SSL: Sweet Spot Learning for Differentiated Guidance in Agentic Optimization

Reinforcement learning with verifiable rewards has emerged as a powerful paradigm for training intelligent agents. However, existing methods typically employ binary rewards that fail to capture quality differences among trajectories achieving identical outcomes, thereby overlooking potential diversity within the solution space. Inspired by the ``sweet spot'' concept in tennis-the racket's core region that produces optimal hitting effects, we introduce \textbf{S}weet \textbf{S}pot \textbf{L}earning (\textbf{SSL}), a novel framework that provides differentiated guidance for agent optimization. SSL follows a simple yet effective principle: progressively amplified, tiered rewards guide policies toward the sweet-spot region of the solution space. This principle naturally adapts across diverse tasks: visual perception tasks leverage distance-tiered modeling to reward proximity, while complex reasoning tasks reward incremental progress toward promising solutions. We theoretically demonstrate that SSL preserves optimal solution ordering and enhances the gradient signal-to-noise ratio, thereby fostering more directed optimization. Extensive experiments across GUI perception, short/long-term planning, and complex reasoning tasks show consistent improvements over strong baselines on 12 benchmarks, achieving up to 2.5X sample efficiency gains and effective cross-task transferability. Our work establishes SSL as a general principle for training capable and robust agents.

CPiRi: Channel Permutation-Invariant Relational Interaction for Multivariate Time Series Forecasting

Current methods for multivariate time series forecasting can be classified into channel-dependent and channel-independent models. Channel-dependent models learn cross-channel features but often overfit the channel ordering, which hampers adaptation when channels are added or reordered. Channel-independent models treat each channel in isolation to increase flexibility, yet this neglects inter-channel dependencies and limits performance. To address these limitations, we propose \textbf{CPiRi}, a \textbf{channel permutation invariant (CPI)} framework that infers cross-channel structure from data rather than memorizing a fixed ordering, enabling deployment in settings with structural and distributional co-drift without retraining. CPiRi couples \textbf{spatio-temporal decoupling architecture} with \textbf{permutation-invariant regularization training strategy}: a frozen pretrained temporal encoder extracts high-quality temporal features, a lightweight spatial module learns content-driven inter-channel relations, while a channel shuffling strategy enforces CPI during training. We further \textbf{ground CPiRi in theory} by analyzing permutation equivariance in multivariate time series forecasting. Experiments on multiple benchmarks show state-of-the-art results. CPiRi remains stable when channel orders are shuffled and exhibits strong \textbf{inductive generalization} to unseen channels even when trained on \textbf{only half} of the channels, while maintaining \textbf{practical efficiency} on large-scale datasets. The source code is released at https://github.com/JasonStraka/CPiRi.

Spark: Strategic Policy-Aware Exploration via Dynamic Branching for Long-Horizon Agentic Learning

Reinforcement learning has empowered large language models to act as intelligent agents, yet training them for long-horizon tasks remains challenging due to the scarcity of high-quality trajectories, especially under limited resources. Existing methods typically scale up rollout sizes and indiscriminately allocate computational resources among intermediate steps. Such attempts inherently waste substantial computation budget on trivial steps while failing to guarantee sample quality. To address this, we propose \textbf{Spark} (\textbf{S}trategic \textbf{P}olicy-\textbf{A}ware explo\textbf{R}ation via \textbf{K}ey-state dynamic branching), a novel framework that selectively branches at critical decision states for resource-efficient exploration. Our key insight is to activate adaptive branching exploration at critical decision points to probe promising trajectories, thereby achieving precise resource allocation that prioritizes sampling quality over blind coverage. This design leverages the agent's intrinsic decision-making signals to reduce dependence on human priors, enabling the agent to autonomously expand exploration and achieve stronger generalization. Experiments across diverse tasks (e.g., embodied planning), demonstrate that \textsc{Spark} achieves superior success rates with significantly fewer training samples, exhibiting robust generalization even in unseen scenarios.

