Yongzhe Chang

Haoyu Wang, Guozheng Ma, Cong Yu et al.

The swift advancement in the scales and capabilities of Large Language Models (LLMs) positions them as promising tools for a variety of downstream tasks. In addition to the pursuit of better performance and the avoidance of violent feedback on a certain prompt, to ensure the responsibility of the LLM, much attention is drawn to the robustness of LLMs. However, existing evaluation methods mostly rely on traditional question answering datasets with predefined supervised labels, which do not align with the superior generation capabilities of contemporary LLMs. To address this issue, we propose a novel rational evaluation approach that leverages pre-trained reward models as diagnostic tools to evaluate the longer conversation generated from more challenging open questions by LLMs, which we refer to as the Reward Model for Reasonable Robustness Evaluation (TREvaL). Longer conversations manifest the comprehensive grasp of language models in terms of their proficiency in understanding questions, a capability not entirely encompassed by individual words or letters, which may exhibit oversimplification and inherent biases. Our extensive empirical experiments demonstrate that TREvaL provides an innovative method for evaluating the robustness of an LLM. Furthermore, our results demonstrate that LLMs frequently exhibit vulnerability to word-level perturbations that are commonplace in daily language usage. Notably, we are surprised to discover that robustness tends to decrease as fine-tuning (SFT and RLHF) is conducted. The code of TREval is available in https://github.com/Harry-mic/TREvaL.

Yilun Kong, Hangyu Mao, Qi Zhao et al.

Prompt engineering has demonstrated remarkable success in enhancing the performance of large language models (LLMs) across diverse tasks. However, most existing prompt optimization methods only focus on the task-level performance, overlooking the importance of query-preferred prompts, which leads to suboptimal performances. Additionally, these methods rely heavily on frequent interactions with LLMs to obtain feedback for guiding the optimization process, incurring substantial redundant interaction costs. In this paper, we introduce Query-dependent Prompt Optimization (QPO), which leverages multi-loop offline reinforcement learning to iteratively fine-tune a small pretrained language model to generate optimal prompts tailored to the input queries, thus significantly improving the prompting effect on the large target LLM. We derive insights from offline prompting demonstration data, which already exists in large quantities as a by-product of benchmarking diverse prompts on open-sourced tasks, thereby circumventing the expenses of online interactions. Furthermore, we continuously augment the offline dataset with the generated prompts in each loop, as the prompts from the fine-tuned model are supposed to outperform the source prompts in the original dataset. These iterative loops bootstrap the model towards generating optimal prompts. Experiments on various LLM scales and diverse NLP and math tasks demonstrate the efficacy and cost-efficiency of our method in both zero-shot and few-shot scenarios.

Qin Zhang, Linrui Zhang, Haoran Xu et al.

Offline safe RL is of great practical relevance for deploying agents in real-world applications. However, acquiring constraint-satisfying policies from the fixed dataset is non-trivial for conventional approaches. Even worse, the learned constraints are stationary and may become invalid when the online safety requirement changes. In this paper, we present a novel offline safe RL approach referred to as SaFormer, which tackles the above issues via conditional sequence modeling. In contrast to existing sequence models, we propose cost-related tokens to restrict the action space and a posterior safety verification to enforce the constraint explicitly. Specifically, SaFormer performs a two-stage auto-regression conditioned by the maximum remaining cost to generate feasible candidates. It then filters out unsafe attempts and executes the optimal action with the highest expected return. Extensive experiments demonstrate the efficacy of SaFormer featuring (1) competitive returns with tightened constraint satisfaction; (2) adaptability to the in-range cost values of the offline data without retraining; (3) generalizability for constraints beyond the current dataset.

Yuxing Wang, Zhiyu Chen, Tiantian Zhang et al.

Brain-body co-evolution enables animals to develop complex behaviors in their environments. Inspired by this biological synergy, embodied co-design (ECD) has emerged as a transformative paradigm for creating intelligent agents-from virtual creatures to physical robots-by jointly optimizing their morphologies and controllers rather than treating control in isolation. This integrated approach facilitates richer environmental interactions and robust task performance. In this survey, we provide a systematic overview of recent advances in ECD. We first formalize the concept of ECD and position it within related fields. We then introduce a hierarchical taxonomy: a lower layer that breaks down agent design into three fundamental components-controlling brain, body morphology, and task environment-and an upper layer that integrates these components into four major ECD frameworks: bi-level, single-level, generative, and open-ended. This taxonomy allows us to synthesize insights from more than one hundred recent studies. We further review notable benchmarks, datasets, and applications in both simulated and real-world scenarios. Finally, we identify significant challenges and offer insights into promising future research directions. A project associated with this survey has been created at https://github.com/Yuxing-Wang-THU/SurveyBrainBody.

