长上下文LLM推理提速新方案，块稀疏注意力优化预填充阶段，值得关注。
arXiv:2608.19758v1 Announce Type: new Abstract: Long-context modeling is a pivotal capability for Large Language Models, yet the quadratic complexity …
LLM不确定性新思路：无需额外训练，仅靠注意力路径的脆弱程度就能暴露模型“心虚”时刻，极具实用潜力。
arXiv:2608.11138v1 Announce Type: new Abstract: We propose that a model's uncertainty about a token is reflected not only in the breadth of its output…
无需训练即可加速注意力计算，二进制主成分构造哈希，大幅降低复杂度，值得关注。
arXiv:2608.04405v1 Announce Type: cross Abstract: Long-context large language models (LLMs) are increasingly deployed in real-world applications, yet …
多模态大模型如何在推理中平衡效率与效果？这篇ACM MM论文提出无训练注意力引导切换，值得关注。
arXiv:2608.03450v1 Announce Type: cross Abstract: Reasoning in Multimodal Large Language Models (MLLMs) requires both fine-grained visual perception a…
稀疏注意力中因果信息流被重新定义，Stem提出新视角，为长上下文模型效率与准确性兼得提供关键思路。
arXiv:2603.06274v2 Announce Type: replace-cross Abstract: The quadratic computational complexity of self-attention remains a fundamental bottleneck fo…
提出边缘感知热红外无人机群跟踪方法，融合图谱卷积与注意力机制，突破多目标遮挡和红外模糊难题。
arXiv:2607.12544v1 Announce Type: new Abstract: Thermal infrared (TIR) imaging is essential for UAV swarm operations in visually degraded environments…
块稀疏注意力硬件友好却未被主流LLM采用？本文剖析原因并提出新型COBS注意力机制。
arXiv:2607.09052v1 Announce Type: new Abstract: Block sparse attention is a hardware friendly way to alleviate the key-value (KV) cache read bottlenec…
多模态大模型定位幻觉问题新解法，无需训练即可评估置信度，基于多令牌局部注意力机制。
arXiv:2607.05978v1 Announce Type: cross Abstract: Multimodal large language models can emit localized predictions, bounding boxes for objects and temp…
提出条件性记忆访问机制，引导长上下文LLM更智能地选择注意力范围，显著提升效率与性能，ICML 2026最新成果。
arXiv:2603.17484v2 Announce Type: replace-cross Abstract: Language models struggle to generalize beyond pretraining context lengths, limiting long-hor…
快手新注意力机制直击长上下文痛点，线性复杂度替代二次方，为下一代LLM推理提速。
arXiv:2604.24432v2 Announce Type: replace-cross Abstract: Long-context ability, has become one of the most important iteration direction of next-gener…
熵引导视觉token剪枝，修正注意力机制，让多模态大模型推理更快更省。
arXiv:2606.31982v1 Announce Type: new Abstract: Multimodal Large Language Models (MLLMs) incur prohibitive inference costs due to long visual token se…
多模态注意力对齐新方法，让GUI定位更精准，值得关注
arXiv:2511.00810v4 Announce Type: replace-cross Abstract: Graphical user interface (GUI) grounding is a key capability for computer-use agents, mappin…
这篇论文提出MatchAttention注意力机制，将显式匹配约束嵌入到Transformer中，显著提升立体匹配的准确性和效率
arXiv:2510.14260v3 Announce Type: replace Abstract: Standard attention mechanisms are not well suited to stereo matching. Global attention scales quad…
全新多尺度注意力UNet3+，引入原型对比损失，精准分割冠状动脉DSA图像。
arXiv:2504.05184v4 Announce Type: replace-cross Abstract: Accurate segmentation of coronary Digital Subtraction Angiography (DSA) images is essential …
深度拆解KV缓存与MQA、GQA、MLA原理，揭示大模型推理加速的关键技术。
LLMs generate text one token at a time. That sounds simple. But without KV Cache, every new token would repeat a lot of old work. That is why inferenc…
语音转文本模型中的交叉注意力并非完整解释，揭示其局限与真正作用机制
arXiv:2509.18010v2 Announce Type: replace-cross Abstract: Cross-attention is a core mechanism in encoder-decoder architectures, widespread in many fie…
用注意力机制检测野外恶意Agent技能，为AI安全提供新思路。
arXiv:2606.23416v1 Announce Type: cross Abstract: LLM agents increasingly load skills, file-based packages of natural-language instructions written by…
GPT-5等大模型在简单的人类注意力测试中败下阵来，Stroop效应暴露出AI注意力架构的致命短板。
Article URL: https://scitechdaily.com/even-gpt-5-failed-this-human-attention-test/ Comments URL: https://news.ycombinator.com/item?id=48626330 Points:…
从同步理论窥见注意力的本质：完美同步无法计算，FSN通过“受阻同步”实现结构化差异，为Transformer架构提供全新数学视角。
arXiv:2606.18694v1 Announce Type: new Abstract: A network of oscillators that synchronizes perfectly computes nothing further, so an attention archite…