Yang Xiang

Yexing Du, Kaiyuan Liu, Youcheng Pan et al.

Multimodal large language models (MLLMs) have demonstrated significant potential for speech-to-text translation (S2TT). However, existing deployment paradigms face critical challenges: pure on-device models suffer from resource constraints, while centralized cloud systems incur severe privacy risks and bandwidth bottlenecks by transmitting raw voice data. Furthermore, most models exhibit English-centric biases, restricting many-to-many translation scaling. In this paper, we propose Edge-cloud Speech Recognition and Translation (ESRT), a privacy-preserving and bandwidth-efficient collaborative edge-cloud MLLM framework. Specifically, we design an edge-cloud split inference architecture that retains a lightweight speech encoder and adapter on the device, transmitting only highly compressed intermediate features to the cloud. This fundamentally prevents voiceprint leakage and reduces bandwidth requirements by up to 10$\times$. To overcome English-centric bottlenecks, we introduce a multi-task weighted curriculum learning strategy with data balancing to ensure robust cross-lingual consistency. Extensive experiments on the FLEURS dataset demonstrate that our models, ESRT-4B and ESRT-12B, achieve state-of-the-art many-to-many S2TT performance across 45 languages ($45 \times 44$ directions). Code and models are released to facilitate reproducible, privacy-aware MLLM S2TT research. The code and models are released at https://github.com/yxduir/esrt.

Yexing Du, Youcheng Pan, Ziyang Ma et al.

Multimodal Large Language Models (MLLMs) have achieved significant success in Speech-to-Text Translation (S2TT) tasks. While most existing research has focused on English-centric translation directions, the exploration of many-to-many translation is still limited by the scarcity of parallel data. To address this, we propose a three-stage curriculum learning strategy that leverages the machine translation capabilities of large language models and adapts them to S2TT tasks, enabling effective learning in low-resource settings. We trained MLLMs with varying parameter sizes (3B, 7B, and 32B) and evaluated the proposed strategy using the FLEURS and CoVoST-2 datasets. Experimental results show that the proposed strategy achieves state-of-the-art average performance in $15\times14$ language pairs, requiring fewer than 10 hours of speech data per language to achieve competitive results. The source code and models are released at https://github.com/yxduir/LLM-SRT.

Bowen Li, Shaotong Guo, Zhen Wang et al.

Building state-of-the-art text-to-speech (TTS) systems typically demands millions of hours of proprietary data and complex multi-stage architectures, creating substantial barriers for resource-constrained research teams. In this report, we present PilotTTS, a lightweight autoregressive TTS system that achieves competitive performance through minimalist architecture and rigorous data engineering. PilotTTS is trained on only 200K hours of data processed entirely with open-source tools. Specifically, our contributions are: (1) a reproducible multi-stage data processing pipeline covering quality assessment, label annotation, and filtering, and (2) a compact model architecture that employs Q-Former-based conditioning to decouple speaker identity from speaking style via cross-sample paired training. Within a unified framework, PilotTTS supports zero-shot voice cloning, emotion synthesis (11 categories), paralinguistic synthesis (4 categories), and Chinese dialect synthesis (14 dialects). On the Seed-TTS Eval benchmark, PilotTTS achieves the lowest WER of 1.50% on test-en, a CER of 0.87% on test-zh, and the highest speaker similarity on both test sets (0.862 and 0.815), outperforming systems trained on significantly larger datasets. We release the complete data pipeline recipe, pretrained weights, and code at https://github.com/AMAPVOICE/PilotTTS.

