Tag
该论文研究了在无限混合可适当损失（mixable proper losses）类别下的 omniprediction 问题，证明了对所有有界适当损失无法实现次线性遗憾，但给出了对所有混合可适当损失的最优 O~(T^{1/3}(log|F|)^{2/3}) 遗憾界，并提出了一种同时兼顾有限损失类别的 best-of-all-worlds 算法。
该论文研究具有已知有向反馈图的对抗性多臂老虎机的一阶遗憾保证，通过固定点采样规则和两层聚合方法，将强可观察图的期望遗憾改进至接近最优的 Õ(√(αL*) + α)，弥合了最坏情况与一阶保证之间的图依赖差距。
The paper proposes an online reinforcement learning framework that trains UI navigation agents on live iOS apps, decoupling them into a planning agent and a UI grounding model with a novel neighbor-conditioned value estimation method (GiGPPO) that improves critic stability and reduces inference costs. Their 8B RL agents outperform SFT baselines and match Gemini-3.1-Pro on real iOS apps while showing strong cross-platform generalization to Android.
The paper proposes WaferHDC, an online open-world wafer defect discovery framework based on hyperdimensional computing that dynamically detects and groups recurring defect patterns without labels or retraining, outperforming the strongest online baseline by 13.3 p.p. in adjusted Rand index on the WM-811K dataset.
This paper studies predictive flow matching under distribution drift, proposing DyProd-FM which refits on dyadic historical prefixes and uses a sleeping Prod master to achieve minimax-optimal dynamic regret under both relative velocity risk variation and target-law variation, without requiring convexity or curvature assumptions.
LEGO-SCD is an online scene change detection method that learns scene-specific statistical models from automatic edits to a reference 3D Gaussian Splatting scene without manual annotations, enabling closed-form Bayesian evidence accumulation with object-guided mask refinement; it achieves 67.64% mIoU on PASLCD and 75.82% on CCS3D at 17.2 FPS.
该论文提出一种基于约束上下文老虎机的个性化 LLM 路由框架，通过 primal-dual GLM-UCB 算法在线学习用户偏好，在最小化推理成本的同时以高概率保证响应质量满意度。
该论文首次研究了可复制的对抗性多臂老虎机（replicable dueling bandits）问题，设计了在共享内部随机性下保证每次运行选择一致比较对的算法，并在多种随机传递性假设下证明了次线性后悔界，实验在合成数据和语言模型偏好上验证了可复制性。
A paper proposing Residual Quantization (RQ) as a representation layer for contextual bandits, enabling nonlinear reward modeling with strictly bounded O(1) memory via gradient-boosted pseudo-targets and online depth adaptation through a shadow mechanism, beating non-RQ counterparts on 11 of 13 datasets while matching XGBoost and neural baselines at up to 1000x less memory.
LVMT (Long-term Video Mask Transformer) introduces a GRU-based temporal propagation module and a Truncated Query Propagation (TQP) training strategy to handle long-term occlusions in online video segmentation, setting new state-of-the-art results across six benchmarks while running 10x faster than prior methods.
This paper proposes a method for improving frozen forecasters by combining static and online correctors to control downside risk, demonstrating gains up to 11.5% in benchmarks and reduced errors in electricity load forecasting.
This paper introduces randomized-pass replay (RPR) to bound rehearsal gaps in online continual learning, showing improved accuracy over independent class-balanced retrieval in experience replay methods like ER-ACE.
This paper introduces the Lightweight Ranking Heads framework to accelerate multi-task experimentation in production recommender systems by enabling dynamic task injection without retraining backbone models, reducing iteration cycles from weeks to days at YouTube scale.
A promotional post for TensorTonic, an online platform for learning machine learning through coding practice, starting with an explanation of the RBF kernel.
This paper improves the optimal expected mistake bound for randomized proper online learning, showing it is O(L(H) log T), which is optimal up to a universal constant for worst-case classes.
The paper introduces RAVEL, a retrieval-aware online reinforcement learning framework for interactive person re-identification that optimizes question selection based on retrieval feedback to improve performance across multiple interaction rounds.
Keet is an educational platform launched on Product Hunt, offering interactive video courses on any topic with AI integration.
This paper introduces HACK GPs, a method that uses online learning with expert advice for kernel selection in Gaussian Processes, enhancing robustness in sequential decision-making tasks such as Bayesian optimization and active learning.
The paper introduces the Free Inference dimension, a combinatorial complexity measure for zero-collision navigation in meta-reinforcement learning under hypothesis mixtures, proving its relationship with VC-dimension and generalization bounds.
The paper introduces MASA, a method that uses frozen multimodal large language models to break the self-referential loop in wild test-time adaptation by providing structured semantic descriptions for more reliable adaptation.