Zhengyang Li

Zhengyang Li, Zhenglin Du, Yi Wen et al.

The CASTLE Challenge @ EgoVis 2026 evaluates long-form egocentric video question answering over 600+ hours of multi-perspective recordings. Each four-choice question requires evidence from videos, transcripts, auxiliary photos, people, days, rooms, and temporal context. We propose an evidence-aware multimodal reasoning pipeline based on Qwen. Our system parses question hints, retrieves ASR chunks, attaches auxiliary images, samples candidate video frames, and routes questions into static visual, speech/text, temporal, and mixed types with specialized prompts. Multiple inference passes are aggregated by confidence-weighted voting and converted into the official Codabench format. In ablation, LoRA improves the score from 0.21 to 0.50, and more sampled frames further raise it to 0.58. Our final system ranks first in the CASTLE Challenge @ EgoVis 2026.

Yanhu Wang, Muhammad Muzammil Afzal, Zhengyang Li et al.

Traditional base station siting (BSS) methods rely heavily on drive testing and user feedback, which are laborious and require extensive expertise in communication, networking, and optimization. As large language models (LLMs) and their associated technologies advance, particularly in the realms of prompt engineering and agent engineering, network optimization will witness a revolutionary approach. This approach entails the strategic use of well-crafted prompts to infuse human experience and knowledge into these sophisticated LLMs, and the deployment of autonomous agents as a communication bridge to seamlessly connect the machine language based LLMs with human users using natural language. Furthermore, our proposed framework incorporates retrieval-augmented generation (RAG) to enhance the system's ability to acquire domain-specific knowledge and generate solutions, thereby enabling the customization and optimization of the BSS process. This integration represents the future paradigm of artificial intelligence (AI) as a service and AI for more ease. This research first develops a novel LLM-empowered BSS optimization framework, and heuristically proposes three different potential implementations: the strategies based on Prompt-optimized LLM (PoL), LLM-empowered autonomous BSS agent (LaBa), and Cooperative multiple LLM-based autonomous BSS agents (CLaBa). Through evaluation on real-world data, the experiments demonstrate that prompt-assisted LLMs and LLM-based agents can generate more efficient and reliable network deployments, noticeably enhancing the efficiency of BSS optimization and reducing trivial manual participation.

Timo Lohrenz, Björn Möller, Zhengyang Li et al.

The powerful modeling capabilities of all-attention-based transformer architectures often cause overfitting and - for natural language processing tasks - lead to an implicitly learned internal language model in the autoregressive transformer decoder complicating the integration of external language models. In this paper, we explore relaxed attention, a simple and easy-to-implement smoothing of the attention weights, yielding a two-fold improvement to the general transformer architecture: First, relaxed attention provides regularization when applied to the self-attention layers in the encoder. Second, we show that it naturally supports the integration of an external language model as it suppresses the implicitly learned internal language model by relaxing the cross attention in the decoder. We demonstrate the benefit of relaxed attention across several tasks with clear improvement in combination with recent benchmark approaches. Specifically, we exceed the former state-of-the-art performance of 26.90% word error rate on the largest public lip-reading LRS3 benchmark with a word error rate of 26.31%, as well as we achieve a top-performing BLEU score of 37.67 on the IWSLT14 (DE$\rightarrow$EN) machine translation task without external language models and virtually no additional model parameters. Code and models will be made publicly available.

Zhengyang Li, Shijing Si, Jianzong Wang et al.

Pre-trained BERT models have achieved impressive performance in many natural language processing (NLP) tasks. However, in many real-world situations, textual data are usually decentralized over many clients and unable to be uploaded to a central server due to privacy protection and regulations. Federated learning (FL) enables multiple clients collaboratively to train a global model while keeping the local data privacy. A few researches have investigated BERT in federated learning setting, but the problem of performance loss caused by heterogeneous (e.g., non-IID) data over clients remain under-explored. To address this issue, we propose a framework, FedSplitBERT, which handles heterogeneous data and decreases the communication cost by splitting the BERT encoder layers into local part and global part. The local part parameters are trained by the local client only while the global part parameters are trained by aggregating gradients of multiple clients. Due to the sheer size of BERT, we explore a quantization method to further reduce the communication cost with minimal performance loss. Our framework is ready-to-use and compatible to many existing federated learning algorithms, including FedAvg, FedProx and FedAdam. Our experiments verify the effectiveness of the proposed framework, which outperforms baseline methods by a significant margin, while FedSplitBERT with quantization can reduce the communication cost by $11.9\times$.

