Jiaxin Gao

Jingliang Duan, Wenxuan Wang, Liming Xiao et al.

Reinforcement learning (RL) has shown remarkable success in solving complex decision-making and control tasks. However, many model-free RL algorithms experience performance degradation due to inaccurate value estimation, particularly the overestimation of Q-values, which can lead to suboptimal policies. To address this issue, we previously proposed the Distributional Soft Actor-Critic (DSAC or DSACv1), an off-policy RL algorithm that enhances value estimation accuracy by learning a continuous Gaussian value distribution. Despite its effectiveness, DSACv1 faces challenges such as training instability and sensitivity to reward scaling, caused by high variance in critic gradients due to return randomness. In this paper, we introduce three key refinements to DSACv1 to overcome these limitations and further improve Q-value estimation accuracy: expected value substitution, twin value distribution learning, and variance-based critic gradient adjustment. The enhanced algorithm, termed DSAC with Three refinements (DSAC-T or DSACv2), is systematically evaluated across a diverse set of benchmark tasks. Without the need for task-specific hyperparameter tuning, DSAC-T consistently matches or outperforms leading model-free RL algorithms, including SAC, TD3, DDPG, TRPO, and PPO, in all tested environments. Additionally, DSAC-T ensures a stable learning process and maintains robust performance across varying reward scales. Its effectiveness is further demonstrated through real-world application in controlling a wheeled robot, highlighting its potential for deployment in practical robotic tasks.

Risheng Liu, Jiaxin Gao, Xuan Liu et al.

The complexity of learning problems, such as Generative Adversarial Network (GAN) and its variants, multi-task and meta-learning, hyper-parameter learning, and a variety of real-world vision applications, demands a deeper understanding of their underlying coupling mechanisms. Existing approaches often address these problems in isolation, lacking a unified perspective that can reveal commonalities and enable effective solutions. Therefore, in this work, we proposed a new framework, named Learning with Constraint Learning (LwCL), that can holistically examine challenges and provide a unified methodology to tackle all the above-mentioned complex learning and vision problems. Specifically, LwCL is designed as a general hierarchical optimization model that captures the essence of these diverse learning and vision problems. Furthermore, we develop a gradient-response based fast solution strategy to overcome optimization challenges of the LwCL framework. Our proposed framework efficiently addresses a wide range of applications in learning and vision, encompassing three categories and nine different problem types. Extensive experiments on synthetic tasks and real-world applications verify the effectiveness of our approach. The LwCL framework offers a comprehensive solution for tackling complex machine learning and computer vision problems, bridging the gap between theory and practice.

Yaohua Liu, Jiaxin Gao, Xianghao Jiao et al.

Adversarial Training (AT), pivotal in fortifying the robustness of deep learning models, is extensively adopted in practical applications. However, prevailing AT methods, relying on direct iterative updates for target model's defense, frequently encounter obstacles such as unstable training and catastrophic overfitting. In this context, our work illuminates the potential of leveraging the target model's historical states as a proxy to provide effective initialization and defense prior, which results in a general proxy guided defense framework, `LAST' ({\bf L}earn from the P{\bf ast}). Specifically, LAST derives response of the proxy model as dynamically learned fast weights, which continuously corrects the update direction of the target model. Besides, we introduce a self-distillation regularized defense objective, ingeniously designed to steer the proxy model's update trajectory without resorting to external teacher models, thereby ameliorating the impact of catastrophic overfitting on performance. Extensive experiments and ablation studies showcase the framework's efficacy in markedly improving model robustness (e.g., up to 9.2\% and 20.3\% enhancement in robust accuracy on CIFAR10 and CIFAR100 datasets, respectively) and training stability. These improvements are consistently observed across various model architectures, larger datasets, perturbation sizes, and attack modalities, affirming LAST's ability to consistently refine both single-step and multi-step AT strategies. The code will be available at~\url{https://github.com/callous-youth/LAST}.

Risheng Liu, Jiaxin Gao, Xuan Liu et al.

In past years, the minimax type single-level optimization formulation and its variations have been widely utilized to address Generative Adversarial Networks (GANs). Unfortunately, it has been proved that these alternating learning strategies cannot exactly reveal the intrinsic relationship between the generator and discriminator, thus easily result in a series of issues, including mode collapse, vanishing gradients and oscillations in the training phase, etc. In this work, by investigating the fundamental mechanism of GANs from the perspective of hierarchical optimization, we propose Best-Response Constraint (BRC), a general learning framework, that can explicitly formulate the potential dependency of the generator on the discriminator. Rather than adopting these existing time-consuming bilevel iterations, we design an implicit gradient scheme with outer-product Hessian approximation as our fast solution strategy. \emph{Noteworthy, we demonstrate that even with different motivations and formulations, a variety of existing GANs ALL can be uniformly improved by our flexible BRC methodology.} Extensive quantitative and qualitative experimental results verify the effectiveness, flexibility and stability of our proposed framework.