Jiaheng Wei, Yanjun Zhang, Leo Yu Zhang et al. · tencent-ai

Federated Learning (FL) enables distributed participants (e.g., mobile devices) to train a global model without sharing data directly to a central server. Recent studies have revealed that FL is vulnerable to gradient inversion attack (GIA), which aims to reconstruct the original training samples and poses high risk against the privacy of clients in FL. However, most existing GIAs necessitate control over the server and rely on strong prior knowledge including batch normalization and data distribution information. In this work, we propose Client-side poisoning Gradient Inversion (CGI), which is a novel attack method that can be launched from clients. For the first time, we show the feasibility of a client-side adversary with limited knowledge being able to recover the training samples from the aggregated global model. We take a distinct approach in which the adversary utilizes a malicious model that amplifies the loss of a specific targeted class of interest. When honest clients employ the poisoned global model, the gradients of samples belonging to the targeted class are magnified, making them the dominant factor in the aggregated update. This enables the adversary to effectively reconstruct the private input belonging to other clients using the aggregated update. In addition, our CGI also features its ability to remain stealthy against Byzantine-robust aggregation rules (AGRs). By optimizing malicious updates and blending benign updates with a malicious replacement vector, our method remains undetected by these defense mechanisms. To evaluate the performance of CGI, we conduct experiments on various benchmark datasets, considering representative Byzantine-robust AGRs, and exploring diverse FL settings with different levels of adversary knowledge about the data. Our results demonstrate that CGI consistently and successfully extracts training input in all tested scenarios.

Wanlun Ma, Derui Wang, Ruoxi Sun et al.

Deep Neural Networks (DNNs) are susceptible to backdoor attacks during training. The model corrupted in this way functions normally, but when triggered by certain patterns in the input, produces a predefined target label. Existing defenses usually rely on the assumption of the universal backdoor setting in which poisoned samples share the same uniform trigger. However, recent advanced backdoor attacks show that this assumption is no longer valid in dynamic backdoors where the triggers vary from input to input, thereby defeating the existing defenses. In this work, we propose a novel technique, Beatrix (backdoor detection via Gram matrix). Beatrix utilizes Gram matrix to capture not only the feature correlations but also the appropriately high-order information of the representations. By learning class-conditional statistics from activation patterns of normal samples, Beatrix can identify poisoned samples by capturing the anomalies in activation patterns. To further improve the performance in identifying target labels, Beatrix leverages kernel-based testing without making any prior assumptions on representation distribution. We demonstrate the effectiveness of our method through extensive evaluation and comparison with state-of-the-art defensive techniques. The experimental results show that our approach achieves an F1 score of 91.1% in detecting dynamic backdoors, while the state of the art can only reach 36.9%.

Bingxin Wang, Xiaowen Fu, Yuan Lan et al.

Pre-trained large transformer models have achieved remarkable performance in the fields of natural language processing and computer vision. However, the limited availability of public electroencephalogram (EEG) data presents a unique challenge for extending the success of these models to EEG-based tasks. To address this gap, we propose AdaCT, plug-and-play Adapters designed for Converting Time series data into spatio-temporal 2D pseudo-images or text forms. Essentially, AdaCT-I transforms multi-channel or lengthy single-channel time series data into spatio-temporal 2D pseudo-images for fine-tuning pre-trained vision transformers, while AdaCT-T converts short single-channel data into text for fine-tuning pre-trained language transformers. The proposed approach allows for seamless integration of pre-trained vision models and language models in time series decoding tasks, particularly in EEG data analysis. Experimental results on diverse benchmark datasets, including Epileptic Seizure Recognition, Sleep-EDF, and UCI HAR, demonstrate the superiority of AdaCT over baseline methods. Overall, we provide a promising transfer learning framework for leveraging the capabilities of pre-trained vision and language models in EEG-based tasks, thereby advancing the field of time series decoding and enhancing interpretability in EEG data analysis. Our code will be available at https://github.com/wangbxj1234/AdaCE.

Chuer Yu, Xuhong Zhang, Yuxuan Duan et al.

Despite the fact that DeepFake forgery detection algorithms have achieved impressive performance on known manipulations, they often face disastrous performance degradation when generalized to an unseen manipulation. Some recent works show improvement in generalization but rely on features fragile to image distortions such as compression. To this end, we propose Diff-ID, a concise and effective approach that explains and measures the identity loss induced by facial manipulations. When testing on an image of a specific person, Diff-ID utilizes an authentic image of that person as a reference and aligns them to the same identity-insensitive attribute feature space by applying a face-swapping generator. We then visualize the identity loss between the test and the reference image from the image differences of the aligned pairs, and design a custom metric to quantify the identity loss. The metric is then proved to be effective in distinguishing the forgery images from the real ones. Extensive experiments show that our approach achieves high detection performance on DeepFake images and state-of-the-art generalization ability to unknown forgery methods, while also being robust to image distortions.

Lin Li, Chao Chen, Lei Pan et al.

