微信内可能无法直接打开本站。请点右上角 ··· → 在浏览器打开，或复制链接后用系统浏览器访问。
XMT
信流 · 上滑连读 · 来源可核
Content Depth Matters in Short-Video Recommendation: Rethinking the Attention Economy
arXiv:2608.…
- 13990v1 Announce Type: new Abstract: Driven by the attention economy, …
- These systems inherently favor shallow-content videos that are effecti…
- However, growing evidence suggests that prolonged exposure to such con…
RSS 官方收录 · 可信分层展示
Simulation-Driven Vehicular Traffic Data Augmentation: Extending Sensor Coverage Through Virtual Sensing
arXiv:2608.13993v1 Announce Type: new Abstract: Urban traffic management relies on sensor networks whose spatial coverage is limited by deployment costs and privacy regulations.…
- Machine learning models trained on such sparse data cannot generalize …
- We propose a simulation-based methodology that addresses this problem …
- Virtual sensors are selected by a graph-search heuristic that jointly …
RSS 官方收录 · 可信分层展示
Nanbeige4.2-3B on Apple Silicon: Fixing Deployment Bugs and Decreasing Looped Transformer Memory Overhead
arXiv:2608.13987v1 Announce Type: new Abstract: Nanbeige4.…
- 2-3B is a 3B-parameter agentic model built around a Looped Transformer…
- Evaluated on Apple Silicon (MPS), we identify five independent bugs wh…
- Furthermore, we show that fixing these bugs is still not sufficient fo…
RSS 官方收录 · 可信分层展示
Implementing Computational Law in Wolfram Language for the Governance of Artificial Intelligence
arXiv:2608.13958v1 Announce Type: new Abstract: How do we govern AI systems whose reasoning we cannot fully inspect?…
- Governance does not require understanding a system's reasoning.
- It requires stating what the system is obliged, permitted, and forbidd…
- I present an implementation of Reified Input/Output Logic, the formali…
RSS 官方收录 · 可信分层展示
HELIX: Model-Harness Co-evolution for Recursive Self-Improvement
arXiv:2608.…
- 13951v1 Announce Type: new Abstract: Scaling agent capability has larg…
- The harness shapes both what a model can accomplish and the trajectori…
- This coupling motivates model-harness co-evolution for recursive self-…
RSS 官方收录 · 可信分层展示
从「3D相册」到「物理底座」：CVPR 2026 开启 3DGS 的具身智能时代
秒级前馈生成、Neural ODE 物理碰撞与真实抗噪建图，3DGS 正演变为机器人模仿学习的数字孪生工厂。作者丨张 璐 编辑丨齐铖湧 在过去两年的 3DGS 研发浪潮中，我们经常看到这样的奇观：算法在论文里跑出了惊艳的 3D 渲染画面，但一拿到真实手机或机器人上，立刻卡成ppt甚至崩溃。…
- “能在实验室发论文”与“能在工业现场干活”之间，隔着一条残酷的工程鸿沟。
- 到了 CVPR 2026，这场画质内卷的狂欢终于踩下了刹车。
- 前沿团队们的视角全面转向工业落地：从秒级前馈生成、GPU 算子重构，到抗噪 SLAM 建图与 Neural ODE 物理碰撞推算，3DGS …
RSS 官方收录 · 可信分层展示
别再乱调图像塔了！浙大 IJCAI 论文揭露 VLM 非对称性真相，给 CLIP 微调「踩刹车」
A3B2自适应非对称适配器，诠释大模型微调的“克制”艺术。作者丨张 璐 编辑丨幸丽娟 在视觉语言大模型（VLM）落地少样本迁移任务时，高效参数微调（PEFT）是大家普遍采用的低成本方案。…
- 然而，业界长期存在一个惯性思维：既然是做图像分类，给图像编码器多挂几个 Adapter、多做微调，模型理应能看懂更多视觉细节。
- 但现实却演化出一个尴尬的矛盾：在面对未知的现实场景（域外数据/OOD）时，在图像编码器上微调得越激进，模型的泛化能力反而破坏得越惨烈。
- 预训练大模型原本拥有一双看懂万物的“好眼睛”，强行局部微调反而把它“改近视了”。
RSS 官方收录 · 可信分层展示
Never the Number: Structural Abstention for AI Systems Whose Answers Are Consumed as Fact
arXiv:2608.…
- 13926v1 Announce Type: new Abstract: Large language models have made n…
- Where the consumer cannot inspect the generated query, as in enterpris…
- This is a reliability problem before it is an accuracy problem.
