arXiv subjects

Xuewen Zhang

Publications and source records attributed to Xuewen Zhang.

17 recordsLinked to original sources

arXiv2026

ARKD: Adaptive Reinforcement Learning-Guided Bidirectional KL Divergence Distillation for Text Generation

Knowledge distillation (KD) is a key technique for compressing Large Language Models (LLMs), yet methods relying on a single KL objective often fail to balance primary distribution fitting with long-tail probability modeling, limiting both generation quality and generalization. To address this, we analyze the complementary roles of forward and reverse KL divergence (FKL/RKL) in distribution alignment from theoretical and empirical perspectives. We then propose a reinforcement-learning-based adaptive KL-weighted distillation framework, in which a policy network dynamically assigns weights to FKL and RKL based on teacher-student distributional characteristics, guided by immediate reward signals to achieve dual alignment on principal and long-tail modes. Extensive experiments demonstrate consistent improvements across Rouge-L and BertScore metrics, surpassing greedy heuristics by 0.4-0.6 points and outperforming other baseline methods on diverse benchmarks.

cs.CL↗

arXiv2026

Lightweight and Fast Backdoor Model Detection

Deep neural networks (DNN), despite their remarkable performance, are highly vulnerable to backdoor attacks. Existing defenses mainly rely on activation anomaly analysis or trigger reverse engineering and often require clean samples or prior knowledge of trigger patterns, resulting in limited efficacy, practicability, and generalizability. More critically, while advanced attacks can implement backdoor implantation in milliseconds, current detection approaches typically demand minutes or even hours. To this end, we propose DFBScanner, a lightweight static parameter inspection framework for fast backdoor scanning. DFBScanner leverages our key observation that backdoor-induced feature perturbations can lead to distinctive and anomalous parameter updates in the final classification layer. Hence, we shift our detection focus from recognizing diverse and attack-specific trigger patterns targeted by prior work, to identifying the unified backdoor manifestation within the final layer, thereby enabling efficient and attack-agnostic detection. Specifically, by constructing and strategically combining multiple anomaly indicators of the final-layer parameters into a Trojan clue, DFBScanner detects backdoors through maximum anomaly scoring. DFBScanner is evaluated on a large-scale backdoor benchmark, including over 5,000 backdoor models trained on 4 datasets, 12 network architectures, 20 types of backdoor triggers, 2 attack strategies (all-to-one and -all), and 3 backdoor injection methods (data poisoning, training pipeline manipulation, and bit-flips). Numerical results show that DFBScanner achieves a 97.17% true-positive rate, 0.95% false-positive rate, and an average detection time of only 1 ms per model, significantly outperforming prior methods.

cs.CR↗

arXiv2026

Evolving Knowledge Distillation for Lightweight Neural Machine Translation

Recent advancements in Neural Machine Translation (NMT) have significantly improved translation quality. However, the increasing size and complexity of state-of-the-art models present significant challenges for deployment on resource-limited devices. Knowledge distillation (KD) is a promising approach for compressing models, but its effectiveness diminishes when there is a large capacity gap between teacher and student models. To address this issue, we propose Evolving Knowledge Distillation (EKD), a progressive training framework in which the student model learns from a sequence of teachers with gradually increasing capacities. Experiments on IWSLT-14, WMT-17, and WMT-23 benchmarks show that EKD leads to consistent improvements at each stage. On IWSLT-14, the final student achieves a BLEU score of 34.24, narrowing the gap to the strongest teacher (34.32 BLEU) to just 0.08 BLEU. Similar trends are observed on other datasets. These results demonstrate that EKD effectively bridges the capacity gap, enabling compact models to achieve performance close to that of much larger teacher models.Code and models are available at https://github.com/agi-content-generation/EKD.

cs.CL↗

arXiv2026

Detecting AI-Generated Content on Social Media with Multi-modal Language Models

Generative AI has enabled the creation of photorealistic images and videos that are increasingly disseminated on social media, often used for spam, misinformation, manipulation, and fraud. Existing AI-generated content (AIGC) detection methods face challenges including poor generalization to new generation models, reliance on single modalities, and lack of interpretable explanations. We present our pipeline that mitigates these issues by continuously curating diverse multi-modal social media data and training a compact vision-language model for detection and explanation. Our model achieves state-of-the-art detection performance on public benchmarks and demonstrates robust detection and explanation capabilities on internal social media datasets across multiple platforms. We deployed our model for post recommendation on social media platforms and observed positive downstream impacts on user engagement, demonstrating that it is feasible to perform effective AIGC detection in dynamic, real-world social media environments.

cs.CL↗

arXiv2025

Learning-based data-enabled economic predictive control with convex optimization for nonlinear systems