Recently, physiological signal-based biometric systems have received wide attention. Unlike traditional biometric features, physiological signals can not be easily compromised (usually unobservable to human eyes). Photoplethysmography (PPG) signal is easy to measure, making it more attractive than many other physiological signals for biometric authentication. However, with the advent of remote PPG (rPPG), unobservability has been challenged when the attacker can remotely steal the rPPG signals by monitoring the victim's face, subsequently posing a threat to PPG-based biometrics. In PPG-based biometric authentication, current attack approaches mandate the victim's PPG signal, making rPPG-based attacks neglected. In this paper, we firstly analyze the security of PPG-based biometrics, including user authentication and communication protocols. We evaluate the signal waveforms, heart rate and inter-pulse-interval information extracted by five rPPG methods, including four traditional optical computing methods (CHROM, POS, LGI, PCA) and one deep learning method (CL_rPPG). We conducted experiments on five datasets (PURE, UBFC_rPPG, UBFC_Phys, LGI_PPGI, and COHFACE) to collect a comprehensive set of results. Our empirical studies show that rPPG poses a serious threat to the authentication system. The success rate of the rPPG signal spoofing attack in the user authentication system reached 0.35. The bit hit rate is 0.6 in inter-pulse-interval-based security protocols. Further, we propose an active defence strategy to hide the physiological signals of the face to resist the attack. It reduces the success rate of rPPG spoofing attacks in user authentication to 0.05. The bit hit rate was reduced to 0.5, which is at the level of a random guess. Our strategy effectively prevents the exposure of PPG signals to protect users' sensitive physiological data.

Yang Xiang, Zhihua Wu, Weibao Gong et al.

The ever-growing model size and scale of compute have attracted increasing interests in training deep learning models over multiple nodes. However, when it comes to training on cloud clusters, especially across remote clusters, huge challenges are faced. In this work, we introduce a general framework, Nebula-I, for collaboratively training deep learning models over remote heterogeneous clusters, the connections between which are low-bandwidth wide area networks (WANs). We took natural language processing (NLP) as an example to show how Nebula-I works in different training phases that include: a) pre-training a multilingual language model using two remote clusters; and b) fine-tuning a machine translation model using knowledge distilled from pre-trained models, which run through the most popular paradigm of recent deep learning. To balance the accuracy and communication efficiency, in Nebula-I, parameter-efficient training strategies, hybrid parallel computing methods and adaptive communication acceleration techniques are jointly applied. Meanwhile, security strategies are employed to guarantee the safety, reliability and privacy in intra-cluster computation and inter-cluster communication. Nebula-I is implemented with the PaddlePaddle deep learning framework, which can support collaborative training over heterogeneous hardware, e.g. GPU and NPU. Experiments demonstrate that the proposed framework could substantially maximize the training efficiency while preserving satisfactory NLP performance. By using Nebula-I, users can run large-scale training tasks over cloud clusters with minimum developments, and the utility of existed large pre-trained models could be further promoted. We also introduced new state-of-the-art results on cross-lingual natural language inference tasks, which are generated based upon a novel learning framework and Nebula-I.

Yuan Lan, Zhen Li, Jie Sun et al.

Deep neural networks (DNNs) recently emerged as a promising tool for analyzing and solving complex differential equations arising in science and engineering applications. Alternative to traditional numerical schemes, learning-based solvers utilize the representation power of DNNs to approximate the input-output relations in an automated manner. However, the lack of physics-in-the-loop often makes it difficult to construct a neural network solver that simultaneously achieves high accuracy, low computational burden, and interpretability. In this work, focusing on a class of evolutionary PDEs characterized by having decomposable operators, we show that the classical ``operator splitting'' numerical scheme of solving these equations can be exploited to design neural network architectures. This gives rise to a learning-based PDE solver, which we name Deep Operator-Splitting Network (DOSnet). Such non-black-box network design is constructed from the physical rules and operators governing the underlying dynamics contains learnable parameters, and is thus more flexible than the standard operator splitting scheme. Once trained, it enables the fast solution of the same type of PDEs. To validate the special structure inside DOSnet, we take the linear PDEs as the benchmark and give the mathematical explanation for the weight behavior. Furthermore, to demonstrate the advantages of our new AI-enhanced PDE solver, we train and validate it on several types of operator-decomposable differential equations. We also apply DOSnet to nonlinear Schrödinger equations (NLSE) which have important applications in the signal processing for modern optical fiber transmission systems, and experimental results show that our model has better accuracy and lower computational complexity than numerical schemes and the baseline DNNs.

