Program (tentative)

Opening (10:00–10:10)

Session 1：Statistical Learning Theory & Generative Models

10:10–10:30
Tomoya Wakayama: Generalization Error of Mean-Field Shallow Neural Networks Trained by Wasserstein Gradient Flow

10:30–10:50
Wei Huang: On the Learnability of Diffusion Models on Low-Dimensional Manifolds

☕ 10:50–11:20 Coffee Break

11:20–11:40
Huanjian Zhou: The Adaptive Complexity of Sampling

11:40–12:00
Kazuma Sawaya: Provable False Discovery Rate Control for Deep Feature Selection (偽発見率制御可能な深層特徴量選択)

Lunch Break 🍱 (12:00–13:20)

Lightning Session (13:20–13:50)

 2 min flash talk × 11 posters

Poster Session (13:50–14:50)

Break ☕ (14:50–15:10)

Session 2：Dynamics, Optimization & Expressivity

15:10–15:30
Shuta Takeuchi: Counting Fixed Points in Randomly Coupled Dynamics: A Grassmann Algebra Approach (ランダム結合ダイナミクスにおける固定点数の評価：グラスマン代数を用いたアプローチ)

15:30–15:50
Tomoei Takahashi: Dynamical regimes of discrete diffusion models (離散拡散モデルの動的特性解析)

☕ 15:50–16:20 CoffeeBreak

16:20–16:40
Taira Tsuchiya: Online Learning and Game Theory: Regret Lower Bounds and Adaptive Learning Dynamics

16:40–17:00
Kevin Xu: On Expressive Power of Looped Language Models (ループ型言語モデルの表現力に関して)

Closing & Banquet (17:00–)

Poster (Lighting talk order)

1. Atsuyoshi Muta: Singularity-Induced Local Complexity of Learning Models via Singular Learning Theory

2. Taishi Kuzumoto: Locally Adaptive Inference on Nonparametric Regression Functions

3. Naoki Yoshida: Zero Generalization Error Theorem for Random Interpolators via Algebraic Geometry (訓練損失が0になるパラメータにおける汎化誤差の代数幾何的解析)

4. Razvan Lascu: Saddle-free Newton Method in Wasserstein Space

5. Shu Tamano: Theoretical Foundations of Generative Adversarial Network-Based Causal Inference

6. Sota Nishiyama: High-Dimensional Limit of Stochastic Gradient Descent: Continuous-Time Approximation and Dynamical Mean-Field Theory (確率的勾配降下法の高次元極限：連続時間近似と動的平均場理論)

7. Ryoya Awano: Pre-trained neural network learns diverse features through weak-to-strong generalization (事前学習ニューラルネットワークにおける弱至強汎化による多様な特徴の学習)

8. Johannes Ackermann: Gradient Regularization Prevents Reward Hacking in RLHF and RLVR

9. Naoki Nishikawa: Inference-time Alignment with Rewards in Besov Spaces: Advantages of Feature Learning and Multi-Step Policy Updates (Besov 空間上の報酬関数に対する推論時アラインメント)

10. Tokio Kajitsuka: Expressive Power of Transformer Architectures from Perspective of Memorization Capacity (記憶容量の観点に基づくTransformerアーキテクチャの表現能力)

11. Takeshi Koshizuka:  Understanding Generalization in Physics Informed Models through Affine Variety Dimensions (アフィン多様体次元に基づく物理情報付きモデルの汎化能力の解析)