works on

From the 1 of 50 linked papers with an AI index.

activity
20242026
collaborators
Showing cs.LGShow all

11 papers · 1 filter

cs.LG2026

The Bridge-Garden Dilemma in LLM Distillation: Why Mixing Hard and Soft Labels Works

Guanghui Wang, Kaiwen Lv Kacuila, Zhiyong Yang +5

Knowledge distillation (KD) transfers knowledge from a large teacher model to a smaller student. In language modeling, the student is trained either on tokens sampled from the teac…

cs.LG2026

The Devil is in the Condition Numbers: Why is GLU Better than non-GLU Structure?

Xingyu Lyu, Qianqian Xu, Zhiyong Yang +2

Gated Linear Units (GLU) and their variants are widely adopted in modern open-source large language model architectures and consistently outperform their non-gated counterparts, ye…

cs.LG2026

Localize and Neutralize: Gradient-guided Token Suppression against Visual Prompt Injection Attack

Dongpeng Zhang, Ke Ma, Yangbangyan Jiang +4

Adversarial images pose a severe security threat to multimodal large language models through prompt injection. Existing defenses largely lack a principled understanding of the unde…

cs.LG2026

DirMixE: Harnessing Test Agnostic Long-tail Recognition with Hierarchical Label Variations

Zhiyong Yang, Qianqian Xu, Sicong Li +3

This paper explores test-agnostic long-tail recognition, a challenging long-tail task where the test label distributions are unknown and arbitrarily imbalanced. We argue that the v…

cs.LG2025

TuckA: Hierarchical Compact Tensor Experts for Efficient Fine-Tuning

Qifeng Lei, Zhiyong Yang, Qianqian Xu +3

Efficiently fine-tuning pre-trained models for downstream tasks is a key challenge in the era of foundation models. Parameter-efficient fine-tuning (PEFT) presents a promising solu…

cs.LG2025

ABKD: Pursuing a Proper Allocation of the Probability Mass in Knowledge Distillation via --Divergence

Guanghui Wang, Zhiyong Yang, Zitai Wang +3

Knowledge Distillation (KD) transfers knowledge from a large teacher model to a smaller student model by minimizing the divergence between their output distributions, typically usi…