AI 基础设施
AI 芯片、训练框架、推理优化、云服务等底层基础设施发展动态。
Using NVFP4 Low-Precision Model Training for Higher Throughput Without Losing Accuracy
As the sizes of AI models and datasets continue to increase, relying only on higher-precision BF16 training is no longer...
Making Softmax More Efficient with NVIDIA Blackwell Ultra
LLM context lengths are exploding, and architectures are moving toward complex attention schemes like Multi-Head Latent ...
利亚德：公司已与国内外诸多知名机器人厂家建立合作
36氪获悉，利亚德在机构调研时表示，截止目前，公司已与国内外诸多知名机器人厂家建立合作，合作内容涵盖硬件销售、动作训练、以及动捕训练中心建设等多个维度。
The public opposition to AI infrastructure is heating up
Public backlash over the data center boom is leading to a variety of draconian policies — including bans on new construc...
Nvidia has another record quarter amid record capex spends
"The demand for tokens in the world has gone completely exponential," Nvidia CEO Jensen Huang said about the company's e...
机构：2025年第四季DRAM产业营收为535.8亿美元，较上季度增加29.4%
36氪获悉，根据TrendForce集邦咨询最新调查显示，由于AI应用由LLM模型训练延伸至推理，推动CSPs业者的数据中心建置重心由AI Server延伸至General Server，进一步推动存储器采购重心由HBM3e、LPDDR5X...
A股三大指数收盘涨跌不一，算力股集体爆发
36氪获悉，A股三大指数收盘涨跌不一，沪指跌0.01%，深成指涨0.19%，创业板指跌0.29%；算力股集体爆发，CPO、PCB方向领涨，沪电股份、深南电路涨停，胜宏科技涨超7%；半导体走强，大族激光涨停，寒武纪涨超7%；动力电池、光伏走弱...
Towards single-shot coherent imaging via overlap-free ptychography
arXiv:2602.21361v1 Announce Type: cross Abstract: Ptychographic imaging at synchrotron and XFEL sources requires dense o...
fEDM+: A Risk-Based Fuzzy Ethical Decision Making Framework with Principle-Level Explainability and Pluralistic Validation
arXiv:2602.21746v1 Announce Type: new Abstract: In a previous work, we introduced the fuzzy Ethical Decision-Making fram...
只要1100美元tokens，一周重写 Next.js！
编辑｜冷猫今天，Web 开发社区爆发了一条令人咋舌的技术新闻。Cloudflare 的一名工程师在一周之内，借助 AI 模型从头重建了 Next.js 。该公司的首席技术官 Dane Knecht 发推庆祝这一史诗级的成就，称之为「Next...
ICLR 2026 | 把视频扩散模型压到4bit，还能接近满血效果？ QVGen让「超低比特视频生成量化」真正可用
视频生成扩散模型越做越大：2B、5B、14B…… 效果提升很快，但训练与推理的成本也随之飙升。社区一直希望用量化把模型 “压小”，把显存和算力成本打下来，真正落到更多卡、更便宜的机器、更多...
RPU -- A Reasoning Processing Unit
arXiv:2602.18568v2 Announce Type: replace-cross Abstract: Large language model (LLM) inference performance is increasing...
Characterizing State Space Model and Hybrid Language Model Performance with Long Context
arXiv:2507.12442v3 Announce Type: replace-cross Abstract: Emerging applications such as AR are driving demands for machi...
Peering into the Unknown: Active View Selection with Neural Uncertainty Maps for 3D Reconstruction
arXiv:2506.14856v2 Announce Type: replace-cross Abstract: Some perspectives naturally provide more information than othe...
An Efficient LiDAR-Camera Fusion Network for Multi-Class 3D Dynamic Object Detection and Trajectory Prediction
arXiv:2504.13647v2 Announce Type: replace-cross Abstract: Service mobile robots are often required to avoid dynamic obje...
Semantic Parallelism: Redefining Efficient MoE Inference via Model-Data Co-Scheduling
arXiv:2503.04398v4 Announce Type: replace-cross Abstract: Prevailing LLM serving engines employ expert parallelism (EP) ...
RMIT-ADM+S at the MMU-RAG NeurIPS 2025 Competition
arXiv:2602.20735v1 Announce Type: cross Abstract: This paper presents the award-winning RMIT-ADM+S system for the Text-t...
Onboard-Targeted Segmentation of Straylight in Space Camera Sensors
arXiv:2602.20709v1 Announce Type: cross Abstract: This study details an artificial intelligence (AI)-based methodology f...
What Drives Students' Use of AI Chatbots? Technology Acceptance in Conversational AI
arXiv:2602.20547v1 Announce Type: cross Abstract: Conversational AI tools have been rapidly adopted by students and are ...
Elimination-compensation pruning for fully-connected neural networks
arXiv:2602.20467v1 Announce Type: cross Abstract: The unmatched ability of Deep Neural Networks in capturing complex pat...
What Matters for Simulation to Online Reinforcement Learning on Real Robots
arXiv:2602.20220v1 Announce Type: cross Abstract: We investigate what specific design choices enable successful online r...
KnapSpec: Self-Speculative Decoding via Adaptive Layer Selection as a Knapsack Problem
arXiv:2602.20217v1 Announce Type: cross Abstract: Self-speculative decoding (SSD) accelerates LLM inference by skipping ...
Analyzing Latency Hiding and Parallelism in an MLIR-based AI Kernel Compiler
arXiv:2602.20204v1 Announce Type: cross Abstract: AI kernel compilation for edge devices depends on the compiler's abili...
Enhancing Heat Sink Efficiency in MOSFETs using Physics Informed Neural Networks: A Systematic Study on Coolant Velocity Estimation
arXiv:2602.20177v1 Announce Type: cross Abstract: In this work, we present a methodology using Physics Informed Neural N...
Meta strikes up to $100B AMD chip deal as it chases ‘personal superintelligence’
Meta is buying billions of dollars in AMD AI chips in a multiyear deal tied to a 160 million-share warrant, deepening it...
Nvidia challenger AI chip startup MatX raised $500M
The startup was founded by former Google TPU engineers in 2023.