Hui Liu
9 papers · Latest:
Model-Agnostic Lifelong LLM Safety via Externalized Attack-Defense Co-Evolution
EvoSafety introduces a novel framework for lifelong, model-agnostic LLM safety via externalized attack-defense co-evolution to counter adversarial prompts.
Characterizing the Failure Modes of LLMs in Resolving Real-World GitHub Issues
This paper analyzes LLM failures in resolving GitHub issues, revealing strategy formulation as the most error-prone stage and localization as the least.
An Embarrassingly Simple Graph Heuristic Reveals Shortcut-Solvable Benchmarks for Sequential Recommendation
A simple graph heuristic reveals many sequential recommendation benchmarks are "shortcut-solvable," outperforming complex generative models.
Crafting Reversible SFT Behaviors in Large Language Models
This paper introduces LCDD to create sparse, controllable "carriers" for SFT behaviors in LLMs, enabling their selective reversal with SFT-Eraser.
Genus-protected higher-order topological phases
This paper introduces a new class of higher-order topological phases protected by the system's global topology (genus), independent of lattice symmetries.
FedCRF: A Federated Cross-domain Recommendation Method with Semantic-driven Deep Knowledge Fusion
FedCRF enables privacy-preserving cross-domain recommendation in non-overlapping scenarios via federated semantic learning and deep knowledge fusion.
Federated User Behavior Modeling for Privacy-Preserving LLM Recommendation
SF-UBM proposes a privacy-preserving federated LLM recommendation system that uses semantic bridging and knowledge distillation for non-overlapping domains.
Task-Aware LLM Routing with Multi-Level Task-Profile-Guided Data Synthesis for Cold-Start Scenarios
This paper introduces a data synthesis framework and TRouter for effective LLM routing in cold-start scenarios, improving performance and cost efficiency.
Baichuan 2: Open Large-scale Language Models
Baichuan 2 is a series of large-scale, open-source multilingual language models that achieve state-of-the-art performance across general and specialized benchmarks.
📬 Weekly AI Paper Digest
Get the top 10 AI/ML arXiv papers from the week — summarized, scored, and delivered to your inbox every Monday.