慣性聚合 高效追蹤和閱讀你感興趣的部落格、新聞、科技資訊
閱讀原文 在慣性聚合中打開

推薦訂閱源

The GitHub Blog
The GitHub Blog
T
ThreatConnect
C
Check Point Blog
T
The Exploit Database - CXSecurity.com
U
Unit 42
云风的 BLOG
云风的 BLOG
freeCodeCamp Programming Tutorials: Python, JavaScript, Git & More
T
Tenable Blog
博客园 - 叶小钗
D
Docker
T
Threatpost
WordPress大学
WordPress大学
腾讯CDC
I
Intezer
T
Tailwind CSS Blog
Engineering at Meta
Engineering at Meta
D
Darknet – Hacking Tools, Hacker News & Cyber Security
Hugging Face - Blog
Hugging Face - Blog
cs.CV updates on arXiv.org
cs.CV updates on arXiv.org
The Register - Security
The Register - Security
Stack Overflow Blog
Stack Overflow Blog
PCI Perspectives
PCI Perspectives
S
Security Archives - TechRepublic
Simon Willison's Weblog
Simon Willison's Weblog
A
Arctic Wolf
MongoDB | Blog
MongoDB | Blog
小众软件
小众软件
Hacker News: Ask HN
Hacker News: Ask HN
O
OpenAI News
博客园 - 【当耐特】
L
LINUX DO - 最新话题
C
Comments on: Blog
S
Securelist
月光博客
月光博客
S
Secure Thoughts
Security Latest
Security Latest
MyScale Blog
MyScale Blog
NISL@THU
NISL@THU
F
Full Disclosure
M
Microsoft Research Blog - Microsoft Research
T
True Tiger Recordings
SecWiki News
SecWiki News
aimingoo的专栏
aimingoo的专栏
Threat Intelligence Blog | Flashpoint
Threat Intelligence Blog | Flashpoint
L
LINUX DO - 热门话题
cs.AI updates on arXiv.org
cs.AI updates on arXiv.org
AWS News Blog
AWS News Blog
Hacker News - Newest:
Hacker News - Newest: "LLM"
L
Lohrmann on Cybersecurity
H
Help Net Security

