Pareto-Lenient Consensus for Efficient Multi-Preference LLM Alignment
Discover Pareto-Lenient Consensus, a game-theoretic method enhancing multi-preference LLM alignment by escaping local trade-offs for better results.
Trustworthy Report Generation with Confidence Estimation
Discover a deep research agent that enhances report trustworthiness using progressive confidence estimation and calibration for reliable AI-generated conte...
MARL-GPT: Unified GPT Model for Multi-Agent RL
Discover MARL-GPT, a scalable GPT-based model excelling in multi-agent reinforcement learning across diverse tasks without task-specific tuning.
Context-Value-Action Architecture for Value-Driven LLMs
Discover the Context-Value-Action architecture that reduces polarization and boosts fidelity in value-driven large language model agents.
HybridKV: Efficient KV Cache Compression for Multimodal LLMs
HybridKV compresses KV caches to boost multimodal LLM inference, reducing memory by 7.9x and speeding decoding by 1.5x without losing accuracy.