The evolution paradigm of Large Language Models (LLMs) is shifting from scaling training compute to scaling inference-time compute.While Reinforcement Learning with Verifiable Rewards (RLVR) has become a key engine for this transition, standard approaches often fail to equip models with the autonomous improvement capabilities required for test-time scaling.Existing critique-guided methods attempt to mitigate this by leveraging external feedback or ground-truth signals; however, these dependencies are unavailable at test time, fundamentally limiting the model's capacity for continuous self-improvement.To bridge this gap, we propose CURE (Critique-driven Unified REinforcement Learning), a framework that jointly optimizes a single policy for standard solving, critiquing, and guided re-exploration.Uniquely, CURE facilitates re-exploration by generating strategic hints while discarding initial incorrect solutions to mitigate anchoring bias.Empirical results across diverse mathematical reasoning and code generation benchmarks demonstrate that CURE not only maintains competitive single-turn performance but, more importantly, unlocks effective inferencetime scaling, enabling the model to significantly boost accuracy through iterative selfimprovement.

Utilizing Graphic User Interfaces (GUIs) for human-computer interaction is essential for accessing various digital tools. Recent advancements in Vision Language Models (VLMs) reveal significant potential for developing versatile agents that assist humans in navigating GUIs. However, current VLMs face challenges related to fundamental abilities, such as OCR and grounding, as well as a lack of knowledge about GUI elements functionalities and control methods. These limitations hinder their effectiveness as practical GUI agents. To address these challenges, we introduce GUICourse, a series of datasets for training visual-based GUI agents using general VLMs. First, we enhance the OCR and grounding capabilities of VLMs using the GUIEnv dataset. Next, we enrich the GUI knowledge of VLMs using the GUIAct and GUIChat datasets. Our experiments demonstrate that even a small-sized GUI agent (with 3.1 billion parameters) performs effectively on both single-step and multi-step GUI tasks. We further finetune our GUI agents on other GUI tasks with different action spaces (AITW and Mind2Web), and the results show that our agents are better than their baseline VLMs. Additionally, we analyze the impact of OCR and grounding capabilities through an ablation study, revealing a positive correlation with GUI navigation ability.

Agents powered by large language models have shown remarkable abilities in solving complex tasks. However, most agent systems remain reactive, limiting their effectiveness in scenarios requiring foresight and autonomous decision-making. In this paper, we tackle the challenge of developing proactive agents capable of anticipating and initiating tasks without explicit human instructions. We propose a novel data-driven approach for this problem. Firstly, we collect real-world human activities to generate proactive task predictions. These predictions are then labeled by human annotators as either accepted or rejected. The labeled data is used to train a reward model that simulates human judgment and serves as an automatic evaluator of the proactiveness of LLM agents. Building on this, we develop a comprehensive data generation pipeline to create a diverse dataset, ProactiveBench, containing 6,790 events. Finally, we demonstrate that fine-tuning models with the proposed ProactiveBench can significantly elicit the proactiveness of LLM agents. Experimental results show that our fine-tuned model achieves an F1-Score of 66.47% in proactively offering assistance, outperforming all open-source and close-source models. These results highlight the potential of our method in creating more proactive and effective agent systems, paving the way for future advancements in human-agent collaboration.

Data aggregation scheduling is a critical issue in WSNs. This paper studies the Delay efficient Data Aggregation scheduling problem in multi-Channel asynchronous Duty-cycled WSNs (DDACD problem), which aims to accomplish data aggregation with minimum delay. Existing studies, nevertheless, either focus on non-sleeping scenarios or assume that nodes communicate with one single channel, and thus may have poor performance if directly applied to multi-channel asynchronous duty-cycled scenarios. We first show that the DDACD problem is NP-hard. Then, we propose two new concepts of candidate active conflict graphs (CACGs) and feasible active conflict graphs (FACGs) to depict the relationship of the data aggregation links and present two coloring methods to well separate the links at different time slots or on different channels. Based on these two new concepts and two coloring methods, we propose an efficient data aggregation scheduling algorithm called EDAS, which exploits the fewest-children-first rule to choose the forwarding nodes to benefit the link scheduling. To reduce unused time slots or channels, we further propose a novel algorithm called NDAS by making full use of the characteristics of multi-channel asynchronous duty-cycled WSNs. We prove that our algorithms can achieve provable performance guarantee. The results of extensive simulations confirm the efficiency of our algorithms.

