Revolutionizing Reinforcement Learning Framework for Diffusion Large Language Models

We propose TraceRL, a trajectory-aware reinforcement learning framework for diffusion language models (DLMs) that incorporates preferred inference trajectory into post-training, and is applicable across different architectures. Equipped with a diffusion-based value model that enhances training stability, we demonstrate improved reasoning performance on complex math and coding tasks. Besides, it can also be applied to adapt block-specific models to larger blocks, which improves sampling flexibility. Employing TraceRL, we derive a series of state-of-the-art diffusion language models, namely TraDo. Although smaller than 7B-scale AR models, TraDo-4B-Instruct still consistently outperforms them across complex math reasoning tasks. TraDo-8B-Instruct achieves relative accuracy improvements of 6.1% over Qwen2.5-7B-Instruct and 51.3% over Llama3.1-8B-Instruct on mathematical reasoning benchmarks. Through curriculum learning, we also derive the first long-CoT DLM, outperforming Qwen2.5-7B-Instruct on MATH500 with an 18.1% relative accuracy gain. To facilitate reproducible research and practical applications, we release a comprehensive open-source framework for building, training, and deploying diffusion LLMs across diverse architectures. The framework integrates accelerated KV-cache techniques and inference engines for both inference and reinforcement learning, and includes implementations of various supervised fine-tuning and RL methods for mathematics, coding, and general tasks. Code and Models: https://github.com/Gen-Verse/dLLM-RL

LMAE4Eth: Generalizable and Robust Ethereum Fraud Detection by Exploring Transaction Semantics and Masked Graph Embedding

Current Ethereum fraud detection methods rely on context-independent, numerical transaction sequences, failing to capture semantic of account transactions. Furthermore, the pervasive homogeneity in Ethereum transaction records renders it challenging to learn discriminative account embeddings. Moreover, current self-supervised graph learning methods primarily learn node representations through graph reconstruction, resulting in suboptimal performance for node-level tasks like fraud account detection, while these methods also encounter scalability challenges. To tackle these challenges, we propose LMAE4Eth, a multi-view learning framework that fuses transaction semantics, masked graph embedding, and expert knowledge. We first propose a transaction-token contrastive language model (TxCLM) that transforms context-independent numerical transaction records into logically cohesive linguistic representations. To clearly characterize the semantic differences between accounts, we also use a token-aware contrastive learning pre-training objective together with the masked transaction model pre-training objective, learns high-expressive account representations. We then propose a masked account graph autoencoder (MAGAE) using generative self-supervised learning, which achieves superior node-level account detection by focusing on reconstructing account node features. To enable MAGAE to scale for large-scale training, we propose to integrate layer-neighbor sampling into the graph, which reduces the number of sampled vertices by several times without compromising training quality. Finally, using a cross-attention fusion network, we unify the embeddings of TxCLM and MAGAE to leverage the benefits of both. We evaluate our method against 21 baseline approaches on three datasets. Experimental results show that our method outperforms the best baseline by over 10% in F1-score on two of the datasets.

Breaking Obfuscation: Cluster-Aware Graph with LLM-Aided Recovery for Malicious JavaScript Detection

With the rapid expansion of web-based applications and cloud services, malicious JavaScript code continues to pose significant threats to user privacy, system integrity, and enterprise security. But, detecting such threats remains challenging due to sophisticated code obfuscation techniques and JavaScript's inherent language characteristics, particularly its nested closure structures and syntactic flexibility. In this work, we propose DeCoda, a hybrid defense framework that combines large language model (LLM)-based deobfuscation with code graph learning: (1) We first construct a sophisticated prompt-learning pipeline with multi-stage refinement, where the LLM progressively reconstructs the original code structure from obfuscated inputs and then generates normalized Abstract Syntax Tree (AST) representations; (2) In JavaScript ASTs, dynamic typing scatters semantically similar nodes while deeply nested functions fracture scope capturing, introducing structural noise and semantic ambiguity. To address these challenges, we then propose to learn hierarchical code graph representations via a Cluster-wise Graph that synergistically integrates graph transformer network, node clustering, and node-to-cluster attention to simultaneously capture both local node-level semantics and global cluster-induced structural relationships from AST graph. Experimental results demonstrate that our method achieves F1-scores of 94.64% and 97.71% on two benchmark datasets, demonstrating absolute improvements of 10.74% and 13.85% over state-of-the-art baselines. In false-positive control evaluation at fixed FPR levels (0.0001, 0.001, 0.01), our approach delivers 4.82, 5.91, and 2.53 higher TPR respectively compared to the best-performing baseline. These results highlight the effectiveness of LLM-based deobfuscation and underscore the importance of modeling cluster-level relationships in detecting malicious code.