Shigang Liu, Di Cao, Junae Kim et al.

Recently, deep learning has demonstrated promising results in enhancing the accuracy of vulnerability detection and identifying vulnerabilities in software. However, these techniques are still vulnerable to attacks. Adversarial examples can exploit vulnerabilities within deep neural networks, posing a significant threat to system security. This study showcases the susceptibility of deep learning models to adversarial attacks, which can achieve 100% attack success rate (refer to Table 5). The proposed method, EaTVul, encompasses six stages: identification of important samples using support vector machines, identification of important features using the attention mechanism, generation of adversarial data based on these features using ChatGPT, preparation of an adversarial attack pool, selection of seed data using a fuzzy genetic algorithm, and the execution of an evasion attack. Extensive experiments demonstrate the effectiveness of EaTVul, achieving an attack success rate of more than 83% when the snippet size is greater than 2. Furthermore, in most cases with a snippet size of 4, EaTVul achieves a 100% attack success rate. The findings of this research emphasize the necessity of robust defenses against adversarial attacks in software vulnerability detection.

Sicen Liu, Xiaolong Wang, JIngcheng Du et al.

Effectively medication recommendation with complex multimorbidity conditions is a critical task in healthcare. Most existing works predicted medications based on longitudinal records, which assumed the information transmitted patterns of learning longitudinal sequence data are stable and intra-visit medical events are serialized. However, the following conditions may have been ignored: 1) A more compact encoder for intra-relationship in the intra-visit medical event is urgent; 2) Strategies for learning accurate representations of the variable longitudinal sequences of patients are different. In this paper, we proposed a novel Sample-adaptive Hierarchical medicAtion Prediction nEtwork, termed SHAPE, to tackle the above challenges in the medication recommendation task. Specifically, we design a compact intra-visit set encoder to encode the relationship in the medical event for obtaining visit-level representation and then develop an inter-visit longitudinal encoder to learn the patient-level longitudinal representation efficiently. To endow the model with the capability of modeling the variable visit length, we introduce a soft curriculum learning method to assign the difficulty of each sample automatically by the visit length. Extensive experiments on a benchmark dataset verify the superiority of our model compared with several state-of-the-art baselines.

Shaohuai Shi, Qing Yang, Yang Xiang et al.

To enable the pre-trained models to be fine-tuned with local data on edge devices without sharing data with the cloud, we design an efficient split fine-tuning (SFT) framework for edge and cloud collaborative learning. We propose three novel techniques in this framework. First, we propose a matrix decomposition-based method to compress the intermediate output of a neural network to reduce the communication volume between the edge device and the cloud server. Second, we eliminate particular links in the model without affecting the convergence performance in fine-tuning. Third, we implement our system atop PyTorch to allow users to easily extend their existing training scripts to enjoy the efficient edge and cloud collaborative learning. Experiments results on 9 NLP datasets show that our framework can reduce the communication traffic by 96 times with little impact on the model accuracy.

Sicen Liu, Xiaolong Wang, Yang Xiang et al.

Medical event prediction (MEP) is a fundamental task in the medical domain, which needs to predict medical events, including medications, diagnosis codes, laboratory tests, procedures, outcomes, and so on, according to historical medical records. The task is challenging as medical data is a type of complex time series data with heterogeneous and temporal irregular characteristics. Many machine learning methods that consider the two characteristics have been proposed for medical event prediction. However, most of them consider the two characteristics separately and ignore the correlations among different types of medical events, especially relations between historical medical events and target medical events. In this paper, we propose a novel neural network based on attention mechanism, called cross-event attention-based time-aware network (CATNet), for medical event prediction. It is a time-aware, event-aware and task-adaptive method with the following advantages: 1) modeling heterogeneous information and temporal information in a unified way and considering temporal irregular characteristics locally and globally respectively, 2) taking full advantage of correlations among different types of events via cross-event attention. Experiments on two public datasets (MIMIC-III and eICU) show CATNet can be adaptive with different MEP tasks and outperforms other state-of-the-art methods on various MEP tasks. The source code of CATNet will be released after this manuscript is accepted.