Zhengyang Li, Thomas Graave, Björn Möller et al.

In audiovisual automatic speech recognition (AV-ASR) systems, information fusion of visual features in a pre-trained ASR has been proven as a promising method to improve noise robustness. In this work, based on the prominent Whisper ASR, first, we propose a simple and effective visual fusion method -- use of visual features both in encoder and decoder (dual-use) -- to learn the audiovisual interactions in the encoder and to weigh modalities in the decoder. Second, we compare visual fusion methods in Whisper models of various sizes. Our proposed dual-use method shows consistent noise robustness improvement, e.g., a 35% relative improvement (WER: 4.41% vs. 6.83%) based on Whisper small, and a 57% relative improvement (WER: 4.07% vs. 9.53%) based on Whisper medium, compared to typical reference middle fusion in babble noise with a signal-to-noise ratio (SNR) of 0dB. Third, we conduct ablation studies examining the impact of various module designs and fusion options. Fine-tuned on 1929 hours of audiovisual data, our dual-use method using Whisper medium achieves 4.08% (MUSAN babble noise) and 4.43% (NoiseX babble noise) average WER across various SNRs, thereby establishing a new state-of-the-art in noisy conditions on the LRS3 AV-ASR benchmark. Our code is at https://github.com/ifnspaml/Dual-Use-AVASR

Wenhao Liang, Zhengyang Li, Weitong Chen

This paper introduces the FinSen dataset that revolutionizes financial market analysis by integrating economic and financial news articles from 197 countries with stock market data. The dataset's extensive coverage spans 15 years from 2007 to 2023 with temporal information, offering a rich, global perspective with 160,000 records on financial market news. Our study leverages causally validated sentiment scores and LSTM models to enhance market forecast accuracy and reliability. Utilizing the FinSen dataset, we introduce an innovative Focal Calibration Loss, reducing Expected Calibration Error (ECE) to 3.34 percent with the DAN 3 model. This not only improves prediction accuracy but also aligns probabilistic forecasts closely with real outcomes, crucial for the financial sector where predicted probability is paramount. Our approach demonstrates the effectiveness of combining sentiment analysis with precise calibration techniques for trustworthy financial forecasting where the cost of misinterpretation can be high. Finsen Data can be found at [this github URL](https://github.com/EagleAdelaide/FinSen_Dataset.git).

Liangwei Nathan Zheng, Zhengyang Li, Chang George Dong et al.

Irregular Time Series Data (IRTS) has shown increasing prevalence in real-world applications. We observed that IRTS can be divided into two specialized types: Natural Irregular Time Series (NIRTS) and Accidental Irregular Time Series (AIRTS). Various existing methods either ignore the impacts of irregular patterns or statically learn the irregular dynamics of NIRTS and AIRTS data and suffer from limited data availability due to the sparsity of IRTS. We proposed a novel transformer-based framework for general irregular time series data that treats IRTS from four views: Locality, Time, Spatio and Irregularity to motivate the data usage to the highest potential. Moreover, we design a sophisticated irregularity-gate mechanism to adaptively select task-relevant information from irregularity, which improves the generalization ability to various IRTS data. We implement extensive experiments to demonstrate the resistance of our work to three highly missing ratio datasets (88.4\%, 94.9\%, 60\% missing value) and investigate the significance of the irregularity information for both NIRTS and AIRTS by additional ablation study. We release our implementation in https://github.com/IcurasLW/MTSFormer-Irregular_Time_Series.git

Miao Zhang, Peng Jia, Zhengyang Li et al.

Sky survey telescopes play a critical role in modern astronomy, but misalignment of their optical elements can introduce significant variations in point spread functions, leading to reduced data quality. To address this, we need a method to obtain misalignment states, aiding in the reconstruction of accurate point spread functions for data processing methods or facilitating adjustments of optical components for improved image quality. Since sky survey telescopes consist of many optical elements, they result in a vast array of potential misalignment states, some of which are intricately coupled, posing detection challenges. However, by continuously adjusting the misalignment states of optical elements, we can disentangle coupled states. Based on this principle, we propose a deep neural network to extract misalignment states from continuously varying point spread functions in different field of views. To ensure sufficient and diverse training data, we recommend employing a digital twin to obtain data for neural network training. Additionally, we introduce the state graph to store misalignment data and explore complex relationships between misalignment states and corresponding point spread functions, guiding the generation of training data from experiments. Once trained, the neural network estimates misalignment states from observation data, regardless of the impacts caused by atmospheric turbulence, noise, and limited spatial sampling rates in the detector. The method proposed in this paper could be used to provide prior information for the active optics system and the optical system alignment.