In this article, we propose a data-enabled economic predictive control method for a class of nonlinear systems, which aims to optimize the economic operational performance while handling hard constraints on the system outputs. Two lifting functions are constructed via training neural networks, which generate mapped input and mapped output in a higher-dimensional space, where the nonlinear economic cost function can be approximated using a quadratic function of the mapped variables. The data-enabled predictive control framework is extended to address nonlinear dynamics by using the mapped input and the mapped output that belong to a virtual linear representation, which serves as an approximation of the original nonlinear system. Additionally, we reconstruct the system output variables from the mapped output, on which hard output constraints are imposed. The online control problem is formulated as a convex optimization problem, despite the nonlinearity of the system dynamics and the original economic cost function. Theoretical analysis is presented to justify the suitability of the proposed method for nonlinear systems. We evaluate the proposed method through two large-scale industrial case studies: (i) a biological water treatment process, and (ii) a solvent-based shipboard post-combustion carbon capture process. These studies demonstrate its effectiveness and advantages.

eess.SY↗

arXiv2025

Economic zone data-enabled predictive control for connected open water systems

The real-time operation of open water systems is essential for ensuring operational safety, satisfying operational requirements, and optimizing energy usage. However, existing rule-based control strategies rely heavily on human experience, while model-based approaches depend on accurate hydrodynamic models, which limit their applicability to water systems with complex dynamics and uncertain disturbances. In this work, we develop a fully data-driven, zone-based control framework with adaptive control target zone selection for safe and energy-efficient operation of connected open water systems. Specifically, we propose a mixed-integer economic zone data-enabled predictive control (DeePC) approach that aims to maintain the water levels of the branches within the desired water-level zone while reducing real-time operational energy consumption. The DeePC-based approach enables direct use of input-output data for predictive control, eliminating the need for explicit dynamic modeling. To handle multiple control objectives with different priorities, we employ lexicographic optimization and reformulate the traditional DeePC cost function to incorporate zone tracking and energy consumption minimization objectives. Additionally, Bayesian optimization is utilized to determine the control target zone, which enables an effective trade-off between zone tracking and energy consumption in the presence of external disturbances. Comprehensive simulations and comparative analyses demonstrate the effectiveness of the proposed method. The proposed method maintains water levels within the desired water-level zone for 97.04% of the operating time, with an average energy consumption of 33.5 kWh per 0.5 hour. Compared to rule-based control method, the proposed method lowers zone-violation frequency by 74.96% and the average energy consumption by 22.44%.

eess.SY↗

arXiv2025

Inference Compute-Optimal Video Vision Language Models

This work investigates the optimal allocation of inference compute across three key scaling factors in video vision language models: language model size, frame count, and the number of visual tokens per frame. While prior works typically focuses on optimizing model efficiency or improving performance without considering resource constraints, we instead identify optimal model configuration under fixed inference compute budgets. We conduct large-scale training sweeps and careful parametric modeling of task performance to identify the inference compute-optimal frontier. Our experiments reveal how task performance depends on scaling factors and finetuning data size, as well as how changes in data size shift the compute-optimal frontier. These findings translate to practical tips for selecting these scaling factors.

cs.CV↗

arXiv2025

Economic data-enabled predictive control using machine learning

In this paper, we propose a convex data-based economic predictive control method within the framework of data-enabled predictive control (DeePC). Specifically, we use a neural network to transform the system output into a new state space, where the nonlinear economic cost function of the underlying nonlinear system is approximated using a quadratic function expressed by the transformed output in the new state space. Both the neural network parameters and the coefficients of the quadratic function are learned from open-loop data of the system. Additionally, we reconstruct constrained output variables from the transformed output through learning an output reconstruction matrix; this way, the proposed economic DeePC can handle output constraints explicitly. The performance of the proposed method is evaluated via a case study in a simulated chemical process.

eess.SY↗

arXiv2025

Machine learning-based hybrid dynamic modeling and economic predictive control of carbon capture process for ship decarbonization

Implementing carbon capture technology on-board ships holds promise as a solution to facilitate the reduction of carbon intensity in international shipping, as mandated by the International Maritime Organization. In this work, we address the energy-efficient operation of shipboard carbon capture processes by proposing a hybrid modeling-based economic predictive control scheme. Specifically, we consider a comprehensive shipboard carbon capture process that encompasses the ship engine system and the shipboard post-combustion carbon capture plant. To accurately and robustly characterize the dynamic behaviors of this shipboard plant, we develop a hybrid dynamic process model that integrates available imperfect physical knowledge with neural networks trained using process operation data. An economic model predictive control approach is proposed based on the hybrid model to ensure carbon capture efficiency while minimizing energy consumption required for the carbon capture process operation. The cross-entropy method is employed to efficiently solve the complex non-convex optimization problem associated with the proposed hybrid model-based economic model predictive control method. Extensive simulations, analyses, and comparisons are conducted to verify the effectiveness and illustrate the superiority of the proposed framework.

