Mastering Massive Multi-Task Reinforcement Learning via Mixture-of-Expert Decision Transformer

Despite recent advancements in offline multi-task reinforcement learning (MTRL) have harnessed the powerful capabilities of the Transformer architecture, most approaches focus on a limited number of tasks, with scaling to extremely massive tasks remaining a formidable challenge. In this paper, we first revisit the key impact of task numbers on current MTRL method, and further reveal that naively expanding the parameters proves insufficient to counteract the performance degradation as the number of tasks escalates. Building upon these insights, we propose M3DT, a novel mixture-of-experts (MoE) framework that tackles task scalability by further unlocking the model's parameter scalability. Specifically, we enhance both the architecture and the optimization of the agent, where we strengthen the Decision Transformer (DT) backbone with MoE to reduce task load on parameter subsets, and introduce a three-stage training mechanism to facilitate efficient training with optimal performance. Experimental results show that, by increasing the number of experts, M3DT not only consistently enhances its performance as model expansion on the fixed task numbers, but also exhibits remarkable task scalability, successfully extending to 160 tasks with superior performance.

Lifelong Safety Alignment for Language Models

LLMs have made impressive progress, but their growing capabilities also expose them to highly flexible jailbreaking attacks designed to bypass safety alignment. While many existing defenses focus on known types of attacks, it is more critical to prepare LLMs for unseen attacks that may arise during deployment. To address this, we propose a lifelong safety alignment framework that enables LLMs to continuously adapt to new and evolving jailbreaking strategies. Our framework introduces a competitive setup between two components: a Meta-Attacker, trained to actively discover novel jailbreaking strategies, and a Defender, trained to resist them. To effectively warm up the Meta-Attacker, we first leverage the GPT-4o API to extract key insights from a large collection of jailbreak-related research papers. Through iterative training, the first iteration Meta-Attacker achieves a 73% attack success rate (ASR) on RR and a 57% transfer ASR on LAT using only single-turn attacks. Meanwhile, the Defender progressively improves its robustness and ultimately reduces the Meta-Attacker's success rate to just 7%, enabling safer and more reliable deployment of LLMs in open-ended environments. The code is available at https://github.com/sail-sg/LifelongSafetyAlignment.

Reinforcement Fine-Tuning Powers Reasoning Capability of Multimodal Large Language Models

Standing in 2025, at a critical juncture in the pursuit of Artificial General Intelligence (AGI), reinforcement fine-tuning (RFT) has demonstrated significant potential in enhancing the reasoning capability of large language models (LLMs) and has led to the development of cutting-edge AI models such as OpenAI-o1 and DeepSeek-R1. Moreover, the efficient application of RFT to enhance the reasoning capability of multimodal large language models (MLLMs) has attracted widespread attention from the community. In this position paper, we argue that reinforcement fine-tuning powers the reasoning capability of multimodal large language models. To begin with, we provide a detailed introduction to the fundamental background knowledge that researchers interested in this field should be familiar with. Furthermore, we meticulously summarize the improvements of RFT in powering reasoning capability of MLLMs into five key points: diverse modalities, diverse tasks and domains, better training algorithms, abundant benchmarks and thriving engineering frameworks. Finally, we propose five promising directions for future research that the community might consider. We hope that this position paper will provide valuable insights to the community at this pivotal stage in the advancement toward AGI. Summary of works done on RFT for MLLMs is available at https://github.com/Sun-Haoyuan23/Awesome-RL-based-Reasoning-MLLMs.

VTire: A Bimodal Visuotactile Tire with High-Resolution Sensing Capability

Developing smart tires with high sensing capability is significant for improving the moving stability and environmental adaptability of wheeled robots and vehicles. However, due to the classical manufacturing design, it is always challenging for tires to infer external information precisely. To this end, this paper introduces a bimodal sensing tire, which can simultaneously capture tactile and visual data. By leveraging the emerging visuotactile techniques, the proposed smart tire can realize various functions, including terrain recognition, ground crack detection, load sensing, and tire damage detection. Besides, we optimize the material and structure of the tire to ensure its outstanding elasticity, toughness, hardness, and transparency. In terms of algorithms, a transformer-based multimodal classification algorithm, a load detection method based on finite element analysis, and a contact segmentation algorithm have been developed. Furthermore, we construct an intelligent mobile platform to validate the system's effectiveness and develop visual and tactile datasets in complex terrains. The experimental results show that our multimodal terrain sensing algorithm can achieve a classification accuracy of 99.2\%, a tire damage detection accuracy of 97\%, a 98\% success rate in object search, and the ability to withstand tire loading weights exceeding 35 kg. In addition, we open-source our algorithms, hardware, and datasets at https://sites.google.com/view/vtire.

Optimizing Multi-Round Enhanced Training in Diffusion Models for Improved Preference Understanding

Generative AI has significantly changed industries by enabling text-driven image generation, yet challenges remain in achieving high-resolution outputs that align with fine-grained user preferences. Consequently, multi-round interactions are necessary to ensure the generated images meet expectations. Previous methods enhanced prompts via reward feedback but did not optimize over a multi-round dialogue dataset. In this work, we present a Visual Co-Adaptation (VCA) framework incorporating human-in-the-loop feedback, leveraging a well-trained reward model aligned with human preferences. Using a diverse multi-turn dialogue dataset, our framework applies multiple reward functions, such as diversity, consistency, and preference feedback, while fine-tuning the diffusion model through LoRA, thus optimizing image generation based on user input. We also construct multi-round dialogue datasets of prompts and image pairs aligned with user intent. Experiments demonstrate that our method outperforms state-of-the-art baselines, significantly improving image consistency and alignment with user intent. Our approach consistently surpasses competing models in user satisfaction, especially in multi-turn dialogue scenarios.

