NIRC is based on the National Key Laboratory of Network and Switching Technology at Beijing University of Posts and Telecommunications. It has built a technology matrix across four AI domains—intelligent cloud networking, natural language processing, computer vision, and machine learning systems—dedicated to solving real‑world problems, creating top‑tier systems, publishing high‑impact papers, and contributing significantly to the rapid advancement of China's network technology.
This article surveys current network load‑balancing approaches—including CONGA, Hula, DRILL, Hermes, MP‑RDMA, ConWeave, Proteus, and CAVER—detailing their granularity, information exchange, signaling methods, and the performance gains they achieve in modern data‑center environments.
This article walks through exporting a PyTorch image‑classification model to ONNX, identifying a hidden layer node, adding it as an output, and using ONNX Runtime to retrieve both the final prediction and the intermediate feature tensor.
This article walks through using DeepQueueNet—a deep‑learning‑enhanced network performance estimator—to set up a device model, train the PyTorch version, configure a fattree16 topology, and run multi‑GPU simulations that deliver minute‑level, packet‑accurate results in as little as 1 minute 27 seconds.
The article recounts the authors' participation in EuroSys'25 in Rotterdam, detailing the conference schedule, their presentation of the Atlas network verification paper, technical insights into distributed verification, interactions with peers, and memorable social and cultural experiences during the five‑day event.
The article analyzes the Model Context Protocol (MCP) as an emerging open standard that unifies how applications provide context to large language models, reviews its rapid ecosystem growth, highlights security and performance challenges, and discusses future directions such as vertical small‑model opportunities and broader protocol integrations.
The article analyzes the anti‑scaling phenomenon in video large‑language models, identifies a “temporal hacking” shortcut where models focus on a few key frames, formalizes it via reward‑hacking theory, introduces the Temporal Perplexity (TPL) metric, and proposes an Unhackable Temporal Rewarding (UTR) framework to mitigate the issue.
This guide introduces Hugging Face's TRL library, explains how to install it alongside Transformers, and walks through modifying LLaVA's trainer, dataset, and data collator to apply the DPO reinforcement‑learning algorithm for multimodal model fine‑tuning.
The paper introduces MHA2MLA, a data‑efficient fine‑tuning framework that converts pre‑trained multi‑head attention LLMs to DeepSeek’s Multi‑Head Latent Attention architecture, achieving up to 92% KV‑cache compression with less than 0.5% performance loss on long‑context tasks.
This guide walks through setting up Meta XR SDK in Unity, using Building Blocks to add camera rigs, hand tracking and passthrough, binding interaction events, accessing hand‑tracking data via OVRSkeleton/OVRHand, and integrating ONNX machine‑learning models for XR experiments.
The article reviews the paper ‘Towards Monosemanticity: Decomposing Language Models With Dictionary Learning’, showing how Anthropic’s sparse autoencoders extract interpretable, monosemantic concepts from transformer layers, enable controlled generation, and reveal trade‑offs such as data‑intensive training and potential performance impacts.