Jianfeng Lu, Yue Wu, Yang Xiang

We use the score-based transport modeling method to solve the mean-field Fokker-Planck equations, which we call MSBTM. We establish an upper bound on the time derivative of the Kullback-Leibler (KL) divergence to MSBTM numerical estimation from the exact solution, thus validates the MSBTM approach. Besides, we provide an error analysis for the algorithm. In numerical experiments, we study two types of mean-field Fokker-Planck equation and their corresponding dynamics of particles in interacting systems. The MSBTM algorithm is numerically validated through qualitative and quantitative comparison between the MSBTM solutions, the results of integrating the associated stochastic differential equation and the analytical solutions if available.

Yang Xiang, Li Fan, Chenke Yin et al.

Foundation models in language and vision benefit from a unified discrete token interface that converts raw inputs into sequences for scalable pre-training and inference. For graphs, an effective tokenizer should yield reusable discrete codes that capture both node semantics and relational structure across scales, yet prior quantization-based graph tokenizers typically combine residual vector quantization (RVQ) levels with fixed rules and often focus on a single structural view, limiting cross-task transfer. We present a hierarchical quantized tokenization framework with task-conditioned routing and dual-view token streams. It produces multi-scale codes and two synchronized sequences: a local stream that preserves node-level information and a diffusion-style multi-hop stream that summarizes connectivity. A lightweight router learns task-dependent mixtures over RVQ depths to select an appropriate granularity, while a gated cross-attention module aligns and fuses the two streams into a single token sequence without altering the downstream backbone encoder. Experiments on node classification and link prediction show consistent gains over strong quantized baselines at matched compute, with ablations verifying contributions from hierarchical quantization, adaptive routing, and fusion.

Xutao Guo, Yanwu Yang, Chenfei Ye et al.

Based on the Denoising Diffusion Probabilistic Model (DDPM), medical image segmentation can be described as a conditional image generation task, which allows to compute pixel-wise uncertainty maps of the segmentation and allows an implicit ensemble of segmentations to boost the segmentation performance. However, DDPM requires many iterative denoising steps to generate segmentations from Gaussian noise, resulting in extremely inefficient inference. To mitigate the issue, we propose a principled acceleration strategy, called pre-segmentation diffusion sampling DDPM (PD-DDPM), which is specially used for medical image segmentation. The key idea is to obtain pre-segmentation results based on a separately trained segmentation network, and construct noise predictions (non-Gaussian distribution) according to the forward diffusion rule. We can then start with noisy predictions and use fewer reverse steps to generate segmentation results. Experiments show that PD-DDPM yields better segmentation results over representative baseline methods even if the number of reverse steps is significantly reduced. Moreover, PD-DDPM is orthogonal to existing advanced segmentation models, which can be combined to further improve the segmentation performance.

Yexing Du, Youcheng Pan, Zekun Wang et al.

Multimodal Large Language Models (MLLMs) have achieved notable success in enhancing translation performance by integrating multimodal information. However, existing research primarily focuses on image-guided methods, whose applicability is constrained by the scarcity of multilingual image-text pairs. The speech modality overcomes this limitation due to its natural alignment with text and the abundance of existing speech datasets, which enable scalable language coverage. In this paper, we propose a Speech-guided Machine Translation (SMT) framework that integrates speech and text as fused inputs into an MLLM to improve translation quality. To mitigate reliance on low-resource data, we introduce a Self-Evolution Mechanism. The core components of this framework include a text-to-speech model, responsible for generating synthetic speech, and an MLLM capable of classifying synthetic speech samples and iteratively optimizing itself using positive samples. Experimental results demonstrate that our framework surpasses all existing methods on the Multi30K multimodal machine translation benchmark, achieving new state-of-the-art results. Furthermore, on general machine translation datasets, particularly the FLORES-200, it achieves average state-of-the-art performance in 108 translation directions. Ablation studies on CoVoST-2 confirms that differences between synthetic and authentic speech have negligible impact on translation quality. The code and models are released at https://github.com/yxduir/LLM-SRT.

Xin Mu, Yu Wang, Yehong Zhang et al.