RSS 官方收录 · 可信分层展示
When Personal Memory Has No Single Answer: Evaluating LLM Agents under Irreducible Conflict
arXiv:2608.13921v1 Announce Type: new Abstract: LLM agents increasingly maintain personal memory across sessions, but it can conflict.…
- Preferences depend on context, behavior evolves, and sources can confl…
- When a query lacks context, time, or source authority to interpret con…
- Existing benchmarks recover one answer from conflicting evidence, over…
RSS 官方收录 · 可信分层展示
Joint Optimization of Memory and Computing Frequency for Energy-Efficient DNN Inference
arXiv:2608.…
- 13863v1 Announce Type: new Abstract: Deep neural network (DNN) inferen…
- To enable energy-efficient DNN inference, most existing studies focus …
- In this paper, we consider the impact of memory frequency and computin…
RSS 官方收录 · 可信分层展示
MemoryLake on MemoryArena: A Matched Study of Agent Memory Backends
arXiv:2608.…
- 13883v1 Announce Type: new Abstract: Most agent-memory benchmarks test…
- We compare MemoryLake, a structured multi-track memory backend, with M…
- The systems share the same agent framework, requested gpt-5-mini model…
RSS 官方收录 · 可信分层展示
SDO: Subspace Deconflicting Operator for Multi-Adapter Composition
arXiv:2608.…
- 13820v1 Announce Type: new Abstract: Composing independently trained a…
- We study this interference from a parameter-space perspective and hypo…
- To address this issue, we propose \textbf{SDO}, a \textbf{S}ubspace \t…
RSS 官方收录 · 可信分层展示
From Passive Delegates to Strategic Negotiators: Reinforcing Social Reasoning in Small Language Models with SocialRL
arXiv:2608.…
- 13787v1 Announce Type: new Abstract: AI agents increasingly act on the…
- These principal-driven tasks routinely place the agent across from a c…
- Yet the dispositions that make an assistant pleasant can make it a poo…
RSS 官方收录 · 可信分层展示
FLARE MCMC: Fidelity-based Layer-Adaptive REcursive proposals for MCMC
arXiv:2608.…
- 13774v1 Announce Type: new Abstract: Markov chain Monte Carlo (MCMC) r…
- However, it can have a slow mixing rate, requiring the generation of m…
- FLARE MCMC is a multi-fidelity layered MCMC method that exploits lower…
RSS 官方收录 · 可信分层展示
Simulation-Aware In-Context Policy Improvement for LLM-Aided Analog Layout Refinement
arXiv:2608.13767v1 Announce Type: new Abstract: Analog IC layout design remains a labor-intensive iterative process dominated by simulation-driven refinement.…
- Although end-to-end layout generators accelerate initial placement and…
- While Bayesian Optimization (BO) is widely adopted for parameter tunin…
- Recently, Large Language Models (LLMs) have demonstrated potential in …
RSS 官方收录 · 可信分层展示
Explanation Multiplicity: Circuit-Level Interpretability Evidence Does Not Survive Defensible Analytic Variation
arXiv:2608.…
- 13754v1 Announce Type: new Abstract: The EU AI Act requires providers …
- Mechanistic interpretability is the obvious source of such evidence, a…
- We ask whether that evidence survives the condition under which it wou…
RSS 官方收录 · 可信分层展示
中国电信领投，觅蜂科技再获数亿元融资，聚焦物理AI数据服务平台
雷峰网获悉，近日，一站式物理AI数据服务平台觅蜂科技完成新一轮数亿元融资。本轮由中国电信领投，张江集团跟投，红杉中国、元启创新等老股东超额追加。资金将用于扩大MEgo系列无本体采集产品量产、提升数据采集—治理—评测全栈能力，并加速千万小时级物理交互数据产能落地。…
- 成立约半年，觅蜂科技已连续完成多轮融资，投资方涵盖国资平台、头部基金及产业方。
- 进入真实场景采集，定位“懂模型与场景”的数据平台觅蜂科技认为，物理AI要实现高阶推理和物理规律掌握，需依赖大规模、高质量真实交互数据。
- 相比大语言模型，物理AI数据有效信息密度更低、噪声更高，需求更大。
RSS 官方收录 · 可信分层展示