Chang Dong, Zhengyang Li, Liangwei Zheng et al.

Recently, the issue of adversarial robustness in the time series domain has garnered significant attention. However, the available defense mechanisms remain limited, with adversarial training being the predominant approach, though it does not provide theoretical guarantees. Randomized Smoothing has emerged as a standout method due to its ability to certify a provable lower bound on robustness radius under $\ell_p$-ball attacks. Recognizing its success, research in the time series domain has started focusing on these aspects. However, existing research predominantly focuses on time series forecasting, or under the non-$\ell_p$ robustness in statistic feature augmentation for time series classification~(TSC). Our review found that Randomized Smoothing performs modestly in TSC, struggling to provide effective assurances on datasets with poor robustness. Therefore, we propose a self-ensemble method to enhance the lower bound of the probability confidence of predicted labels by reducing the variance of classification margins, thereby certifying a larger radius. This approach also addresses the computational overhead issue of Deep Ensemble~(DE) while remaining competitive and, in some cases, outperforming it in terms of robustness. Both theoretical analysis and experimental results validate the effectiveness of our method, demonstrating superior performance in robustness testing compared to baseline approaches.

Björn Möller, Zhengyang Li, Malte Stelzer et al.

Recent successful video generation systems that predict and create realistic automotive driving scenes from short video inputs assign tokenization, future state prediction (world model), and video decoding to dedicated models. These approaches often utilize large models that require significant training resources, offer limited insight into design choices, and lack publicly available code and datasets. In this work, we address these deficiencies and present OpenViGA, an open video generation system for automotive driving scenes. Our contributions are: Unlike several earlier works for video generation, such as GAIA-1, we provide a deep analysis of the three components of our system by separate quantitative and qualitative evaluation: Image tokenizer, world model, video decoder. Second, we purely build upon powerful pre-trained open source models from various domains, which we fine-tune by publicly available automotive data (BDD100K) on GPU hardware at academic scale. Third, we build a coherent video generation system by streamlining interfaces of our components. Fourth, due to public availability of the underlying models and data, we allow full reproducibility. Finally, we also publish our code and models on Github. For an image size of 256x256 at 4 fps we are able to predict realistic driving scene videos frame-by-frame with only one frame of algorithmic latency.

Zhengyang Li, Wenhao Liang, Chang Dong et al.

This study investigates the vulnerability of time series classification models to adversarial attacks, with a focus on how these models process local versus global information under such conditions. By leveraging the Normalized Auto Correlation Function (NACF), an exploration into the inclination of neural networks is conducted. It is demonstrated that regularization techniques, particularly those employing Fast Fourier Transform (FFT) methods and targeting frequency components of perturbations, markedly enhance the effectiveness of attacks. Meanwhile, the defense strategies, like noise introduction and Gaussian filtering, are shown to significantly lower the Attack Success Rate (ASR), with approaches based on noise introducing notably effective in countering high-frequency distortions. Furthermore, models designed to prioritize global information are revealed to possess greater resistance to adversarial manipulations. These results underline the importance of designing attack and defense mechanisms, informed by frequency domain analysis, as a means to considerably reinforce the resilience of neural network models against adversarial threats.

Wei Luo, Yiting Lu, Xin Li et al.

This paper reports on the LoViF 2026 PhyScore challenge, a competition on holistic quality assessment of world-model-generated videos across both 2D and 4D generation settings. The challenge is motivated by a central gap in current evaluation practice: perceptual quality alone is insufficient to judge whether generated dynamics are physically plausible, temporally coherent, and consistent with input conditions. Participants are required to build a metric that jointly predicts four dimensions, i.e., Video Quality, Physical Realism, Condition-Video Alignment, and Temporal Consistency. Depart from that, participants also need to localize physical anomaly timestamps for fine-grained diagnosis. The benchmark dataset contains 1,554 videos generated by seven representative world generative models, organized into three tracks (text-2D, image-to-4D, and video-to-4D) and spanning 26 categories. These categories explicitly cover physics-relevant scenarios, including dynamics, optics, and thermodynamics, together with diverse real-world and creative content. To ensure label reliability, scores and anomaly timestamps are produced through trained human annotation with an additional automated quality-control pass. Evaluation is based on both score prediction and anomaly localization, with a composite protocol that combines TimeStamp_IOU and SRCC/PLCC. This report summarizes the challenge design and provides method-level insights from submitted solutions.