Twin Co-Adaptive Dialogue for Progressive Image Generation

Modern text-to-image generation systems have enabled the creation of remarkably realistic and high-quality visuals, yet they often falter when handling the inherent ambiguities in user prompts. In this work, we present Twin-Co, a framework that leverages synchronized, co-adaptive dialogue to progressively refine image generation. Instead of a static generation process, Twin-Co employs a dynamic, iterative workflow where an intelligent dialogue agent continuously interacts with the user. Initially, a base image is generated from the user's prompt. Then, through a series of synchronized dialogue exchanges, the system adapts and optimizes the image according to evolving user feedback. The co-adaptive process allows the system to progressively narrow down ambiguities and better align with user intent. Experiments demonstrate that Twin-Co not only enhances user experience by reducing trial-and-error iterations but also improves the quality of the generated images, streamlining the creative process across various applications.

A Novel Radar Constant False Alarm Rate Detection Algorithm Based on VAMP Deep Unfolding

The combination of deep unfolding with vector approximate message passing (VAMP) algorithm, results in faster convergence and higher sparse recovery accuracy than traditional compressive sensing approaches. However, deep unfolding alters the parameters in traditional VAMP algorithm, resulting in the unattainable distribution parameter of the recovery error of non-sparse noisy estimation via traditional VAMP, which hinders the utilization of VAMP deep unfolding in constant false alarm rate (CFAR) detection in sub-Nyquist radar system. Based on VAMP deep unfolding, we provide a parameter convergence detector (PCD) to estimate the recovery error distribution parameter and implement CFAR detection. Compared to the state-of-the-art approaches, both the sparse solution and non-sparse noisy estimation are utilized to estimate the distribution parameter and implement CFAR detection in PCD, which leverages both the VAMP distribution property and the improved sparse recovery accuracy provided by deep unfolding. Simulation results indicate that PCD offers improved false alarm rate control performance and higher target detection rate.

Parameter Convergence Detector Based on VAMP Deep Unfolding: A Novel Radar Constant False Alarm Rate Detection Algorithm

The sub-Nyquist radar framework exploits the sparsity of signals, which effectively alleviates the pressure on system storage and transmission bandwidth. Compressed sensing (CS) algorithms, such as the VAMP algorithm, are used for sparse signal processing in the sub-Nyquist radar framework. By combining deep unfolding techniques with VAMP, faster convergence and higher accuracy than traditional CS algorithms are achieved. However, deep unfolding disrupts the parameter constrains in traditional VAMP algorithm, leading to the distribution of non-sparse noisy estimation in VAMP deep unfolding unknown, and its distribution parameter unable to be obtained directly using method of traditional VAMP, which prevents the application of VAMP deep unfolding in radar constant false alarm rate (CFAR) detection. To address this problem, we explore the distribution of the non-sparse noisy estimation and propose a parameter convergence detector (PCD) to achieve CFAR detection based on VAMP deep unfolding. Compared to the state-of-the-art methods, PCD leverages not only the sparse solution, but also the non-sparse noisy estimation, which is used to iteratively estimate the distribution parameter and served as the test statistic in detection process. In this way, the proposed algorithm takes advantage of both the enhanced sparse recovery accuracy from deep unfolding and the distribution property of VAMP, thereby achieving superior CFAR detection performance. Additionally, the PCD requires no information about the power of AWGN in the environment, which is more suitable for practical application. The convergence performance and effectiveness of the proposed PCD are analyzed based on the Banach Fixed-Point Theorem. Numerical simulations and practical data experiments demonstrate that PCD can achieve better false alarm control and target detection performance.

MARS: Memory-Enhanced Agents with Reflective Self-improvement

Large language models (LLMs) have made significant advances in the field of natural language processing, but they still face challenges such as continuous decision-making, lack of long-term memory, and limited context windows in dynamic environments. To address these issues, this paper proposes an innovative framework Memory-Enhanced Agents with Reflective Self-improvement. The MARS framework comprises three agents: the User, the Assistant, and the Checker. By integrating iterative feedback, reflective mechanisms, and a memory optimization mechanism based on the Ebbinghaus forgetting curve, it significantly enhances the agents capabilities in handling multi-tasking and long-span information.

PromptLNet: Region-Adaptive Aesthetic Enhancement via Prompt Guidance in Low-Light Enhancement Net

Learning and improving large language models through human preference feedback has become a mainstream approach, but it has rarely been applied to the field of low-light image enhancement. Existing low-light enhancement evaluations typically rely on objective metrics (such as FID, PSNR, etc.), which often result in models that perform well objectively but lack aesthetic quality. Moreover, most low-light enhancement models are primarily designed for global brightening, lacking detailed refinement. Therefore, the generated images often require additional local adjustments, leading to research gaps in practical applications. To bridge this gap, we propose the following innovations: 1) We collect human aesthetic evaluation text pairs and aesthetic scores from multiple low-light image datasets (e.g., LOL, LOL2, LOM, DCIM, MEF, etc.) to train a low-light image aesthetic evaluation model, supplemented by an optimization algorithm designed to fine-tune the diffusion model. 2) We propose a prompt-driven brightness adjustment module capable of performing fine-grained brightness and aesthetic adjustments for specific instances or regions. 3) We evaluate our method alongside existing state-of-the-art algorithms on mainstream benchmarks. Experimental results show that our method not only outperforms traditional methods in terms of visual quality but also provides greater flexibility and controllability, paving the way for improved aesthetic quality.