Packet-Level DDoS Data Augmentation Using Dual-Stream Temporal-Field Diffusion

In response to Distributed Denial of Service (DDoS) attacks, recent research efforts increasingly rely on Machine Learning (ML)-based solutions, whose effectiveness largely depends on the quality of labeled training datasets. To address the scarcity of such datasets, data augmentation with synthetic traces is often employed. However, current synthetic trace generation methods struggle to capture the complex temporal patterns and spatial distributions exhibited in emerging DDoS attacks. This results in insufficient resemblance to real traces and unsatisfied detection accuracy when applied to ML tasks. In this paper, we propose Dual-Stream Temporal-Field Diffusion (DSTF-Diffusion), a multi-view, multi-stream network traffic generative model based on diffusion models, featuring two main streams: The field stream utilizes spatial mapping to bridge network data characteristics with pre-trained realms of stable diffusion models, effectively translating complex network interactions into formats that stable diffusion can process, while the spatial stream adopts a dynamic temporal modeling approach, meticulously capturing the intrinsic temporal patterns of network traffic. Extensive experiments demonstrate that data generated by our model exhibits higher statistical similarity to originals compared to current state-of-the-art solutions, and enhance performances on a wide range of downstream tasks.

WebGuard++:Interpretable Malicious URL Detection via Bidirectional Fusion of HTML Subgraphs and Multi-Scale Convolutional BERT

URL+HTML feature fusion shows promise for robust malicious URL detection, since attacker artifacts persist in DOM structures. However, prior work suffers from four critical shortcomings: (1) incomplete URL modeling, failing to jointly capture lexical patterns and semantic context; (2) HTML graph sparsity, where threat-indicative nodes (e.g., obfuscated scripts) are isolated amid benign content, causing signal dilution during graph aggregation; (3) unidirectional analysis, ignoring URL-HTML feature bidirectional interaction; and (4) opaque decisions, lacking attribution to malicious DOM components. To address these challenges, we present WebGuard++, a detection framework with 4 novel components: 1) Cross-scale URL Encoder: Hierarchically learns local-to-global and coarse to fine URL features based on Transformer network with dynamic convolution. 2) Subgraph-aware HTML Encoder: Decomposes DOM graphs into interpretable substructures, amplifying sparse threat signals via Hierarchical feature fusion. 3) Bidirectional Coupling Module: Aligns URL and HTML embeddings through cross-modal contrastive learning, optimizing inter-modal consistency and intra-modal specificity. 4) Voting Module: Localizes malicious regions through consensus voting on malicious subgraph predictions. Experiments show WebGuard++ achieves significant improvements over state-of-the-art baselines, achieving 1.1x-7.9x higher TPR at fixed FPR of 0.001 and 0.0001 across both datasets.

MudiNet: Task-guided Disentangled Representation Learning for 5G Indoor Multipath-assisted Positioning

In the fifth-generation communication system (5G), multipath-assisted positioning (MAP) has emerged as a promising approach. With the enhancement of signal resolution, multipath component (MPC) are no longer regarded as noise but rather as valuable information that can contribute to positioning. However, existing research often treats reflective surfaces as ideal reflectors, while being powerless in the face of indistinguishable multipath caused by diffuse reflectors. This study approaches diffuse reflectors from the perspective of uncertainty, investigating the statistical distribution characteristics of indoor diffuse and specular reflectors. Based on these insights, a task-guided disentangled representation learning method leveraging multi-time channel impulse response (CIR) observations is designed to directly map CIRs to positions, while mitigating the adverse effects of components that contribute minimally to localization accuracy (e.g., diffuse multipath).In this semi-supervised learning framework, a global feature extraction architecture based on self-attention is proposed to capture location-independent wireless environmental information, while an MLP is employed to extract the time-varying features related to user equipment (UE) positions. Variational inference based on a latent variable model (LVM) is applied to separate independent features within the CIR, with position labels guiding the LVM to express components more beneficial for localization. Additionally, we provide a feasibility proof for the separability of diffuse and specular environmental features in CIRs. Simulation results demonstrate that the proposed method achieves higher localization accuracy compared to conventional search-based localization methods, with enhanced robustness against indistinguishable multipath from diffuse reflectors.