eess.SY↗

arXiv2024

CompCap: Improving Multimodal Large Language Models with Composite Captions

How well can Multimodal Large Language Models (MLLMs) understand composite images? Composite images (CIs) are synthetic visuals created by merging multiple visual elements, such as charts, posters, or screenshots, rather than being captured directly by a camera. While CIs are prevalent in real-world applications, recent MLLM developments have primarily focused on interpreting natural images (NIs). Our research reveals that current MLLMs face significant challenges in accurately understanding CIs, often struggling to extract information or perform complex reasoning based on these images. We find that existing training data for CIs are mostly formatted for question-answer tasks (e.g., in datasets like ChartQA and ScienceQA), while high-quality image-caption datasets, critical for robust vision-language alignment, are only available for NIs. To bridge this gap, we introduce Composite Captions (CompCap), a flexible framework that leverages Large Language Models (LLMs) and automation tools to synthesize CIs with accurate and detailed captions. Using CompCap, we curate CompCap-118K, a dataset containing 118K image-caption pairs across six CI types. We validate the effectiveness of CompCap-118K by supervised fine-tuning MLLMs of three sizes: xGen-MM-inst.-4B and LLaVA-NeXT-Vicuna-7B/13B. Empirical results show that CompCap-118K significantly enhances MLLMs' understanding of CIs, yielding average gains of 1.7%, 2.0%, and 2.9% across eleven benchmarks, respectively.

cs.CV↗

arXiv2024

Deep DeePC: Data-enabled predictive control with low or no online optimization using deep learning

Data-enabled predictive control (DeePC) is a data-driven control algorithm that utilizes data matrices to form a non-parametric representation of the underlying system, predicting future behaviors and generating optimal control actions. DeePC typically requires solving an online optimization problem, the complexity of which is heavily influenced by the amount of data used, potentially leading to expensive online computation. In this paper, we leverage deep learning to propose a highly computationally efficient DeePC approach for general nonlinear processes, referred to as Deep DeePC. Specifically, a deep neural network is employed to learn the DeePC vector operator, which is an essential component of the non-parametric representation of DeePC. This neural network is trained offline using historical open-loop input and output data of the nonlinear process. With the trained neural network, the Deep DeePC framework is formed for online control implementation. At each sampling instant, this neural network directly outputs the DeePC operator, eliminating the need for online optimization as conventional DeePC. The optimal control action is obtained based on the DeePC operator updated by the trained neural network. To address constrained scenarios, a constraint handling scheme is further proposed and integrated with the Deep DeePC to handle hard constraints during online implementation. The efficacy and superiority of the proposed Deep DeePC approach are demonstrated using two benchmark process examples.

eess.SY↗

arXiv2024

Data-driven parallel Koopman subsystem modeling and distributed moving horizon state estimation for large-scale nonlinear processes

In this work, we consider a state estimation problem for large-scale nonlinear processes in the absence of first-principles process models. By exploiting process operation data, both process modeling and state estimation design are addressed within a distributed framework. By leveraging the Koopman operator concept, a parallel subsystem modeling approach is proposed to establish interactive linear subsystem process models in higher-dimensional subspaces, each of which correlates with the original nonlinear subspace of the corresponding process subsystem via a nonlinear mapping. The data-driven linear subsystem models can be used to collaboratively characterize and predict the dynamical behaviors of the entire nonlinear process. Based on the established subsystem models, local state estimators that can explicitly handle process operation constraints are designed using moving horizon estimation. The local estimators are integrated via information exchange to form a distributed estimation scheme, which provides estimates of the unmeasured/unmeasurable state variables of the original nonlinear process in a linear manner. The proposed framework is applied to a chemical process and an agro-hydrological process to illustrate its effectiveness and applicability. Good open-loop predictability of the linear subsystem models is confirmed, and accurate estimates of the process states are obtained without requiring a first-principles process model.

eess.SY↗

arXiv2024

Reduced-order Koopman modeling and predictive control of nonlinear processes

In this paper, we propose an efficient data-driven predictive control approach for general nonlinear processes based on a reduced-order Koopman operator. A Kalman-based sparse identification of nonlinear dynamics method is employed to select lifting functions for Koopman identification. The selected lifting functions are used to project the original nonlinear state-space into a higher-dimensional linear function space, in which Koopman-based linear models can be constructed for the underlying nonlinear process. To curb the significant increase in the dimensionality of the resulting full-order Koopman models caused by the use of lifting functions, we propose a reduced-order Koopman modeling approach based on proper orthogonal decomposition. A computationally efficient linear robust predictive control scheme is established based on the reduced-order Koopman model. A case study on a benchmark chemical process is conducted to illustrate the effectiveness of the proposed method. Comprehensive comparisons are conducted to demonstrate the advantage of the proposed method.

