Irminsul: Efficient Position-Independent Caching for Agentic LLMs
Discover Irminsul, a novel MLA-native caching framework boosting efficiency and energy savings in agentic LLM serving with position-independent caching.
CFE-PPAR: Efficient Encryption for Privacy Action Recognition
Discover CFE-PPAR, a compression-friendly encryption method enhancing privacy-preserving action recognition with video transformers.
Temporal Functional Circuits for Accurate KAN Forecasting
Discover how Temporal Functional Circuits improve KAN forecasting with interpretable spline plots and enhanced prediction accuracy in time-series models.
PersonaTeaming: Enhancing AI Red-Teaming with Personas
Discover how PersonaTeaming boosts generative AI safety by integrating diverse human personas for more effective red-teaming and risk detection.
Mitigating Cross-Task Interference in Multi-Task LLM Training
Discover BADIT, a novel method to reduce cross-task interference in multi-task instruct-tuning of large language models for improved performance.