Cross-Task Experiential Learning on LLM-based Multi-Agent Collaboration

Large Language Model-based multi-agent systems (MAS) have shown remarkable progress in solving complex tasks through collaborative reasoning and inter-agent critique. However, existing approaches typically treat each task in isolation, resulting in redundant computations and limited generalization across structurally similar tasks. To address this, we introduce multi-agent cross-task experiential learning (MAEL), a novel framework that endows LLM-driven agents with explicit cross-task learning and experience accumulation. We model the task-solving workflow on a graph-structured multi-agent collaboration network, where agents propagate information and coordinate via explicit connectivity. During the experiential learning phase, we quantify the quality for each step in the task-solving workflow and store the resulting rewards along with the corresponding inputs and outputs into each agent's individual experience pool. During inference, agents retrieve high-reward, task-relevant experiences as few-shot examples to enhance the effectiveness of each reasoning step, thereby enabling more accurate and efficient multi-agent collaboration. Experimental results on diverse datasets demonstrate that MAEL empowers agents to learn from prior task experiences effectively-achieving faster convergence and producing higher-quality solutions on current tasks.

Co-Saving: Resource Aware Multi-Agent Collaboration for Software Development

Recent advancements in Large Language Models (LLMs) and autonomous agents have demonstrated remarkable capabilities across various domains. However, standalone agents frequently encounter limitations when handling complex tasks that demand extensive interactions and substantial computational resources. Although Multi-Agent Systems (MAS) alleviate some of these limitations through collaborative mechanisms like task decomposition, iterative communication, and role specialization, they typically remain resource-unaware, incurring significant inefficiencies due to high token consumption and excessive execution time. To address these limitations, we propose a resource-aware multi-agent system -- Co-Saving (meaning that multiple agents collaboratively engage in resource-saving activities), which leverages experiential knowledge to enhance operational efficiency and solution quality. Our key innovation is the introduction of "shortcuts" -- instructional transitions learned from historically successful trajectories -- which allows to bypass redundant reasoning agents and expedite the collective problem-solving process. Experiments for software development tasks demonstrate significant advantages over existing methods. Specifically, compared to the state-of-the-art MAS ChatDev, our method achieves an average reduction of 50.85% in token usage, and improves the overall code quality by 10.06%.

Multi-Agent Collaboration via Evolving Orchestration

Large language models (LLMs) have achieved remarkable results across diverse downstream tasks, but their monolithic nature restricts scalability and efficiency in complex problem-solving. While recent research explores multi-agent collaboration among LLMs, most approaches rely on static organizational structures that struggle to adapt as task complexity and agent numbers grow, resulting in coordination overhead and inefficiencies. To this end, we propose a puppeteer-style paradigm for LLM-based multi-agent collaboration, where a centralized orchestrator ("puppeteer") dynamically directs agents ("puppets") in response to evolving task states. This orchestrator is trained via reinforcement learning to adaptively sequence and prioritize agents, enabling flexible and evolvable collective reasoning. Experiments on closed- and open-domain scenarios show that this method achieves superior performance with reduced computational costs. Analyses further reveal that the key improvements consistently stem from the emergence of more compact, cyclic reasoning structures under the orchestrator's evolution.

A Feasibility Study of Task-Based fMRI at 0.55 T

0.55T MRI offers advantages compared to conventional field strengths, including reduced susceptibility artifacts and better compatibility with simultaneous EEG recordings. However, reliable task-based fMRI at 0.55T has not been significantly demonstrated. In this study, we establish a robust task-based fMRI protocol and analysis pipeline at 0.55T that achieves full brain coverage and results comparable to what is expected for activation extent and location. We attempted fMRI at 0.55T by combining EPI acquisition with custom analysis techniques. Finger-tapping and visual tasks were used, comparing 5- and 10-minute runs to enhance activation detection. The results show significant activations, demonstrating that high-quality task-based fMRI is achievable at 0.55T in single subjects. This study demonstrates that reliable task-based fMRI is feasible on 0.55T scanners, potentially broadening functional neuroimaging access in clinical and research settings where high-field MRI is unavailable or impractical, supporting broader diagnostic and research applications.