看学术

AI 安全方向的论文、会议、比赛、教程与资料，按内容形态筛选，支持搜索。

共 1,382 条 · 更新于 15:39

10月20日周二

10月20日星期二

12:58
arXiv· 12:58
Cross-Fitting for Neural Posterior Estimation
12:58
arXiv· 12:58
Parallel Language Models via Variational Decoding
12:58
arXiv· 12:58
Hierarchical Token Merging with Recursive Autoencoders for Unsupervised Grammar Induction
12:58
arXiv· 12:58
CAVA: Complementary-Label-Aware Data Valuation
12:58
arXiv· 12:58
Polar Switch: Fast Muon Orthogonalization with Polynomial Circuits
12:58
arXiv· 12:58
HoloCosm: Beyond Single-Metric Benchmark Compression Across Multiple Metrics
12:58
arXiv· 12:58
Copying Under Repetition: From Prefix Matching to Position Hashing
12:58
arXiv· 12:58
Gradient descent reaches arbitrary accuracy in a fixed number of large steps on multiclass logistic regression
12:58
arXiv· 12:58
EpiLens: Lightweight RAG for Epidemiological Knowledge Acquisition
12:58
arXiv· 12:58
Trajectory Replay: Continuous-Time Mixture Transport for Lifelong Generative Models
12:58
arXiv· 12:58
Simulation Priors for Rapid Adaptation in Model-Based RL for Autonomous Racing
12:58
arXiv· 12:58
recAMPLIFY: Large-Scale Pretraining of Recursive Protein Language Models
12:58
arXiv· 12:58
Topological Sheaves of Irreducibility: A Theoretical Framework for Multi-Agent Task Obstructions
12:58
arXiv· 12:58
TokenGate: Adapting LLMs for Fast and Memory-Efficient Prefill with Learned Token Pruning
12:58
arXiv· 12:58
HarnessPro: Recursive Harness Self-Improvement with Pareto Optimization for Performance-Cost Trade-offs
12:58
arXiv· 12:58
Cooperative Multi-Agent Reinforcement Learning for Market Making on Prediction-Market Order Books
12:58
arXiv· 12:58
Monsoon: Scalable Spontaneous Speech Collection for Low-Resource ASR in 50 Languages
12:58
arXiv· 12:58
Training with ML-labeled Data via Stochastic First- and Zeroth-Order Optimization
12:58
arXiv· 12:58推荐指数3232
Game of Phones：用博弈论与 LLM 在“Bot Wars”中反击电话诈骗
研究者提出一个博弈论框架，让 LLM 无需微调即可通过上下文经验学习反诈骗策略，将诈骗者与反诈机器人互动建模为双人博弈，并以纳什均衡作为评估学习行为的正式基准。在 300 段模拟对话和三种 LLM 架构上，自适应策略选择相比非自适应基线将纳什均衡对齐度提升 28.9-79.7%。代码与含逐轮策略标签和效用分数的 300 段对话标注数据集已开源，成果获 ECML PKDD 2026 接收。
安全论文安全研究 #论文/研究 #C02 AI对抗防御 #C11 AI安全事件管理检测与护栏垂直领域 Agent 红队与自动化测试基准测试集
12:58
arXiv· 12:58
On Data Scaling in Tabular ML: More Data, Better Generalization?
12:58
arXiv· 12:58
RESIN: Multi-Chain Reasoning with Entity-Structured Indexing
12:58
arXiv· 12:58
Two hidden layer suffice: the caterpillar net
12:58
arXiv· 12:58
Teacher Supervision, Fixed Representations and Calibration: Exact Risk Calculations for Candidate-Scoring Decision Models
12:58
arXiv· 12:58
Safety Mid-Training: Internalizing LLM Safety as a Foundational Capability
12:58
arXiv· 12:58
Learning to Refer: Inference-Time Collaboration in Decentralized Federated Learning
12:58
arXiv· 12:58
ASTOR: Benchmarking and Improving Autoformulation for Soft-Constraint Optimization with a Formal Grammar
12:58
arXiv· 12:58
Scaling Parameter-Efficient Fine-Tuning for Brain Encoding
12:58
arXiv· 12:58
AfriBED: a Speech Recognition Bias Evaluation Dataset for Low Resource African Languages
12:58
arXiv· 12:58
Next Best Action: Causal Constrained Offline RL for Business Action Targeting
12:58
arXiv· 12:58
SpecMRL: Redesign Matryoshka Representation Learning within Parametric Spectrum
12:58
arXiv· 12:58
MemT-1.0: Towards Memory-Native Evolution
12:58
arXiv· 12:58
RefCraft: Benchmarking Visual-Reference-Conditioned Agentic Generation through Web Development
12:58
arXiv· 12:58
ProteinGeoBench: Frontier Structural-Biology Agents Are Fluent at Lookup but Lost in Geometry
12:58
arXiv· 12:58
KEEP-Bench: Evaluating Evidence Preservation in Context Compression
12:58
arXiv· 12:58
Supervised Learning with Quantum Hamiltonian Embeddings
12:58
arXiv· 12:58
Agent World Model in the Wild: Computer-Use Agent Environment Synthesis
12:58
arXiv· 12:58
Competing Learning in Hybrid Architectures
12:58
arXiv· 12:58
Kronecker Delta Rule for Long Context Linear Attention
12:58
arXiv· 12:58
LatComm: Reusable Context Compression for Latent Communication between Large Language Models
12:58
arXiv· 12:58
Newton-Based Methods for Bilevel Optimization Beyond Lower-Level Strong Convexity