Picture for Taiji Suzuki

Taiji Suzuki

Learning sum of diverse features: computational hardness and efficient gradient-based training for ridge combinations

Add code
Jun 17, 2024
Viaarxiv icon

Provably Neural Active Learning Succeeds via Prioritizing Perplexing Samples

Add code
Jun 06, 2024
Figure 1 for Provably Neural Active Learning Succeeds via Prioritizing Perplexing Samples
Figure 2 for Provably Neural Active Learning Succeeds via Prioritizing Perplexing Samples
Figure 3 for Provably Neural Active Learning Succeeds via Prioritizing Perplexing Samples
Figure 4 for Provably Neural Active Learning Succeeds via Prioritizing Perplexing Samples
Viaarxiv icon

High-Dimensional Kernel Methods under Covariate Shift: Data-Dependent Implicit Regularization

Add code
Jun 05, 2024
Viaarxiv icon

Neural network learns low-dimensional polynomials with SGD near the information-theoretic limit

Add code
Jun 03, 2024
Viaarxiv icon

Flow matching achieves minimax optimal convergence

Add code
May 31, 2024
Viaarxiv icon

State Space Models are Comparable to Transformers in Estimating Functions with Dynamic Smoothness

Add code
May 29, 2024
Viaarxiv icon

State-Free Inference of State-Space Models: The Transfer Function Approach

Add code
May 10, 2024
Viaarxiv icon

Mean-field Analysis on Two-layer Neural Networks from a Kernel Perspective

Add code
Apr 07, 2024
Viaarxiv icon

Mechanistic Design and Scaling of Hybrid Architectures

Add code
Mar 26, 2024
Figure 1 for Mechanistic Design and Scaling of Hybrid Architectures
Figure 2 for Mechanistic Design and Scaling of Hybrid Architectures
Figure 3 for Mechanistic Design and Scaling of Hybrid Architectures
Figure 4 for Mechanistic Design and Scaling of Hybrid Architectures
Viaarxiv icon

How do Transformers perform In-Context Autoregressive Learning?

Add code
Feb 08, 2024
Viaarxiv icon