Jiaxin Gao, Ziyu Yue, Yaohua Liu et al.

Super-resolution tasks oriented to images captured in ultra-dark environments is a practical yet challenging problem that has received little attention. Due to uneven illumination and low signal-to-noise ratio in dark environments, a multitude of problems such as lack of detail and color distortion may be magnified in the super-resolution process compared to normal-lighting environments. Consequently, conventional low-light enhancement or super-resolution methods, whether applied individually or in a cascaded manner for such problem, often encounter limitations in recovering luminance, color fidelity, and intricate details. To conquer these issues, this paper proposes a specialized dual-modulated learning framework that, for the first time, attempts to deeply dissect the nature of the low-light super-resolution task. Leveraging natural image color characteristics, we introduce a self-regularized luminance constraint as a prior for addressing uneven lighting. Expanding on this, we develop Illuminance-Semantic Dual Modulation (ISDM) components to enhance feature-level preservation of illumination and color details. Besides, instead of deploying naive up-sampling strategies, we design the Resolution-Sensitive Merging Up-sampler (RSMU) module that brings together different sampling modalities as substrates, effectively mitigating the presence of artifacts and halos. Comprehensive experiments showcases the applicability and generalizability of our approach to diverse and challenging ultra-low-light conditions, outperforming state-of-the-art methods with a notable improvement (i.e., $\uparrow$5\% in PSNR, and $\uparrow$43\% in LPIPS). Especially noteworthy is the 19-fold increase in the RMSE score, underscoring our method's exceptional generalization across different darkness levels. The code will be available online upon publication of the paper.

Jiaxin Gao, Wenbo Hu, Dongxiao Zhang et al.

Electrical energy is essential in today's society. Accurate electrical load forecasting is beneficial for better scheduling of electricity generation and saving electrical energy. In this paper, we propose theory-guided deep-learning load forecasting 2.0 (TgDLF2.0) to solve this issue, which is an improved version of the theory-guided deep-learning framework for load forecasting via ensemble long short-term memory (TgDLF). TgDLF2.0 introduces the deep-learning model Transformer and transfer learning on the basis of dividing the electrical load into dimensionless trends and local fluctuations, which realizes the utilization of domain knowledge, captures the long-term dependency of the load series, and is more appropriate for realistic scenarios with scarce samples. Cross-validation experiments on different districts show that TgDLF2.0 is approximately 16% more accurate than TgDLF and saves more than half of the training time. TgDLF2.0 with 50% weather noise has the same accuracy as TgDLF without noise, which proves its robustness. We also preliminarily mine the interpretability of Transformer in TgDLF2.0, which may provide future potential for better theory guidance. Furthermore, experiments demonstrate that transfer learning can accelerate convergence of the model in half the number of training epochs and achieve better performance.

Yaohua Liu, Jiaxin Gao, Xuan Liu et al.

Transfer attacks generate significant interest for real-world black-box applications by crafting transferable adversarial examples through surrogate models. Whereas, existing works essentially directly optimize the single-level objective w.r.t. the surrogate model, which always leads to poor interpretability of attack mechanism and limited generalization performance over unknown victim models. In this work, we propose the \textbf{B}il\textbf{E}vel \textbf{T}ransfer \textbf{A}ttac\textbf{K} (BETAK) framework by establishing an initialization derived bilevel optimization paradigm, which explicitly reformulates the nested constraint relationship between the Upper-Level (UL) pseudo-victim attacker and the Lower-Level (LL) surrogate attacker. Algorithmically, we introduce the Hyper Gradient Response (HGR) estimation as an effective feedback for the transferability over pseudo-victim attackers, and propose the Dynamic Sequence Truncation (DST) technique to dynamically adjust the back-propagation path for HGR and reduce computational overhead simultaneously. Meanwhile, we conduct detailed algorithmic analysis and provide convergence guarantee to support non-convexity of the LL surrogate attacker. Extensive evaluations demonstrate substantial improvement of BETAK (e.g., $\mathbf{53.41}$\% increase of attack success rates against IncRes-v$2_{ens}$) against different victims and defense methods in targeted and untargeted attack scenarios. The source code is available at https://github.com/callous-youth/BETAK.

