Visual Degradation Risks in MLLM Safety and Jailbreaking
Explore how visual degradation weakens MLLM safety, enabling jailbreaking and learn about Structured Cognitive Offloading to enhance model security.
Mutual Reinforcement Learning for Diverse Language Models
Discover how Mutual Reinforcement Learning enables experience sharing among heterogeneous language models to boost AI learning efficiency.
Efficient KV Cache Eviction for Long-Context LLMs
Discover LaProx, a novel KV cache eviction method that boosts long-context LLM inference efficiency with reduced memory and runtime overhead.
CASCADE: Fast Context-Aware Speculative Image Decoding
CASCADE boosts image synthesis speed up to 3.6x using context-aware speculative decoding without extra training, improving AI image generation efficiency.
Improving EEG Decoding Reliability by Optimizing Preprocessing
Discover how preprocessing choices impact EEG decoding reliability and explore tools to measure and reduce prediction instability in brain data analysis.