Tan Minh Nguyen

Assistant Professor@National University of Singapore·新加坡·OpenReview

研究方向

deep learning · control theory · PDEs · optimization · transformers · graph neural networks · mixture of experts · state-space models · mixture models · generative models

Tan Minh Nguyen

Expert Merging in Sparse Mixture of Experts with Nash Bargaining

Quasi-Equivariant Metanetworks

Mixed-Curvature Tree-Sliced Wasserstein Distance

Activation Steering with a Feedback Controller

Tree-sliced Sobolev IPM

Functional Equivalence in Attention: A Comprehensive Study with Applications to Linear Mode Connectivity

Discontinuities in Sparse Mixture-of-Experts: A Measure-Stochastic Analysis

Revisiting Tree-Sliced Wasserstein Distance Through the Lens of the Fermat–Weber Problem

Momentum Steering: Activation Steering Meets Optimization

Accelerating Transformer Training: Architectural Symmetry, Positional Encoding, and Teleportation

Modeling Expert Interactions in Sparse Mixture of Experts via Graph Structures

Equivariant Metanetworks for Mixture-of-Experts Weights

Token Dynamics on Spheres in Mamba Models

On Linear Mode Connectivity of Mixture-of-Experts Architectures

Angular Steering: Behavior Control via Rotation in Activation Space

Demystifying the Token Dynamics of Deep Selective State Space Models

Tight Clusters Make Specialized Experts

Equivariant Neural Functional Networks for Transformers

Dynamical Properties of Tokens in Self-Attention and Effects of Positional Encoding

CAMEx: Curvature-aware Merging of Experts

The Blessing and Curse of Dimensionality in Safety Alignment

Distance-Based Tree-Sliced Wasserstein Distance

Tree-Sliced Entropy Partial Transport

Spherical Tree-Sliced Wasserstein Distance

MoLEx: Mixture of Layer Experts for Fine-tuning with Sparse Upcycling

Transformer Meets Twicing: Harnessing Unattended Residual Information

Promoting Ensemble Diversity with Interactive Bayesian Distributional Robustness for Fine-tuning Foundation Models

Tree-Sliced Wasserstein Distance: A Geometric Perspective

Equivariant Polynomial Functional Networks

Projection Optimal Transport on Tree-Ordered Lines

Tree-Sliced Wasserstein Distance with Nonlinear Projection

Equivariant Polynomial Functional Networks

L-MSA: Layer-wise Fine-tuning using the Method of Successive Approximations