DiSa: Saliency-Aware Foreground-Background Disentangled Framework for Open-Vocabulary Semantic Segmentation

Open-vocabulary semantic segmentation aims to assign labels to every pixel in an image based on text labels. Existing approaches typically utilize vision-language models (VLMs), such as CLIP, for dense prediction. However, VLMs, pre-trained on image-text pairs, are biased toward salient, object-centric regions and exhibit two critical limitations when adapted to segmentation: (i) Foreground Bias, which tends to ignore background regions, and (ii) Limited Spatial Localization, resulting in blurred object boundaries. To address these limitations, we introduce DiSa, a novel saliency-aware foreground-background disentangled framework. By explicitly incorporating saliency cues in our designed Saliency-aware Disentanglement Module (SDM), DiSa separately models foreground and background ensemble features in a divide-and-conquer manner. Additionally, we propose a Hierarchical Refinement Module (HRM) that leverages pixel-wise spatial contexts and enables channel-wise feature refinement through multi-level updates. Extensive experiments on six benchmarks demonstrate that DiSa consistently outperforms state-of-the-art methods.

FP8-RL: A Practical and Stable Low-Precision Stack for LLM Reinforcement Learning

Reinforcement learning (RL) for large language models (LLMs) is increasingly bottlenecked by rollout (generation), where long output sequence lengths make attention and KV-cache memory dominate end-to-end step time. FP8 offers an attractive lever for accelerating RL by reducing compute cost and memory traffic during rollout, but applying FP8 in RL introduces unique engineering and algorithmic challenges: policy weights change every step (requiring repeated quantization and weight synchronization into the inference engine) and low-precision rollouts can deviate from the higher-precision policy assumed by the trainer, causing train-inference mismatch and potential instability. This report presents a practical FP8 rollout stack for LLM RL, implemented in the veRL ecosystem with support for common training backends (e.g., FSDP/Megatron-LM) and inference engines (e.g., vLLM/SGLang). We (i) enable FP8 W8A8 linear-layer rollout using blockwise FP8 quantization, (ii) extend FP8 to KV-cache to remove long-context memory bottlenecks via per-step QKV scale recalibration, and (iii) mitigate mismatch using importance-sampling-based rollout correction (token-level TIS/MIS variants). Across dense and MoE models, these techniques deliver up to 44% rollout throughput gains while preserving learning behavior comparable to BF16 baselines.

Atlas: Orchestrating Heterogeneous Models and Tools for Multi-Domain Complex Reasoning

The integration of large language models (LLMs) with external tools has significantly expanded the capabilities of AI agents. However, as the diversity of both LLMs and tools increases, selecting the optimal model-tool combination becomes a high-dimensional optimization challenge. Existing approaches often rely on a single model or fixed tool-calling logic, failing to exploit the performance variations across heterogeneous model-tool pairs. In this paper, we present ATLAS (Adaptive Tool-LLM Alignment and Synergistic Invocation), a dual-path framework for dynamic tool usage in cross-domain complex reasoning. ATLAS operates via a dual-path approach: (1) \textbf{training-free cluster-based routing} that exploits empirical priors for domain-specific alignment, and (2) \textbf{RL-based multi-step routing} that explores autonomous trajectories for out-of-distribution generalization. Extensive experiments across 15 benchmarks demonstrate that our method outperforms closed-source models like GPT-4o, surpassing existing routing methods on both in-distribution (+10.1%) and out-of-distribution (+13.1%) tasks. Furthermore, our framework shows significant gains in visual reasoning by orchestrating specialized multi-modal tools.

Adaptive Dependency-aware Prompt Optimization Framework for Multi-Step LLM Pipeline

Multi-step LLM pipelines invoke large language models multiple times in a structured sequence and can effectively solve complex tasks, but their performance heavily depends on the prompts used at each step. Jointly optimizing these prompts is difficult due to missing step-level supervision and inter-step dependencies. Existing end-to-end prompt optimization methods struggle under these conditions and often yield suboptimal or unstable updates. We propose ADOPT, an Adaptive Dependency-aware Prompt Optimization framework for multi-step LLM pipelines. ADOPT explicitly models the dependency between each LLM step and the final task outcome, enabling precise text-gradient estimation analogous to computing analytical derivatives. It decouples textual gradient estimation from gradient updates, reducing multi-prompt optimization to flexible single-prompt optimization steps, and employs a Shapley-based mechanism to adaptively allocate optimization resources. Experiments on real-world datasets and diverse pipeline structures show that ADOPT is effective and robust, consistently outperforming state-of-the-art prompt optimization baselines.