Jinghui Xu, Boyi Shangguan, Mengke Zhu et al.

Autonomous Earth Observation (EO) agents are transitioning from passive perception to complex, multi-step task execution. However, current architectures that integrate planning and execution within a single model often struggle with combinatorial complexity and reasoning errors in dynamic EO scenarios. To resolve these challenges, we propose the Lightweight Multimodal Meta-Planner (LMMP) framework. LMMP incorporates a dual-awareness mechanism that grounds strategic plans in both multimodal image features and high-level task semantics. Crucially, we introduce a Meta Task Library to inject remote sensing expert knowledge directly into the workflow, which standardizes domain logic and ensures plans are physically feasible. We further implement a two-stage training pipeline, initializing the Meta-Planner via expert-distilled Supervised Fine-Tuning and refining it through Direct Preference Optimization based on execution feedback. Extensive experiments on a dataset derived from EarthBench and ThinkGeo demonstrate that LMMP significantly improves tool-calling accuracy and task success rates. Moreover, the framework exhibits strong ``plug-and-play'' versatility, consistently enhancing the performance of diverse executor backbones across previously unseen EO missions.

Haoyuan Sun, Bo Xia, Yongzhe Chang et al.

Direct Preference Optimization (DPO) has recently expanded its successful application from aligning large language models (LLMs) to aligning text-to-image models with human preferences, which has generated considerable interest within the community. However, we have observed that these approaches rely solely on minimizing the reverse Kullback-Leibler divergence during alignment process between the fine-tuned model and the reference model, neglecting the incorporation of other divergence constraints. In this study, we focus on extending reverse Kullback-Leibler divergence in the alignment paradigm of text-to-image models to $f$-divergence, which aims to garner better alignment performance as well as good generation diversity. We provide the generalized formula of the alignment paradigm under the $f$-divergence condition and thoroughly analyze the impact of different divergence constraints on alignment process from the perspective of gradient fields. We conduct comprehensive evaluation on image-text alignment performance, human value alignment performance and generation diversity performance under different divergence constraints, and the results indicate that alignment based on Jensen-Shannon divergence achieves the best trade-off among them. The option of divergence employed for aligning text-to-image models significantly impacts the trade-off between alignment performance (especially human value alignment) and generation diversity, which highlights the necessity of selecting an appropriate divergence for practical applications.

Haoyu Wang, Bingzhe Wu, Yatao Bian et al.

Large Language Models (LLMs) are implicit troublemakers. While they provide valuable insights and assist in problem-solving, they can also potentially serve as a resource for malicious activities. Implementing safety alignment could mitigate the risk of LLMs generating harmful responses. We argue that: even when an LLM appears to successfully block harmful queries, there may still be hidden vulnerabilities that could act as ticking time bombs. To identify these underlying weaknesses, we propose to use a cost value model as both a detector and an attacker. Trained on external or self-generated harmful datasets, the cost value model could successfully influence the original safe LLM to output toxic content in decoding process. For instance, LLaMA-2-chat 7B outputs 39.18% concrete toxic content, along with only 22.16% refusals without any harmful suffixes. These potential weaknesses can then be exploited via prompt optimization such as soft prompts on images. We name this decoding strategy: Jailbreak Value Decoding (JVD), emphasizing that seemingly secure LLMs may not be as safe as we initially believe. They could be used to gather harmful data or launch covert attacks.

Haoyuan Sun, Jiaqi Wu, Bo Xia et al.

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.

Yuxing Wang, Jie Li, Cong Yu et al.