eess.SY↗

arXiv2024

M2fNet: Multi-modal Forest Monitoring Network on Large-scale Virtual Dataset

Forest monitoring and education are key to forest protection, education and management, which is an effective way to measure the progress of a country's forest and climate commitments. Due to the lack of a large-scale wild forest monitoring benchmark, the common practice is to train the model on a common outdoor benchmark (e.g., KITTI) and evaluate it on real forest datasets (e.g., CanaTree100). However, there is a large domain gap in this setting, which makes the evaluation and deployment difficult. In this paper, we propose a new photorealistic virtual forest dataset and a multimodal transformer-based algorithm for tree detection and instance segmentation. To the best of our knowledge, it is the first time that a multimodal detection and segmentation algorithm is applied to large-scale forest scenes. We believe that the proposed dataset and method will inspire the simulation, computer vision, education, and forestry communities towards a more comprehensive multi-modal understanding.

cs.GR↗

arXiv2021

Deep Learning Based Defect Detection for Solder Joints on Industrial X-Ray Circuit Board Images

Quality control is of vital importance during electronics production. As the methods of producing electronic circuits improve, there is an increasing chance of solder defects during assembling the printed circuit board (PCB). Many technologies have been incorporated for inspecting failed soldering, such as X-ray imaging, optical imaging, and thermal imaging. With some advanced algorithms, the new technologies are expected to control the production quality based on the digital images. However, current algorithms sometimes are not accurate enough to meet the quality control. Specialists are needed to do a follow-up checking. For automated X-ray inspection, joint of interest on the X-ray image is located by region of interest (ROI) and inspected by some algorithms. Some incorrect ROIs deteriorate the inspection algorithm. The high dimension of X-ray images and the varying sizes of image dimensions also challenge the inspection algorithms. On the other hand, recent advances on deep learning shed light on image-based tasks and are competitive to human levels. In this paper, deep learning is incorporated in X-ray imaging based quality control during PCB quality inspection. Two artificial intelligence (AI) based models are proposed and compared for joint defect detection. The noised ROI problem and the varying sizes of imaging dimension problem are addressed. The efficacy of the proposed methods are verified through experimenting on a real-world 3D X-ray dataset. By incorporating the proposed methods, specialist inspection workload is largely saved.

eess.IV↗

arXiv2021

Time-Series Regeneration with Convolutional Recurrent Generative Adversarial Network for Remaining Useful Life Estimation

For health prognostic task, ever-increasing efforts have been focused on machine learning-based methods, which are capable of yielding accurate remaining useful life (RUL) estimation for industrial equipment or components without exploring the degradation mechanism. A prerequisite ensuring the success of these methods depends on a wealth of run-to-failure data, however, run-to-failure data may be insufficient in practice. That is, conducting a substantial amount of destructive experiments not only is high costs, but also may cause catastrophic consequences. Out of this consideration, an enhanced RUL framework focusing on data self-generation is put forward for both non-cyclic and cyclic degradation patterns for the first time. It is designed to enrich data from a data-driven way, generating realistic-like time-series to enhance current RUL methods. First, high-quality data generation is ensured through the proposed convolutional recurrent generative adversarial network (CR-GAN), which adopts a two-channel fusion convolutional recurrent neural network. Next, a hierarchical framework is proposed to combine generated data into current RUL estimation methods. Finally, the efficacy of the proposed method is verified through both non-cyclic and cyclic degradation systems. With the enhanced RUL framework, an aero-engine system following non-cyclic degradation has been tested using three typical RUL models. State-of-art RUL estimation results are achieved by enhancing capsule network with generated time-series. Specifically, estimation errors evaluated by the index score function have been reduced by 21.77%, and 32.67% for the two employed operating conditions, respectively. Besides, the estimation error is reduced to zero for the Lithium-ion battery system, which presents cyclic degradation.

cs.LG↗

arXiv2018

Anisotropic EM Segmentation by 3D Affinity Learning and Agglomeration

The field of connectomics has recently produced neuron wiring diagrams from relatively large brain regions from multiple animals. Most of these neural reconstructions were computed from isotropic (e.g., FIBSEM) or near isotropic (e.g., SBEM) data. In spite of the remarkable progress on algorithms in recent years, automatic dense reconstruction from anisotropic data remains a challenge for the connectomics community. One significant hurdle in the segmentation of anisotropic data is the difficulty in generating a suitable initial over-segmentation. In this study, we present a segmentation method for anisotropic EM data that agglomerates a 3D over-segmentation computed from the 3D affinity prediction. A 3D U-net is trained to predict 3D affinities by the MALIS approach. Experiments on multiple datasets demonstrates the strength and robustness of the proposed method for anisotropic EM segmentation.

cs.CV↗