2 位,共 2 位作者NeurIPS 2022 · 42 次引用
  14. Disentanglement Analysis with Partial Information Decomposition第 2 位,共 2 位作者ICLR 2022 · 16 次引用
  15. Analyzing Lottery Ticket Hypothesis from PAC-Bayesian Theory Perspective第 2 位,共 2 位作者NeurIPS 2022 · 11 次引用
  16. Understanding Negative Samples in Instance Discriminative Self-supervised Representation Learning第 2 位,共 2 位作者NeurIPS 2021 · 56 次引用
黄昏的海面,两侧是细线勾勒的悬崖