KBBQ: A Predictive Noise Law and the Limits of Spectrum Flattening in FP4 Quantization
本文提出了一种新的量化噪声理论,并基于此开发了KBBQ方法,该方法在FP4量化中通过参数化变换接近上限的程度来提高性能。
本文提出了一种新的量化噪声理论,并基于此开发了KBBQ方法,该方法在FP4量化中通过参数化变换接近上限的程度来提高性能。
We present EmoMed - a multimodal medical consultation agent that adapts its responses based on users'emotional states while maintaining clinical accuracy. The system processes text and medical images, detects affect indicators (anxiety, confusion, urgency) from user input, and adjusts response tone, structure, and detail level accordingly. To ensure factual reliability, the agent grounds clinical information through a dual retrieval mechanism: web-based fact-checking and an API-connected, continuously updated medical knowledge base. We evaluate our approach across seven state-of-the-art language models (GPT-4/5, Qwen3, Llama 4, Gemini 2.5, Grok4, Claude3) using comprehensive metrics including LLM-as-judge assessments, MedQA style accuracy tests, BERT Score, safety/helpfulness ratings, and multimodal medical benchmarks. The results demonstrate that emotionally adaptive responses consistently outperform neutral baseline across evaluation dimensions, without compromising clinical accuracy. A controlled user study validated these findings, with participants reporting improved perceived empathy and communication clarity, while maintaining trust in factual accuracy. Source code: https://github.com/NasonovIvan/EmoMed-Agent
本文提出LoRA-TSD,通过在低秩矩阵流形的切空间中采用谱范数最速下降法优化LoRA,解决了独立训练因素忽略几何结构的问题。
研究通过因果神经元级分析,揭示视觉-语言模型中安全机制的问题,并提出两阶段检测管道和两个基准来解决跨模态的安全性问题。
This work addresses the cold-start problem in biomedical knowledge graphs, where out-of-graph molecules—provided only as SMILES strings—cannot be linked or reasoned over. To tackle this, the authors propose MolBioKG, a two-tier system that first constructs a static index of 2.74 million molecules based on scaffolds, fragments, functional groups, and molecular fingerprints, enabling multi-resolution structural anchoring and static multi-anchor retrieval via Reciprocal Rank Fusion. Building upon this, an adaptive, interpretable knowledge graph traversal is performed using a tool-augmented large language model (Adapt-KG). Requiring no task-specific training, the approach significantly improves multi-hop reasoning performance, raising Hits@10 from 0.585 to 0.876, and boosts target recall for out-of-graph molecules from 0.145 to 0.269, while ensuring predictions are grounded in structural anchors and traceable evidence.
本文提出了一种新的量化噪声理论,并基于此开发了KBBQ方法,该方法在FP4量化中通过参数化变换接近上限的程度来提高性能。
We present EmoMed - a multimodal medical consultation agent that adapts its responses based on users'emotional states while maintaining clinical accuracy. The system processes text and medical images, detects affect indicators (anxiety, confusion, urgency) from user input, and adjusts response tone, structure, and detail level accordingly. To ensure factual reliability, the agent grounds clinical information through a dual retrieval mechanism: web-based fact-checking and an API-connected, continuously updated medical knowledge base. We evaluate our approach across seven state-of-the-art language models (GPT-4/5, Qwen3, Llama 4, Gemini 2.5, Grok4, Claude3) using comprehensive metrics including LLM-as-judge assessments, MedQA style accuracy tests, BERT Score, safety/helpfulness ratings, and multimodal medical benchmarks. The results demonstrate that emotionally adaptive responses consistently outperform neutral baseline across evaluation dimensions, without compromising clinical accuracy. A controlled user study validated these findings, with participants reporting improved perceived empathy and communication clarity, while maintaining trust in factual accuracy. Source code: https://github.com/NasonovIvan/EmoMed-Agent
本文提出LoRA-TSD,通过在低秩矩阵流形的切空间中采用谱范数最速下降法优化LoRA,解决了独立训练因素忽略几何结构的问题。
研究通过因果神经元级分析,揭示视觉-语言模型中安全机制的问题,并提出两阶段检测管道和两个基准来解决跨模态的安全性问题。
This work addresses the cold-start problem in biomedical knowledge graphs, where out-of-graph molecules—provided only as SMILES strings—cannot be linked or reasoned over. To tackle this, the authors propose MolBioKG, a two-tier system that first constructs a static index of 2.74 million molecules based on scaffolds, fragments, functional groups, and molecular fingerprints, enabling multi-resolution structural anchoring and static multi-anchor retrieval via Reciprocal Rank Fusion. Building upon this, an adaptive, interpretable knowledge graph traversal is performed using a tool-augmented large language model (Adapt-KG). Requiring no task-specific training, the approach significantly improves multi-hop reasoning performance, raising Hits@10 from 0.585 to 0.876, and boosts target recall for out-of-graph molecules from 0.145 to 0.269, while ensuring predictions are grounded in structural anchors and traceable evidence.