Recursive self-modification has often been discussed as a cornerstone for building continually improving ML systems (Y ampolskiy, 2015). Modern ML already hints at this trend: reinforcement learning agents tune hyperparameters online, AutoML loops search over training recipes, and optimization pipelines reconfigure code and settings during runs. Y et these procedures often adopt changes on the basis of noisy gains, creating the risk of harmful edits - modifications that seems beneficial in finite trials but ultimately degrade true performance. Such risks are especially concerning in high-stakes scientific domains such as drug design, protein engineering, or climate modeling, where spurious gains can misdirect costly pipelines. G odel machines (Schmidhuber, 2007) offer a conceptually clean answer: an agent rewrites its code only when it can prove the rewrite increases expected utility. But in stochastic, high-dimensional ML, such formal proofs are unattainable. At the other extreme, practical AutoML and RL systems adopt edits using heuristics such as rolling averages, best-of-seeds, or bandit rules, which lack guarantees and may silently accumulate regressions.

The three-dimensional bin packing problem (3D-BPP) is widely applied in logistics and warehousing. Existing learning-based approaches often neglect practical stability-related constraints and exhibit limitations in generalizing across diverse bin dimensions. To address these limitations, we propose a novel deep reinforcement learning framework, One4Many-StablePacker (O4M-SP). The primary advantage of O4M-SP is its ability to handle various bin dimensions in a single training process while incorporating support and weight constraints common in practice. Our training method introduces two innovative mechanisms. First, it employs a weighted reward function that integrates loading rate and a new height difference metric for packing layouts, promoting improved bin utilization through flatter packing configurations. Second, it combines clipped policy gradient optimization with a tailored policy drifting method to mitigate policy entropy collapse, encouraging exploration at critical decision nodes during packing to avoid suboptimal solutions. Extensive experiments demonstrate that O4M-SP generalizes successfully across diverse bin dimensions and significantly outperforms baseline methods. Furthermore, O4M-SP exhibits strong practical applicability by effectively addressing packing scenarios with stability constraints.

Abstract--The rapid advancement of Low-Altitude Economy Networks (LAENets) has enabled a variety of applications, including aerial surveillance, environmental sensing, and semantic data collection. T o support these scenarios, unmanned aerial vehicles (UA Vs) equipped with onboard vision-language models (VLMs) offer a promising solution for real-time multimodal inference. However, ensuring both inference accuracy and communication efficiency remains a significant challenge due to limited onboard resources and dynamic network conditions. In this paper, we first propose a UA V-enabled LAENet system model that jointly captures UA V mobility, user-UA V communication, and the onboard visual question answering (VQA) pipeline. Based on this model, we formulate a mixed-integer non-convex optimization problem to minimize task latency and power consumption under user-specific accuracy constraints. T o solve the problem, we design a hierarchical optimization framework composed of two parts: (i) an Alternating Resolution and Power Optimization (ARPO) algorithm for resource allocation under accuracy constraints, and (ii) a Large Language Model-augmented Reinforcement Learning Approach (LLaRA) for adaptive UA V trajectory optimization. The large language model (LLM) serves as an expert in refining reward design of reinforcement learning in an offline fashion, introducing no additional latency in real-time decision-making. Numerical results demonstrate the efficacy of our proposed framework in improving inference performance and communication efficiency under dynamic LAENet conditions. Low-Altitude Economy Networks (LAENets) have recently garnered growing attention as a novel paradigm that leverages the low-altitude airspace (typically below 1000 meters) to deliver digital services [1]. Li and G. Liu are with the College of Computing and Data Science, the Energy Research Institute @ NTU, Interdisciplinary Graduate Program, Nanyang Technological University, Singapore (e-mail: yang048@e.ntu.edu.sg; Liu and D. Niyato are with the College of Computing and Data Science, Nanyang Technological University, Singapore (e-mails: ruichen.zhang@ntu.edu.sg; X. Wang is with the Department of Electrical and Computer Engineering, Western University, London, Canada (e-mail: xianbin.wang@uwo.ca).

Reasoning-augmented search agents, such as Search-R1, are trained to reason, search, and generate the final answer iteratively. Nevertheless, due to their limited capabilities in reasoning and search, their performance on multi-hop QA benchmarks remains far from satisfactory. To handle complex or compound queries, we train an LLM-based search agent with the native capability of query expansion through reinforcement learning. In each turn, our search agent proposes several query variants, which are searched simultaneously to cover more relevant information. Meanwhile, given limited post-training data and computing resources, it is very challenging for a search agent to master multiple tasks, including query generation, retrieved information understanding, and answer generation. Therefore, we propose incorporating a pre-trained squeezer model that helps the search agent understand the retrieved documents, allowing the search agent to focus on query generation for high retrieval recall. With the assistance of the squeezer model, we discover that even a small-scale 3B LLM can demonstrate a strong capability of query expansion and achieve state-of-the-art accuracy on the multi-hop QA benchmarks. To be specific, our experiments across seven question-answering benchmarks demonstrate that our method, named ExpandSearch, achieves an average improvement of 4.4% compared to state-of-the-art baselines, with strong gains on multi-hop reasoning tasks requiring diverse evidence aggregation.

