每日AI论文精选-2026-11-11

AI 论文精选速递

2026年11月11日 · 追踪前沿研究

1

Harness^2: Recursive Agent Harnessing for an Open World

Ruiyao Xu, Yanfei Chen 等9人

Harness² improves agent execution harnesses through recursive task-level refinement and continual harness-level learning, boosting performance across benchmarks without ground-truth verifiers.

#recursive self-improvement#LLM agents#execution harnesses

https://arxiv.org/abs/2610.14224

https://hjfy.top/arxiv/2610.14224

https://github.com/google-research/harness2 (23369)

2

Learning Heterogeneous Tactile Representation for General Dexterous Manipulation

Bing Han, Yichuan Peng 等19人

LingBot-Touch is a tactile foundation model that learns a unified, extensible representation across heterogeneous tactile sensors, yielding plug-and-play tactile tokens that consistently enhance manipulation policies such as diffusion policy and VLA models.

#tactile foundation model#unified representation#heterogeneous tactile data

https://arxiv.org/abs/2611.12806

https://hjfy.top/arxiv/2611.12806

https://github.com/Robbyant/lingbot-touch (137)

3

Flide: Accelerating Decision Inference with Exact State Forking

Bobo Li, Jiajia Song 等5人

Flide is an exact inference engine that accelerates decision-model LLM agents by decoupling logical request trees from physical execution and using ForkAttention to preserve exact branch evaluations.

#LLM agent systems#categorical reasoning#System One execution

https://arxiv.org/abs/2610.42148

https://hjfy.top/arxiv/2610.42148

https://github.com/OmniJev/Flide (344)

4

FreeToken: Efficient Edge-Native MoE Serving with Bandwidth-Adaptive Execution

Shuo Yang, Xiaoze Fan 等11人

FreeToken is an edge-native Mixture-of-Experts serving system that dynamically maps computation and model state onto heterogeneous local hardware to run large open-weight models on personal machines.

#MoE serving#expert residency#CPU-GPU execution

https://arxiv.org/abs/2608.16157

https://hjfy.top/arxiv/2608.16157

https://github.com/FlashML-org/FreeToken (16985)

5

SplatAnything: Gaussian Spacing Normalization for Feed-forward 3DGS across Arbitrary Scenes

Hao Yu, Zhenwei Wang 等10人

SplatAnything applies Gaussian Spacing Normalization to canonicalize scene-level Gaussian spacing, yielding more consistent supervision and improved feed-forward 3D Gaussian Splatting under varying scene scales and viewpoints.

#feed-forward 3DGS#3D Gaussian Splatting#Gaussian attributes

https://arxiv.org/abs/2611.08264

https://hjfy.top/arxiv/2611.08264

https://github.com/zju3dv/SplatAnything (138)

6

Efficient Memory Management for Large Language Model Serving with PagedAttention

Woosuk Kwon, Zhuohan Li 等9人

PagedAttention algorithm and vLLM system enhance the throughput of large language models by efficiently managing memory and reducing waste in the key-value cache.

#PagedAttention#key-value cache#KV cache

https://arxiv.org/abs/2309.06180

https://hjfy.top/arxiv/2309.06180

https://github.com/vllm-project/vllm (86094)

7

OpenDevin: An Open Platform for AI Software Developers as Generalist Agents

Xingyao Wang, Boxuan Li 等24人

OpenDevin is a platform for developing AI agents that interact with the world by writing code, using command lines, and browsing the web, with support for multiple agents and evaluation benchmarks.

#large language models#OpenDevin#AI agents

https://arxiv.org/abs/2407.16741

https://hjfy.top/arxiv/2407.16741

https://github.com/opendevin/opendevin (93144)

8

Very Large-Scale Multi-Agent Simulation in AgentScope

Xuchen Pan, Dawei Gao 等8人

Enhancements to the AgentScope platform improve scalability, efficiency, and ease of use for large-scale multi-agent simulations through distributed mechanisms, flexible environments, and user-friendly tools.

#actor-based distributed mechanism#multi-agent platform#centralized workflow orchestration

https://arxiv.org/abs/2407.17789

https://hjfy.top/arxiv/2407.17789

https://github.com/modelscope/agentscope (35750)

9

AgentScope 1.0: A Developer-Centric Framework for Building Agentic Applications

Dawei Gao, Zitao Li 等23人

AgentScope enhances agentic applications by providing flexible tool-based interactions, unified interfaces, and advanced infrastructure based on the ReAct paradigm, supporting efficient and safe development and deployment.

#Large Language Models#LLMs#AgentScope

https://arxiv.org/abs/2508.16279

https://hjfy.top/arxiv/2508.16279

https://github.com/agentscope-ai/agentscope (35750)

论文摘要由 AI 生成 · 仅供学习参考

生成时间: 2026-11-11 08:30

© 2026 AI 论文精选 · 共收录 9 篇