Jiaxin Gao, Wenbo Hu, Yuntian Chen

Long-term time series forecasting (LTSF) is a crucial aspect of modern society, playing a pivotal role in facilitating long-term planning and developing early warning systems. While many Transformer-based models have recently been introduced for LTSF, a doubt have been raised regarding the effectiveness of attention modules in capturing cross-time dependencies. In this study, we design a mask-series experiment to validate this assumption and subsequently propose the "Cross-variable Linear Integrated ENhanced Transformer for Multivariate Long-Term Time Series Forecasting" (Client), an advanced model that outperforms both traditional Transformer-based models and linear models. Client employs linear modules to learn trend information and attention modules to capture cross-variable dependencies. Meanwhile, it simplifies the embedding and position encoding layers and replaces the decoder module with a projection layer. Essentially, Client incorporates non-linearity and cross-variable dependencies, which sets it apart from conventional linear models and Transformer-based models. Extensive experiments with nine real-world datasets have confirmed the SOTA performance of Client with the least computation time and memory consumption compared with the previous Transformer-based models. Our code is available at https://github.com/daxin007/Client.

Xiaofeng Liu, Jiaxin Gao, Xin Fan et al.

Contemporary Low-Light Image Enhancement (LLIE) techniques have made notable advancements in preserving image details and enhancing contrast, achieving commendable results on specific datasets. Nevertheless, these approaches encounter persistent challenges in efficiently mitigating dynamic noise and accommodating diverse low-light scenarios. Insufficient constraints on complex pixel-wise mapping learning lead to overfitting to specific types of noise and artifacts associated with low-light conditions, reducing effectiveness in variable lighting scenarios. To this end, we first propose a method for estimating the noise level in low light images in a quick and accurate way. This facilitates precise denoising, prevents over-smoothing, and adapts to dynamic noise patterns. Subsequently, we devise a Learnable Illumination Interpolator (LII), which employs learnlable interpolation operations between the input and unit vector to satisfy general constraints between illumination and input. Finally, we introduce a self-regularization loss that incorporates intrinsic image properties and essential visual attributes to guide the output towards meeting human visual expectations. Comprehensive experiments validate the competitiveness of our proposed algorithm in both qualitative and quantitative assessments. Notably, our noise estimation method, with linear time complexity and suitable for various denoisers, significantly improves both denoising and enhancement performance. Benefiting from this, our approach achieves a 0.675dB PSNR improvement on the LOL dataset and 0.818dB on the MIT dataset on LLIE task, even compared to supervised methods. The source code is available at \href{https://doi.org/10.5281/zenodo.11463142}{this DOI repository} and the specific code for noise estimation can be found at \href{https://github.com/GoogolplexGoodenough/noise_estimate}{this separate GitHub link}.

Weike Peng, Jiaxin Gao, Yuntian Chen et al.

Machine learning algorithms emerge as a promising approach in energy fields, but its practical is hindered by data barriers, stemming from high collection costs and privacy concerns. This study introduces a novel federated learning (FL) framework based on XGBoost models, enabling safe collaborative modeling with accessible yet concealed data from multiple parties. Hyperparameter tuning of the models is achieved through Bayesian Optimization. To ascertain the merits of the proposed FL-XGBoost method, a comparative analysis is conducted between separate and centralized models to address a classical binary classification problem in geoenergy sector. The results reveal that the proposed FL framework strikes an optimal balance between privacy and accuracy. FL models demonstrate superior accuracy and generalization capabilities compared to separate models, particularly for participants with limited data or low correlation features and offers significant privacy benefits compared to centralized model. The aggregated optimization approach within the FL agreement proves effective in tuning hyperparameters. This study opens new avenues for assessing unconventional reservoirs through collaborative and privacy-preserving FL techniques.

Chen Chen, Yuchen Sun, Jiaxin Gao et al.

Large language models (LLMs) have seen significant advancements, achieving superior performance in various Natural Language Processing (NLP) tasks. However, they remain vulnerable to backdoor attacks, where models behave normally for standard queries but generate harmful responses or unintended output when specific triggers are activated. Existing backdoor defenses either lack comprehensiveness, focusing on narrow trigger settings, detection-only mechanisms, and limited domains, or fail to withstand advanced scenarios like model-editing-based, multi-trigger, and triggerless attacks. In this paper, we present LETHE, a novel method to eliminate backdoor behaviors from LLMs through knowledge dilution using both internal and external mechanisms. Internally, LETHE leverages a lightweight dataset to train a clean model, which is then merged with the backdoored model to neutralize malicious behaviors by diluting the backdoor impact within the model's parametric memory. Externally, LETHE incorporates benign and semantically relevant evidence into the prompt to distract LLM's attention from backdoor features. Experimental results on classification and generation domains across 5 widely used LLMs demonstrate that LETHE outperforms 8 state-of-the-art defense baselines against 8 backdoor attacks. LETHE reduces the attack success rate of advanced backdoor attacks by up to 98% while maintaining model utility. Furthermore, LETHE has proven to be cost-efficient and robust against adaptive backdoor attacks.