Zhengyang Li, Qijin Ji, Xinghong Ling et al.

Recent advancements in multi-agent reinforcement learning (MARL) have demonstrated its application potential in modern games. Beginning with foundational work and progressing to landmark achievements such as AlphaStar in StarCraft II and OpenAI Five in Dota 2, MARL has proven capable of achieving superhuman performance across diverse game environments through techniques like self-play, supervised learning, and deep reinforcement learning. With its growing impact, a comprehensive review has become increasingly important in this field. This paper aims to provide a thorough examination of MARL's application from turn-based two-agent games to real-time multi-agent video games including popular genres such as Sports games, First-Person Shooter (FPS) games, Real-Time Strategy (RTS) games and Multiplayer Online Battle Arena (MOBA) games. We further analyze critical challenges posed by MARL in video games, including nonstationary, partial observability, sparse rewards, team coordination, and scalability, and highlight successful implementations in games like Rocket League, Minecraft, Quake III Arena, StarCraft II, Dota 2, Honor of Kings, etc. This paper offers insights into MARL in video game AI systems, proposes a novel method to estimate game complexity, and suggests future research directions to advance MARL and its applications in game development, inspiring further innovation in this rapidly evolving field.

Haomin Zuo, Zhengyang Li, Jiangchuan Gong et al.

As urbanization speeds up and traffic flow increases, the issue of pavement distress is becoming increasingly pronounced, posing a severe threat to road safety and service life. Traditional methods of pothole detection rely on manual inspection, which is not only inefficient but also costly. This paper proposes an intelligent road crack detection and analysis system, based on the enhanced YOLOv8 deep learning framework. A target segmentation model has been developed through the training of 4029 images, capable of efficiently and accurately recognizing and segmenting crack regions in roads. The model also analyzes the segmented regions to precisely calculate the maximum and minimum widths of cracks and their exact locations. Experimental results indicate that the incorporation of ECA and CBAM attention mechanisms substantially enhances the model's detection accuracy and efficiency, offering a novel solution for road maintenance and safety monitoring.

Peng Xu, Shengwu Xiong, Jiajun Zhang et al.

This paper reviews the MARS2 2025 Challenge on Multimodal Reasoning. We aim to bring together different approaches in multimodal machine learning and LLMs via a large benchmark. We hope it better allows researchers to follow the state-of-the-art in this very dynamic area. Meanwhile, a growing number of testbeds have boosted the evolution of general-purpose large language models. Thus, this year's MARS2 focuses on real-world and specialized scenarios to broaden the multimodal reasoning applications of MLLMs. Our organizing team released two tailored datasets Lens and AdsQA as test sets, which support general reasoning in 12 daily scenarios and domain-specific reasoning in advertisement videos, respectively. We evaluated 40+ baselines that include both generalist MLLMs and task-specific models, and opened up three competition tracks, i.e., Visual Grounding in Real-world Scenarios (VG-RS), Visual Question Answering with Spatial Awareness (VQA-SA), and Visual Reasoning in Creative Advertisement Videos (VR-Ads). Finally, 76 teams from the renowned academic and industrial institutions have registered and 40+ valid submissions (out of 1200+) have been included in our ranking lists. Our datasets, code sets (40+ baselines and 15+ participants' methods), and rankings are publicly available on the MARS2 workshop website and our GitHub organization page https://github.com/mars2workshop/, where our updates and announcements of upcoming events will be continuously provided.

Zhengyang Li, Sawyer Campos, Nana Wang

This paper introduces LLM-MARL, a unified framework that incorporates large language models (LLMs) into multi-agent reinforcement learning (MARL) to enhance coordination, communication, and generalization in simulated game environments. The framework features three modular components of Coordinator, Communicator, and Memory, which dynamically generate subgoals, facilitate symbolic inter-agent messaging, and support episodic recall. Training combines PPO with a language-conditioned loss and LLM query gating. LLM-MARL is evaluated in Google Research Football, MAgent Battle, and StarCraft II. Results show consistent improvements over MAPPO and QMIX in win rate, coordination score, and zero-shot generalization. Ablation studies demonstrate that subgoal generation and language-based messaging each contribute significantly to performance gains. Qualitative analysis reveals emergent behaviors such as role specialization and communication-driven tactics. By bridging language modeling and policy learning, this work contributes to the design of intelligent, cooperative agents in interactive simulations. It offers a path forward for leveraging LLMs in multi-agent systems used for training, games, and human-AI collaboration.

