High-quality long-context data is essential for training large language models (LLMs) capable of processing extensive documents, yet existing synthesis approaches using relevance-based aggregation face challenges of computational efficiency. We present LiteLong, a resource-efficient method for synthesizing long-context data through structured topic organization and multi-agent debate. Our approach leverages the BISAC book classification system to provide a comprehensive hierarchical topic organization, and then employs a debate mechanism with multiple LLMs to generate diverse, high-quality topics within this structure. For each topic, we use lightweight BM25 retrieval to obtain relevant documents and concatenate them into 128K-token training samples. Experiments on HELMET and Ruler benchmarks demonstrate that LiteLong achieves competitive long-context performance and can seamlessly integrate with other long-dependency enhancement methods. LiteLong makes high-quality long-context data synthesis more accessible by reducing both computational and data engineering costs, facilitating further research in long-context language training.

Large Language Models (LLMs) often struggle with mathematical reasoning tasks requiring precise, verifiable computation. While Reinforcement Learning (RL) from outcome-based rewards enhances text-based reasoning, understanding how agents autonomously learn to leverage external tools like code execution remains crucial. We investigate RL from outcome-based rewards for Tool-Integrated Reasoning, ZeroTIR, training base LLMs to spontaneously generate and execute Python code for mathematical problems without supervised tool-use examples. Our central contribution is we demonstrate that as RL training progresses, key metrics scale predictably. Specifically, we observe strong positive correlations where increased training steps lead to increases in the spontaneous code execution frequency, the average response length, and, critically, the final task accuracy. This suggests a quantifiable relationship between computational effort invested in training and the emergence of effective, tool-augmented reasoning strategies. We implement a robust framework featuring a decoupled code execution environment and validate our findings across standard RL algorithms and frameworks. Experiments show ZeroTIR significantly surpasses non-tool ZeroRL baselines on challenging math benchmarks. Our findings provide a foundational understanding of how autonomous tool use is acquired and scales within Agent RL, offering a reproducible benchmark for future studies. Code is released at \href{https://github.com/yyht/openrlhf_async_pipline}{https://github.com/yyht/openrlhf\_async\_pipline}.

Recent interactive segmentation methods iteratively take source image, user guidance and previously predicted mask as the input without considering the invariant nature of the source image. As a result, the process of extracting features from the source image is repeated in each interaction, resulting in substantial computational redundancy. In this work, we propose the Feature Decoupling-Recycling Network (FDRN), which decouples the modeling components based on their intrinsic discrepancies and then recycles components that can be reused for each user interaction. Thus, the efficiency of the whole interactive process can be significantly improved. To be specific, we apply the Decoupling-Recycling strategy from three perspectives to address three types of discrepancies, respectively. First, our model decouples the learning of source image semantics from the encoding of user guidance to process two types of input domains separately. Second, FDRN decouples high-level and low-level features from stratified semantic representations to enhance feature learning. Third, during the encoding of user guidance, current user guidance is decoupled from historical guidance to highlight the effect of current user guidance. We conduct extensive experiments on 6 datasets from different domains and modalities, which demonstrate the following merits of our model: 1) superior efficiency than other methods, particularly advantageous in the challenging scenarios requiring long-term interactions (up to 4.25x faster), while achieving favorable segmentation performance; 2) strong applicability to various methods serving as a universal enhancement technique; 3) well cross-task generalizability, e.g., to medical image segmentation, and robustness against misleading user guidance.

Person re-identification aims to identify whether pairs of images belong to the same person or not. This problem is challenging due to large differences in camera views, lighting and background. One of the mainstream in learning CNN features is to design loss functions which reinforce both the class separation and intra-class compactness. In this paper, we propose a novel Orthogonal Center Learning method with Subspace Masking for person re-identification. We make the following contributions: 1) we develop a center learning module to learn the class centers by simultaneously reducing the intra-class differences and inter-class correlations by orthogonalization; 2) we introduce a subspace masking mechanism to enhance the generalization of the learned class centers; and 3) we propose to integrate the average pooling and max pooling in a regularizing manner that fully exploits their powers. Extensive experiments show that our proposed method consistently outperforms the state-of-the-art methods on large-scale ReID datasets including Market-1501, DukeMTMC-ReID, CUHK03 and MSMT17.

This paper extends previous work on modeling stochastic decision discrete-event systems (DDES) through a generalized semi-Markov decision process (GSMDP), which discards any restrictive unrealistic assumptions and can be applied to complex cases including non-Markovian environment. Moreover, as a typical example, we develop a GSMDP model for the optimal call admission control (CAC) problem in an integrated voice/data wireless network supporting multiple traffic types with different resource requirements. In contrast to existing methods, this approach can better model the real world of the next generation wireless network behaviors. Besides, through a form of reinforcement learning algorithm known as Q-learning, we can solve the Bellman optimality equation with requiring neither explicit state transition probabilities nor any assumptions behind the network model.

Influenced by end-host overlay systems, the overlay multicast routing problem is studied under a singlehomed model in most literatures. This model is no longer applicable to those fixed MSNs(multicast service node), which may have multiple access links to the physical network. In order to solve the problem correctly, we propose a multihomed proxy model in this paper to describe the overlay network more accurately. We also develop a general solution to the routing problem under our model, which includes a path selection process and a multicast routing process. In the end we design an OMMP (overlay multicast over multihomed proxy) architecture, which integrates all MSNs into one framework to realize the overlay multicast service under our model.

For scalability and reliability, gossip protocol is widely used in distributed database updates, information dissemination, membership maintenance et al. Spatial gossip [9], which means gossiping within a finite distance independent of the network size, was also proposed to rapidly locate resource in some networks, such as wireless sensor networks. The previous study on spatial gossip used a percolation model, where in each step, a node x chooses a random node y to forward message with probability \beta /(d_{xy} +1)^8 . The study showed a fine result that if D \le s \le 2D, where D is the dimension of the network, the propagation time steps for a message propagated to a node within distance d is O(\log ^{2+\varepsilon}d), with probability 1 -o(\log ^{-k}d). Our work uses a similar model, but let s = D, we show through explicit computation that the propagation time is better, i.e, O(log d) with probability 1 - O(d^k ). We further discuss other cases where s is chosen different values, and point out the limitation of the percolation model.

The paper presents secure protocols and new secure challenges based on looser radio propagation assumptions. According to these, a new secure enhancement mechanism for secure protocols is proposed. Then we exemplify the DoS attacks and protections as the illustration. The research indicates it will increase the usability of secure protocols in common applications.