Jiaxin Gao, Yuxiao Hu, Qinglong Cao et al.

Time series forecasting (TSF) holds significant importance in modern society, spanning numerous domains. Previous representation learning-based TSF algorithms typically embrace a contrastive learning paradigm featuring segregated trend-periodicity representations. Yet, these methodologies disregard the inherent high-impact noise embedded within time series data, resulting in representation inaccuracies and seriously demoting the forecasting performance. To address this issue, we propose CLeaRForecast, a novel contrastive learning framework to learn high-purity time series representations with proposed sample, feature, and architecture purifying methods. More specifically, to avoid more noise adding caused by the transformations of original samples (series), transformations are respectively applied for trendy and periodic parts to provide better positive samples with obviously less noise. Moreover, we introduce a channel independent training manner to mitigate noise originating from unrelated variables in the multivariate series. By employing a streamlined deep-learning backbone and a comprehensive global contrastive loss function, we prevent noise introduction due to redundant or uneven learning of periodicity and trend. Experimental results show the superior performance of CLeaRForecast in various downstream TSF tasks.

Xianghao Jiao, Yaohua Liu, Jiaxin Gao et al.

In light of the significant progress made in the development and application of semantic segmentation tasks, there has been increasing attention towards improving the robustness of segmentation models against natural degradation factors (e.g., rain streaks) or artificially attack factors (e.g., adversarial attack). Whereas, most existing methods are designed to address a single degradation factor and are tailored to specific application scenarios. In this work, we present the first attempt to improve the robustness of semantic segmentation tasks by simultaneously handling different types of degradation factors. Specifically, we introduce the Preprocessing Enhanced Adversarial Robust Learning (PEARL) framework based on the analysis of our proposed Naive Adversarial Training (NAT) framework. Our approach effectively handles both rain streaks and adversarial perturbation by transferring the robustness of the segmentation model to the image derain model. Furthermore, as opposed to the commonly used Negative Adversarial Attack (NAA), we design the Auxiliary Mirror Attack (AMA) to introduce positive information prior to the training of the PEARL framework, which improves defense capability and segmentation performance. Our extensive experiments and ablation studies based on different derain methods and segmentation models have demonstrated the significant performance improvement of PEARL with AMA in defense against various adversarial attacks and rain streaks while maintaining high generalization performance across different datasets.

Xiaofeng Liu, Jiaxin Gao, Yaohua Liu et al.

Recently significant progress has been made in human action recognition and behavior prediction using deep learning techniques, leading to improved vision-based semantic understanding. However, there is still a lack of high-quality motion datasets for small bio-robotics, which presents more challenging scenarios for long-term movement prediction and behavior control based on third-person observation. In this study, we introduce RatPose, a bio-robot motion prediction dataset constructed by considering the influence factors of individuals and environments based on predefined annotation rules. To enhance the robustness of motion prediction against these factors, we propose a Dual-stream Motion-Scenario Decoupling (\textit{DMSD}) framework that effectively separates scenario-oriented and motion-oriented features and designs a scenario contrast loss and motion clustering loss for overall training. With such distinctive architecture, the dual-branch feature flow information is interacted and compensated in a decomposition-then-fusion manner. Moreover, we demonstrate significant performance improvements of the proposed \textit{DMSD} framework on different difficulty-level tasks. We also implement long-term discretized trajectory prediction tasks to verify the generalization ability of the proposed dataset.

Risheng Liu, Jiaxin Gao, Jin Zhang et al.

Bi-Level Optimization (BLO) is originated from the area of economic game theory and then introduced into the optimization community. BLO is able to handle problems with a hierarchical structure, involving two levels of optimization tasks, where one task is nested inside the other. In machine learning and computer vision fields, despite the different motivations and mechanisms, a lot of complex problems, such as hyper-parameter optimization, multi-task and meta-learning, neural architecture search, adversarial learning and deep reinforcement learning, actually all contain a series of closely related subproblms. In this paper, we first uniformly express these complex learning and vision problems from the perspective of BLO. Then we construct a best-response-based single-level reformulation and establish a unified algorithmic framework to understand and formulate mainstream gradient-based BLO methodologies, covering aspects ranging from fundamental automatic differentiation schemes to various accelerations, simplifications, extensions and their convergence and complexity properties. Last but not least, we discuss the potentials of our unified BLO framework for designing new algorithms and point out some promising directions for future research.