The emergence of modular satellites marks a significant transformation in spacecraft engineering, introducing a new paradigm of flexibility, resilience, and scalability in space exploration endeavors. In addressing complex challenges such as attitude control, both the satellite's morphological architecture and the controller are crucial for optimizing performance. Despite substantial research on optimal control, there remains a significant gap in developing optimized and practical assembly strategies for modular satellites tailored to specific mission constraints. This research gap primarily arises from the inherently complex nature of co-optimizing design and control, a process known for its notorious bi-level optimization loop. Conventionally tackled through artificial evolution, this issue involves optimizing the morphology based on the fitness of individual controllers, which is sample-inefficient and computationally expensive. In this paper, we introduce a novel gradient-based approach to simultaneously optimize both morphology and control for modular satellites, enhancing their performance and efficiency in attitude control missions. Our Monte Carlo simulations demonstrate that this co-optimization approach results in modular satellites with better mission performance compared to those designed by evolution-based approaches. Furthermore, this study discusses potential avenues for future research.

Yifu Luo, Xinhao Hu, Keyu Fan et al.

Reinforcement learning (RL) has garnered increasing attention in text-to-image (T2I) generation. However, most existing RL approaches are tailored to either diffusion models or autoregressive models, overlooking an important alternative: masked generative models. In this work, we propose Mask-GRPO, the first method to incorporate Group Relative Policy Optimization (GRPO)-based RL into this overlooked paradigm. Our core insight is to redefine the transition probability, which is different from current approaches, and formulate the unmasking process as a multi-step decision-making problem. To further enhance our method, we explore several useful strategies, including removing the KL constraint, applying the reduction strategy, and filtering out low-quality samples. Using Mask-GRPO, we improve a base model, Show-o, with substantial improvements on standard T2I benchmarks and preference alignment, outperforming existing state-of-the-art approaches. The code is available on https://github.com/xingzhejun/Mask-GRPO

Guoxin Shi, Haoyu Wang, Zaihui Yang et al.

Adversarial behavior plays a central role in aligning large language models with human values. However, existing alignment methods largely rely on static adversarial settings, which fundamentally limit robustness, particularly in multimodal settings with a larger attack surface. In this work, we move beyond static adversarial supervision and introduce co-evolutionary alignment with evolving attacks, instantiated by CEMMA (Co-Evolutionary Multi-Modal Alignment), an automated and adaptive framework for multimodal safety alignment. We introduce an Evolutionary Attacker that decomposes adversarial prompts into method templates and harmful intents. By employing genetic operators, including mutation, crossover, and differential evolution, it enables simple seed attacks to inherit the structural efficacy of sophisticated jailbreaks. The Adaptive Defender is iteratively updated on the synthesized hard negatives, forming a closed-loop process that adapts alignment to evolving attacks. Experiments show that the Evolutionary Attacker substantially increases red-teaming jailbreak attack success rate (ASR), while the Adaptive Defender improves robustness and generalization across benchmarks with higher data efficiency, without inducing excessive benign refusal, and remains compatible with inference-time defenses such as AdaShield.

Haoyuan Sun, Zihao Wu, Bo Xia et al.

The success of artificial neural networks (ANNs) hinges greatly on the judicious selection of an activation function, introducing non-linearity into network and enabling them to model sophisticated relationships in data. However, the search of activation functions has largely relied on empirical knowledge in the past, lacking theoretical guidance, which has hindered the identification of more effective activation functions. In this work, we offer a proper solution to such issue. Firstly, we theoretically demonstrate the existence of the worst activation function with boundary conditions (WAFBC) from the perspective of information entropy. Furthermore, inspired by the Taylor expansion form of information entropy functional, we propose the Entropy-based Activation Function Optimization (EAFO) methodology. EAFO methodology presents a novel perspective for designing static activation functions in deep neural networks and the potential of dynamically optimizing activation during iterative training. Utilizing EAFO methodology, we derive a novel activation function from ReLU, known as Correction Regularized ReLU (CRReLU). Experiments conducted with vision transformer and its variants on CIFAR-10, CIFAR-100 and ImageNet-1K datasets demonstrate the superiority of CRReLU over existing corrections of ReLU. Extensive empirical studies on task of large language model (LLM) fine-tuning, CRReLU exhibits superior performance compared to GELU, suggesting its broader potential for practical applications.