Data aggregation scheduling is a critical issue in wireless sensor networks (WSNs). This paper studies the Delay efficient Data Aggregation scheduling problem in multi-Channel Duty-cycled WSNs (DDACD problem), which aims to accomplish data aggregation with minimum delay. Existing researches, nevertheless, either focus on non-sleeping scenarios, or assume that nodes communicate on one single channel, and thus have poor performance in multi-channel duty-cycled scenarios. In this paper, we first show that DDACD problem is NP-hard. We then propose two new concepts of Candidate Active Conflict Graphs (CACG) and Feasible Active Conflict Graphs (FACG) to depict the relationship of the data aggregation links, and present two coloring methods to well separate the links at different time-slots or on different channels. Based on these two new concepts and two coloring methods, we propose an Efficient Data Aggregation Scheduling algorithm called EDAS, which exploits the fewest-children-first rule to choose the forwarding nodes to benefit the link scheduling. We theoretically prove that our proposed EDAS algorithm can achieve provable performance guarantee. The results of extensive simulations confirm the efficiency of our algorithm.

This paper investigates the problem of maximizing uniform multicast throughput (MUMT) for multi-channel dense wireless sensor networks, where all nodes locate within one-hop transmission range and can communicate with each other on multiple orthogonal channels. This kind of networks show wide application in the real world, and maximizing uniform multicast throughput for these networks is worth deep studying. Previous researches have proved MUMT problem is NP-hard. However, previous researches are either hard to implement, or use too many relay nodes to complete the multicast task, and thus incur high overhead or poor performance. To efficiently solve MUMT problem, we adopt the concept of the maximum independent set with the size constraint, and present one novel Single-Broadcast based Multicast algorithm called SBM based on the concept. We prove that SBM algorithm achieves a constant ratio to the theoretical throughput upper bound. Extensive experimental results demonstrate that, SBM performs better than existing work in terms of both the uniform multicast throughput and the total number of transmissions.

Quantitative understanding of people's behaviors on social networks has significant meaning to reveal the origins of many socioeconomic phenomena. This paper focuses on the time characteristic of human behavior online. Four famous web forums of China were analyzed, including Sina, NetEase, HuBeiDongHu and LiXiang. The empirical analysis result presents some statistical features on the behaviors of forum users. We found that the intervals of user's post and reply behaviors both follow heavy-tailed distribution and have high burstiness and low memory. But the day activities do not follow heavy-tailed distribution and the values concentrate between 10 and 100. Further, we studied the intra-day post and reply behaviors and showed that people's behaviors online have similar time characteristic with that of real life, and they submit obviously more posts and replies on work time and seldom do actions after midnight. The findings of this paper reveal some interesting features of human behaviors on online forums and are helpful to the behavior-based detection of spam and spammer.

Water army means a special group of online users who get paid for posting comments and new threads or articles on different online communities and websites for some hidden purposes. Due to the fact that the nature of the posting behavior of water army is not fully and understood, the driving force detection of the bursty topic for web forum is still a difficult problem to solve. According to the analysis of bursty topics evolution and the posting behavior of water army, it is found that the topics driven by water army exhibit the characteristics different from general topics in their latency stage. Based on this discovery, the paper proposes a novel bursty topic classification algorithm, based on SVM active learning, which transforms the water army detection issue to a SVM-based classification decision issue. The experimental results show that the proposed algorithm has higher detection accuracy and detection efficiency.