Timo Lohrenz, Patrick Schwarz, Zhengyang Li et al.

Recently, attention-based encoder-decoder (AED) models have shown high performance for end-to-end automatic speech recognition (ASR) across several tasks. Addressing overconfidence in such models, in this paper we introduce the concept of relaxed attention, which is a simple gradual injection of a uniform distribution to the encoder-decoder attention weights during training that is easily implemented with two lines of code. We investigate the effect of relaxed attention across different AED model architectures and two prominent ASR tasks, Wall Street Journal (WSJ) and Librispeech. We found that transformers trained with relaxed attention outperform the standard baseline models consistently during decoding with external language models. On WSJ, we set a new benchmark for transformer-based end-to-end speech recognition with a word error rate of 3.65%, outperforming state of the art (4.20%) by 13.1% relative, while introducing only a single hyperparameter.

Timo Lohrenz, Zhengyang Li, Tim Fingscheidt

Stream fusion, also known as system combination, is a common technique in automatic speech recognition for traditional hybrid hidden Markov model approaches, yet mostly unexplored for modern deep neural network end-to-end model architectures. Here, we investigate various fusion techniques for the all-attention-based encoder-decoder architecture known as the transformer, striving to achieve optimal fusion by investigating different fusion levels in an example single-microphone setting with fusion of standard magnitude and phase features. We introduce a novel multi-encoder learning method that performs a weighted combination of two encoder-decoder multi-head attention outputs only during training. Employing then only the magnitude feature encoder in inference, we are able to show consistent improvement on Wall Street Journal (WSJ) with language model and on Librispeech, without increase in runtime or parameters. Combining two such multi-encoder trained models by a simple late fusion in inference, we achieve state-of-the-art performance for transformer-based models on WSJ with a significant WER reduction of 19% relative compared to the current benchmark approach.

Peng Jia, Xuebo Wu, Zhengyang Li et al.

The point spread function (PSF) reflects states of a telescope and plays an important role in development of data processing methods, such as PSF based astrometry, photometry and image restoration. However, for wide field small aperture telescopes (WFSATs), estimating PSF in any position of the whole field of view is hard, because aberrations induced by the optical system are quite complex and the signal to noise ratio of star images is often too low for PSF estimation. In this paper, we further develop our deep neural network (DNN) based PSF modelling method and show its applications in PSF estimation. During the telescope alignment and testing stage, our method collects system calibration data through modification of optical elements within engineering tolerances (tilting and decentering). Then we use these data to train a DNN (Tel--Net). After training, the Tel--Net can estimate PSF in any field of view from several discretely sampled star images. We use both simulated and experimental data to test performance of our method. The results show that the Tel--Net can successfully reconstruct PSFs of WFSATs of any states and in any positions of the FoV. Its results are significantly more precise than results obtained by the compared classic method - Inverse Distance Weight (IDW) interpolation. Our method provides foundations for developing of deep neural network based data processing methods for WFSATs, which require strong prior information of PSFs.

Peng Jia, Xiyu Li, Zhengyang Li et al.

The point spread function reflects the state of an optical telescope and it is important for data post-processing methods design. For wide field small aperture telescopes, the point spread function is hard to model, because it is affected by many different effects and has strong temporal and spatial variations. In this paper, we propose to use the denoising autoencoder, a type of deep neural network, to model the point spread function of wide field small aperture telescopes. The denoising autoencoder is a pure data based point spread function modelling method, which uses calibration data from real observations or numerical simulated results as point spread function templates. According to real observation conditions, different levels of random noise or aberrations are added to point spread function templates, making them as realizations of the point spread function, i.e., simulated star images. Then we train the denoising autoencoder with realizations and templates of the point spread function. After training, the denoising autoencoder learns the manifold space of the point spread function and can map any star images obtained by wide field small aperture telescopes directly to its point spread function, which could be used to design data post-processing or optical system alignment methods.