cs.LG updates on arXiv.org

Personalized Generative Models for Contextual Debiasing From Privacy to Generalization: Linear Max-Information Bounds for DP-SGD When Does Deep RL Beat Calibrated Baselines? A Benchmark Study on Adaptive Resource Control Amortized Factor Inference Networks for Posterior Inference Classification and detection of multiple UAVs using rational Gaussian wavelet neural networks Planning Neural Dynamics with Lie Group Embedding through Supervised Projective Manifold Learning Modeling Dynamic Mixtures of Time-Delay Systems from Streaming Time Series AirCast-SR: A Foundation Model for Kilometer-Scale Atmospheric Super-Resolution via Latent Consistency Diffusion Neural Bayesian Sequential Routing GAC: Noise-Aware Adaptive Mixing for Hybrid SFT-RL Post-Training Provably Communication-Efficient and Privacy-Preserving Federated Graph Neural Networks Function-Valued Causal Influence in Nonlinear Time Series The Bridge-Garden Dilemma in LLM Distillation: Why Mixing Hard and Soft Labels Works Balancing Plasticity and Stability with Fast and Slow Successor Features InfoQuant: Shaping Activation Distributions for Low-Bit LLM Quantization FM-fMRI: Event Conditioned Flow Matching for Rest-to-Task fMRI Time-Series Synthesis TrackRef3D: Multi-View Consistent Track-then-Label for Open-World Referring Segmentation in 3D Gaussian Splatting TSFMAudit: Data Contamination Auditing in Forecasting Time Series Foundation Models On the Push-Based Asynchronous Federated Learning: A Bias-Correction Aggregation Approach CSV-ViT: A Vision Transformer with the Variable-sized Cortical Supervertices for Detection of Alzheimer's Disease Pathologies Max-Window Scale Estimation for Near-Lossless HiF8 W8A8 Quantization-Aware Training Online Learning on Hidden-Convex Losses via Algorithmic Equivalence: Optimal Regret, Geometric Barrier, and Bandit Feedback Curriculum Learning for Safety Alignment A PAC-Bayesian View of Generalisation for Physics-Informed Machine Learning Dynamic Link Prediction with Temporally Enhanced Signed Graph Neural Networks GEM: Geometric Entropy Mixing for Optimal LLM Data Curation Reparametrizing Shampoo and SOAP for Subspace Basis Updates and BFloat16 Storage Unified Neural Scaling Laws Semigroup Consistency as a Diagnostic for Learned Physics Simulators QAM-W: Joint 2D Codebook Quantization for LLM Weights via Hadamard Rotation and Activation-Aware Scaling HRVConformer: Neonatal Hypoxic-Ischemic Encephalopathy Classification from the Heart Rate signals Scaling World-Model Reinforcement Learning Through Diffusion Policy Optimization A Hybrid Vision-Language Architecture for Automated Defect Reasoning and Report Generation in Industrial Inspection Quantized Keys Steal Attention: Bias Correction for KV-Cache Compression in Video Diffusion BioFact-MoE: Biologically Factorized Mixture of Experts for Vision-Language Prognostic Modeling in Hepatocellular Carcinoma Bridging Classification and Reconstruction: Cooperative Time Series Anomaly Detection SilIF: Silhouette-Augmented Isolation Forest for Unsupervised Transaction Fraud Detection Co-folding model guided by structural proteomics Energy-Gated Attention and Wavelet Positional Encoding: Complementary Inductive Biases for Transformer Attention Stateful Inference for Low-Latency Multi-Agent Tool Calling Two-Parameter Flows for Learning Population Dynamics of Physical Systems On the Role of Inductive Bias in Time-Series Pretraining: A Case Study in Learning Generalizable Representations for Clinical Time Series A Fast and Generic Energy-Shifting Transformer for Hybrid Monte Carlo Radiotherapy Calculation ARBITER: Reasoning Trajectory Basins and Majority Vote Failures in Test-Time Sampling When Rule Violations Are Rare: Chimera Training for Logical Anomaly Detection The Constraint Tax: Measuring Validity-Correctness Tradeoffs in Structured Outputs for Small Language Models When Correct Demonstrations Hurt: Rethinking the Role of Exemplars in In-Context Learning Benchmarking Convolutional, Transformer, Hybrid, and Vision Language Models for Multi Disease Retinal Screening Rotation-Invariant Spherical Watermarking via Third-Order SO(3) Representation Coupling MULTISEISMO: A Multimodal Seismic Dataset and Model for Cross-Modal Seismic Understanding
透支樣本變異推理來對齊少步生成模型
Jaewoo Lee, · 2026-05-27 · via cs.LG updates on arXiv.org

查看PDF HTML (實驗性)

摘要:對齊多步生成模型是挑戰性的,因為現有的對齊框架通常依賴於限制性假設:一個易處理的似然性、一個特定的常微分方程/隨機微分方程求解器,或是一個特定的模型家族。我們介紹 FAV,基於樣本的變分推理進行多步生成模型對齊,是一個通用對齊框架,只需要對生成器和參考分佈進行樣本訪問。我們將對齊視為從一個以參考分佈為基準的獎勵偏斜分佈中進行抽樣。我們利用 Stein 變分梯度下降作為基於樣本的變分推理方案,並通過固定點回歸將其粒子更新分攤到生成器參數中。我們在兩個領域上評估 FAV:機器人操作和圖像生成器對齊。在機器人操作生成策略對齊方面,FAV 在 56 個離線和 30 個離線到線上 RL 任務上優於當前的策略提取基線。對於圖像生成器對齊,FAV 微調多樣化的多步主幹,包括 GAN、漂移模型、一致性模型和流圖,規模從 ImageNet-$256$ 到 1024$^2$ 文本到圖像合成。程式碼可在此 https URL
評論: 審核中
主題: 機器學習 (cs.LG); 人工智慧 (cs.AI)
引用格式: arXiv:2605.26552 [cs.LG]
  (或 arXiv:2605.26552v1 [cs.LG]) for this version)
  https://doi.org/10.48550/arXiv.2605.26552

arXiv發行的DOI透過DataCite(待登記)

提交通過歷史

From: Jaewoo Lee [查看郵件]
[v1] 周二,2026年5月26日 05:02:49 UTC (18,419 KB)