Yifu Luo, Yongzhe Chang, Xueqian Wang

Diffusion probability models have shown significant promise in offline reinforcement learning by directly modeling trajectory sequences. However, existing approaches primarily focus on time-domain features while overlooking frequency-domain features, leading to frequency shift and degraded performance according to our observation. In this paper, we investigate the RL problem from a new perspective of the frequency domain. We first observe that time-domain-only approaches inadvertently introduce shifts in the low-frequency components of the frequency domain, which results in trajectory instability and degraded performance. To address this issue, we propose Wavelet Fourier Diffuser (WFDiffuser), a novel diffusion-based RL framework that integrates Discrete Wavelet Transform to decompose trajectories into low- and high-frequency components. To further enhance diffusion modeling for each component, WFDiffuser employs Short-Time Fourier Transform and cross attention mechanisms to extract frequency-domain features and facilitate cross-frequency interaction. Extensive experiment results on the D4RL benchmark demonstrate that WFDiffuser effectively mitigates frequency shift, leading to smoother, more stable trajectories and improved decision-making performance over existing methods.

Jiaxi Wu, Tiantian Zhang, Yuxing Wang et al.

Robust adversarial reinforcement learning has emerged as an effective paradigm for training agents to handle uncertain disturbance in real environments, with critical applications in sequential decision-making domains such as autonomous driving and robotic control. Within this paradigm, agent training is typically formulated as a zero-sum Markov game between a protagonist and an adversary to enhance policy robustness. However, the trainable nature of the adversary inevitably induces non-stationarity in the learning dynamics, leading to exacerbated training instability and convergence difficulties, particularly in high-dimensional complex environments. In this paper, we propose a novel approach, Uncertainty-Adaptive Critic Ensemble for robust adversarial Reinforcement learning (UACER), which consists of two components: 1) Diversified critic ensemble: A diverse set of K critic networks is employed in parallel to stabilize Q-value estimation in robust adversarial reinforcement learning, reducing variance and enhancing robustness compared to conventional single-critic designs. 2) Time-varying Decay Uncertainty (TDU) mechanism: Moving beyond simple linear combinations, we propose a variance-derived Q-value aggregation strategy that explicitly incorporates epistemic uncertainty to adaptively regulate the exploration-exploitation trade-off while stabilizing the training process. Comprehensive experiments across several challenging MuJoCo control problems validate the superior effectiveness of UACER, outperforming state-of-the-art methods in terms of overall performance, stability, and efficiency.

Yifu Luo, Penghui Du, Bo Li et al.

Group Relative Policy Optimization (GRPO) has shown strong potential for flow-matching-based text-to-image (T2I) generation, but it faces two key limitations: inaccurate advantage attribution, and the neglect of temporal dynamics of generation. In this work, we argue that shifting the optimization paradigm from the step level to the chunk level can effectively alleviate these issues. Building on this idea, we propose Chunk-GRPO, the first chunk-level GRPO-based approach for T2I generation. The insight is to group consecutive steps into coherent 'chunk's that capture the intrinsic temporal dynamics of flow matching, and to optimize policies at the chunk level. In addition, we introduce an optional weighted sampling strategy to further enhance performance. Extensive experiments show that ChunkGRPO achieves superior results in both preference alignment and image quality, highlighting the promise of chunk-level optimization for GRPO-based methods.

Kai Qin, Jiaqi Wu, Jianxiang He et al.

As Large Language Models (LLMs) demonstrate remarkable capabilities learned from vast corpora, concerns regarding data privacy and safety are receiving increasing attention. LLM unlearning, which aims to remove the influence of specific data while preserving overall model utility, is becoming an important research area. One of the mainstream unlearning classes is optimization-based methods, which achieve forgetting directly through fine-tuning, exemplified by Negative Preference Optimization (NPO). However, NPO's effectiveness is limited by its inherent lack of explicit positive preference signals. Attempts to introduce such signals by constructing preferred responses often necessitate domain-specific knowledge or well-designed prompts, fundamentally restricting their generalizability. In this paper, we shift the focus to the distribution-level, directly targeting the next-token probability distribution instead of entire responses, and derive a novel unlearning algorithm termed \textbf{Di}stribution \textbf{P}reference \textbf{O}ptimization (DiPO). We show that the requisite preference distribution pairs for DiPO, which are distributions over the model's output tokens, can be constructed by selectively amplifying or suppressing the model's high-confidence output logits, thereby effectively overcoming NPO's limitations. We theoretically prove the consistency of DiPO's loss function with the desired unlearning direction. Extensive experiments demonstrate that DiPO achieves a strong trade-off between model utility and forget quality. Notably, DiPO attains the highest forget quality on the TOFU benchmark, and maintains leading scalability and sustainability in utility preservation on the MUSE benchmark.