Vision-Language-Action (VLA) models such as OpenVLA, Octo, and $π_0$ have shown strong generalization by leveraging large-scale demonstrations, yet their performance is still fundamentally constrained by the quality and coverage of supervised data. Reinforcement learning (RL) provides a promising path for improving and fine-tuning VLAs through online interaction. However, conventional policy gradient methods are computationally infeasible in the context of flow-matching based models due to the intractability of the importance sampling process, which requires explicit computation of policy ratios. To overcome this limitation, we propose Flow Policy Optimization (FPO) algorithm, which reformulates importance sampling by leveraging per-sample changes in the conditional flow-matching objective. Furthermore, FPO achieves stable and scalable online reinforcement fine-tuning of the $π_0$ model by integrating structure-aware credit assignment to enhance gradient efficiency, clipped surrogate objectives to stabilize optimization, multi-step latent exploration to encourage diverse policy updates, and a Q-ensemble mechanism to provide robust value estimation. We evaluate FPO on the LIBERO benchmark and the ALOHA simulation task against supervised, preference-aligned, diffusion-based, autoregressive online RL, and $π_0$-FAST baselines, observing consistent improvements over the imitation prior and strong alternatives with stable learning under sparse rewards. In addition, ablation studies and analyses of the latent space dynamics further highlight the contributions of individual components within FPO, validating the effectiveness of the proposed computational modules and the stable convergence of the conditional flow-matching objective during online RL.

Weather forecasting is a fundamental task in spatiotemporal data analysis, with broad applications across a wide range of domains. Existing data-driven forecasting methods typically model atmospheric dynamics over a fixed short time interval (e.g., 6 hours) and rely on naive autoregression-based rollout for long-term forecasting (e.g., 138 hours). However, this paradigm suffers from two key limitations: (1) it often inadequately models the spatial and multi-scale temporal dependencies inherent in global weather systems, and (2) the rollout strategy struggles to balance error accumulation with the capture of fine-grained atmospheric variations. In this study, we propose ARROW, an Adaptive-Rollout Multi-scale temporal Routing method for Global Weather Forecasting. To contend with the first limitation, we construct a multi-interval forecasting model that forecasts weather across different time intervals. Within the model, the Shared-Private Mixture-of-Experts captures both shared patterns and specific characteristics of atmospheric dynamics across different time scales, while Ring Positional Encoding accurately encodes the circular latitude structure of the Earth when representing spatial information. For the second limitation, we develop an adaptive rollout scheduler based on reinforcement learning, which selects the most suitable time interval to forecast according to the current weather state. Experimental results demonstrate that ARROW achieves state-of-the-art performance in global weather forecasting, establishing a promising paradigm in this field.

Diffusion large language models (dLLMs) are emerging as an efficient alternative to autoregressive models due to their ability to decode multiple tokens in parallel. However, aligning dLLMs with human preferences or task-specific rewards via reinforcement learning (RL) is challenging because their intractable log-likelihood precludes the direct application of standard policy gradient methods. While prior work uses surrogates like the evidence lower bound (ELBO), these one-sided approximations can introduce significant policy gradient bias. To address this, we propose the Sandwiched Policy Gradient (SPG) that leverages both an upper and a lower bound of the true log-likelihood. Experiments show that SPG significantly outperforms baselines based on ELBO or one-step estimation. Specifically, SPG improves the accuracy over state-of-the-art RL methods for dLLMs by 3.6% in GSM8K, 2.6% in MATH500, 18.4% in Countdown and 27.0% in Sudoku.

Flow Matching (FM) has shown remarkable ability in modeling complex distributions and achieves strong performance in offline imitation learning for cloning expert behaviors. However, despite its behavioral cloning expressiveness, FM-based policies are inherently limited by their lack of environmental interaction and exploration. This leads to poor generalization in unseen scenarios beyond the expert demonstrations, underscoring the necessity of online interaction with environment. Unfortunately, optimizing FM policies via online interaction is challenging and inefficient due to instability in gradient computation and high inference costs. To address these issues, we propose to let a student policy with simple MLP structure explore the environment and be online updated via RL algorithm with a reward model. This reward model is associated with a teacher FM model, containing rich information of expert data distribution. Furthermore, the same teacher FM model is utilized to regularize the student policy's behavior to stabilize policy learning. Due to the student's simple architecture, we avoid the gradient instability of FM policies and enable efficient online exploration, while still leveraging the expressiveness of the teacher FM model. Extensive experiments show that our approach significantly enhances learning efficiency, generalization, and robustness, especially when learning from suboptimal expert data.

We propose FSPO (Fair Sequence Policy Optimization), a sequence-level reinforcement learning method for LLMs that enforces length-fair clipping on the importance-sampling (IS) weight. We study RL methods with sequence-level IS and identify a mismatch when PPO/GRPO-style clipping is transplanted to sequences: a fixed clip range systematically reweights short vs. long responses, distorting the optimization direction. FSPO introduces a simple remedy: we clip the sequence log-IS ratio with a band that scales as $\sqrt{L}$. Theoretically, we formalize length fairness via a Length Reweighting Error (LRE) and prove that small LRE yields a cosine directional guarantee between the clipped and true updates. Empirically, FSPO flattens clip rates across length bins, stabilizes training, and outperforms baselines across model sizes and evaluation datasets, with the largest gains on the Qwen3-8B-Base model.

We derive a novel PAC-Bayesian generalization bound for reinforcement learning that explicitly accounts for Markov dependencies in the data, through the chain's mixing time. This contributes to overcoming challenges in obtaining generalization guarantees for reinforcement learning, where the sequential nature of data breaks the independence assumptions underlying classical bounds. Our bound provides non-vacuous certificates for modern off-policy algorithms like Soft Actor-Critic. We demonstrate the bound's practical utility through PB-SAC, a novel algorithm that optimizes the bound during training to guide exploration. Experiments across continuous control tasks show that our approach provides meaningful confidence certificates while maintaining competitive performance.