FLEX-MoE: Federated Mixture-of-Experts with Load-balanced Expert Assignment

Mixture-of-Experts (MoE) models enable scalable neural networks through conditional computation. However, their deployment with federated learning (FL) faces two critical challenges: 1) resource-constrained edge devices cannot store full expert sets, and 2) non-IID data distributions cause severe expert load imbalance that degrades model performance. To this end, we propose \textbf{FLEX-MoE}, a novel federated MoE framework that jointly optimizes expert assignment and load balancing under limited client capacity. Specifically, our approach introduces client-expert fitness scores that quantify the expert suitability for local datasets through training feedback, and employs an optimization-based algorithm to maximize client-expert specialization while enforcing balanced expert utilization system-wide. Unlike existing greedy methods that focus solely on personalization while ignoring load imbalance, our FLEX-MoE is capable of addressing the expert utilization skew, which is particularly severe in FL settings with heterogeneous data. Our comprehensive experiments on three different datasets demonstrate the superior performance of the proposed FLEX-MoE, together with its ability to maintain balanced expert utilization across diverse resource-constrained scenarios.

DeltaMIL: Gated Memory Integration for Efficient and Discriminative Whole Slide Image Analysis

Whole Slide Images (WSIs) are typically analyzed using multiple instance learning (MIL) methods. However, the scale and heterogeneity of WSIs generate highly redundant and dispersed information, making it difficult to identify and integrate discriminative signals. Existing MIL methods either fail to discard uninformative cues effectively or have limited ability to consolidate relevant features from multiple patches, which restricts their performance on large and heterogeneous WSIs. To address this issue, we propose DeltaMIL, a novel MIL framework that explicitly selects semantically relevant regions and integrates the discriminative information from WSIs. Our method leverages the gated delta rule to efficiently filter and integrate information through a block combining forgetting and memory mechanisms. The delta mechanism dynamically updates the memory by removing old values and inserting new ones according to their correlation with the current patch. The gating mechanism further enables rapid forgetting of irrelevant signals. Additionally, DeltaMIL integrates a complementary local pattern mixing mechanism to retain fine-grained pathological locality. Our design enhances the extraction of meaningful cues and suppresses redundant or noisy information, which improves the model's robustness and discriminative power. Experiments demonstrate that DeltaMIL achieves state-of-the-art performance. Specifically, for survival prediction, DeltaMIL improves performance by 3.69\% using ResNet-50 features and 2.36\% using UNI features. For slide-level classification, it increases accuracy by 3.09\% with ResNet-50 features and 3.75\% with UNI features. These results demonstrate the strong and consistent performance of DeltaMIL across diverse WSI tasks.

Beyond a Single Light: A Large-Scale Aerial Dataset for Urban Scene Reconstruction Under Varying Illumination

Recent advances in Neural Radiance Fields and 3D Gaussian Splatting have demonstrated strong potential for large-scale UAV-based 3D reconstruction tasks by fitting the appearance of images. However, real-world large-scale captures are often based on multi-temporal data capture, where illumination inconsistencies across different times of day can significantly lead to color artifacts, geometric inaccuracies, and inconsistent appearance. Due to the lack of UAV datasets that systematically capture the same areas under varying illumination conditions, this challenge remains largely underexplored. To fill this gap, we introduceSkyLume, a large-scale, real-world UAV dataset specifically designed for studying illumination robust 3D reconstruction in urban scene modeling: (1) We collect data from 10 urban regions data comprising more than 100k high resolution UAV images (four oblique views and nadir), where each region is captured at three periods of the day to systematically isolate illumination changes. (2) To support precise evaluation of geometry and appearance, we provide per-scene LiDAR scans and accurate 3D ground-truth for assessing depth, surface normals, and reconstruction quality under varying illumination. (3) For the inverse rendering task, we introduce the Temporal Consistency Coefficient (TCC), a metric that measuress cross-time albedo stability and directly evaluates the robustness of the disentanglement of light and material. We aim for this resource to serve as a foundation that advances research and real-world evaluation in large-scale inverse rendering, geometry reconstruction, and novel view synthesis.