Bo Xia, Yilun Kong, Yongzhe Chang et al.

Classic reinforcement learning (RL) frequently confronts challenges in tasks involving delays, which cause a mismatch between received observations and subsequent actions, thereby deviating from the Markov assumption. Existing methods usually tackle this issue with end-to-end solutions using state augmentation. However, these black-box approaches often involve incomprehensible processes and redundant information in the information states, causing instability and potentially undermining the overall performance. To alleviate the delay challenges in RL, we propose $\textbf{DEER (Delay-resilient Encoder-Enhanced RL)}$, a framework designed to effectively enhance the interpretability and address the random delay issues. DEER employs a pretrained encoder to map delayed states, along with their variable-length past action sequences resulting from different delays, into hidden states, which is trained on delay-free environment datasets. In a variety of delayed scenarios, the trained encoder can seamlessly integrate with standard RL algorithms without requiring additional modifications and enhance the delay-solving capability by simply adapting the input dimension of the original algorithms. We evaluate DEER through extensive experiments on Gym and Mujoco environments. The results confirm that DEER is superior to state-of-the-art RL algorithms in both constant and random delay settings.

Yuxing Wang, Tiantian Zhang, Yongzhe Chang et al.

The integration of Reinforcement Learning (RL) and Evolutionary Algorithms (EAs) aims at simultaneously exploiting the sample efficiency as well as the diversity and robustness of the two paradigms. Recently, hybrid learning frameworks based on this principle have achieved great success in various challenging robot control tasks. However, in these methods, policies from the genetic population are evaluated via interactions with the real environments, limiting their applicability in computationally expensive problems. In this work, we propose Surrogate-assisted Controller (SC), a novel and efficient module that can be integrated into existing frameworks to alleviate the computational burden of EAs by partially replacing the expensive policy evaluation. The key challenge in applying this module is to prevent the optimization process from being misled by the possible false minima introduced by the surrogate. To address this issue, we present two strategies for SC to control the workflow of hybrid frameworks. Experiments on six continuous control tasks from the OpenAI Gym platform show that SC can not only significantly reduce the cost of fitness evaluations, but also boost the performance of the original hybrid frameworks with collaborative learning and evolutionary processes.

Yang Liu, Yongzhe Chang, Shilei Jiang et al.

Imitation Learning (IL) is an effective learning paradigm exploiting the interactions between agents and environments. It does not require explicit reward signals and instead tries to recover desired policies using expert demonstrations. In general, IL methods can be categorized into Behavioral Cloning (BC) and Inverse Reinforcement Learning (IRL). In this work, a novel reward function based on probability density estimation is proposed for IRL, which can significantly reduce the complexity of existing IRL methods. Furthermore, we prove that the theoretically optimal policy derived from our reward function is identical to the expert policy as long as it is deterministic. Consequently, an IRL problem can be gracefully transformed into a probability density estimation problem. Based on the proposed reward function, we present a "watch-try-learn" style framework named Probability Density Estimation based Imitation Learning (PDEIL), which can work in both discrete and continuous action spaces. Finally, comprehensive experiments in the Gym environment show that PDEIL is much more efficient than existing algorithms in recovering rewards close to the ground truth.

Liang Xu, Liying Zheng, Weijun Li et al.

In recent studies, Lots of work has been done to solve time series anomaly detection by applying Variational Auto-Encoders (VAEs). Time series anomaly detection is a very common but challenging task in many industries, which plays an important role in network monitoring, facility maintenance, information security, and so on. However, it is very difficult to detect anomalies in time series with high accuracy, due to noisy data collected from real world, and complicated abnormal patterns. From recent studies, we are inspired by Nouveau VAE (NVAE) and propose our anomaly detection model: Time series to Image VAE (T2IVAE), an unsupervised model based on NVAE for univariate series, transforming 1D time series to 2D image as input, and adopting the reconstruction error to detect anomalies. Besides, we also apply the Generative Adversarial Networks based techniques to T2IVAE training strategy, aiming to reduce the overfitting. We evaluate our model performance on three datasets, and compare it with other several popular models using F1 score. T2IVAE achieves 0.639 on Numenta Anomaly Benchmark, 0.651 on public dataset from NASA, and 0.504 on our dataset collected from real-world scenario, outperforms other comparison models.