Understanding the life cycle of the machine learning (ML) model is an intriguing area of research (e.g., understanding where the model comes from, how it is trained, and how it is used). This paper focuses on a novel problem within this field, namely Model Provenance (MP), which concerns the relationship between a target model and its pre-training model and aims to determine whether a source model serves as the provenance for a target model. This is an important problem that has significant implications for ensuring the security and intellectual property of machine learning models but has not received much attention in the literature. To fill in this gap, we introduce a novel concept of Model DNA which represents the unique characteristics of a machine learning model. We utilize a data-driven and model-driven representation learning method to encode the model's training data and input-output information as a compact and comprehensive representation (i.e., DNA) of the model. Using this model DNA, we develop an efficient framework for model provenance identification, which enables us to identify whether a source model is a pre-training model of a target model. We conduct evaluations on both computer vision and natural language processing tasks using various models, datasets, and scenarios to demonstrate the effectiveness of our approach in accurately identifying model provenance.

Yuxuan Han, Jialin Zeng, Yang Wang et al.

We study the stochastic contextual bandit with knapsacks (CBwK) problem, where each action, taken upon a context, not only leads to a random reward but also costs a random resource consumption in a vector form. The challenge is to maximize the total reward without violating the budget for each resource. We study this problem under a general realizability setting where the expected reward and expected cost are functions of contexts and actions in some given general function classes $\mathcal{F}$ and $\mathcal{G}$, respectively. Existing works on CBwK are restricted to the linear function class since they use UCB-type algorithms, which heavily rely on the linear form and thus are difficult to extend to general function classes. Motivated by online regression oracles that have been successfully applied to contextual bandits, we propose the first universal and optimal algorithmic framework for CBwK by reducing it to online regression. We also establish the lower regret bound to show the optimality of our algorithm for a variety of function classes.

Yexing Du, Kaiyuan Liu, Youcheng Pan et al.

Multimodal Large Language Models (MLLMs) have achieved great success in Speech-to-Text Translation (S2TT) tasks. However, current research is constrained by two key challenges: language coverage and efficiency. Most of the popular S2TT datasets are substantially English-centric, which restricts the scaling-up of MLLMs' many-to-many translation capabilities. Moreover, the inference speed of MLLMs degrades dramatically when the speech is converted into long sequences (e.g., 750 tokens). To address these limitations, we propose a Multilingual Cost-effective Accelerated Speech-to-Text Translator (MCAT) framework, which includes two innovations. First, a language scaling method that leverages curriculum learning and a data balancing strategy is introduced to extend the language coverage supported by MLLMs to 70 languages and achieve mutual translation among these languages. Second, an optimized speech adapter module is designed to reduce the length of the speech sequence to only 30 tokens. Extensive experiments were conducted on MLLMs of different scales (9B and 27B). The experimental results demonstrate that MCAT not only surpasses state-of-the-art end-to-end models on the FLEURS dataset across 70x69 directions but also enhances batch inference efficiency. This is achieved with only ~100M trainable parameters and by using only 10 hours of S2TT data per language. Furthermore, we have released MCAT as open-source to promote the development of MLLMs for robust S2TT capabilities. The code and models are released at https://github.com/yxduir/m2m-70.

Yexing Du, Kaiyuan Liu, Bihe Zhang et al.

With the rapid advancement of Multimodal Large Language Models (MLLMs), their potential has garnered significant attention in Chinese Classical Studies (CCS). While existing research has primarily focused on text and visual modalities, the audio corpus within this domain remains largely underexplored. To bridge this gap, we propose the Multi-task Classical Chinese Literary Genre Audio Corpus (MCGA). It encompasses a diverse range of literary genres across six tasks: Automatic Speech Recognition (ASR), Speech-to-Text Translation (S2TT), Speech Emotion Captioning (SEC), Spoken Question Answering (SQA), Speech Understanding (SU), and Speech Reasoning (SR). Through the evaluation of ten MLLMs, our experimental results demonstrate that current models still face substantial challenges when processed on the MCGA test set. Furthermore, we introduce an evaluation metric for SEC and a metric to measure the consistency between the speech and text capabilities of MLLMs. We release MCGA and our code to the public to facilitate the development of MLLMs with more robust multidimensional audio capabilities in CCS. MCGA Corpus: https://github.com/yxduir/MCGA

Xiaowen Fu, Bingxin Wang, Xinzhou Guo et al.

Recently, multimodal electroencephalogram (EEG) learning has shown great promise in disease detection. At the same time, ensuring privacy in clinical studies has become increasingly crucial due to legal and ethical concerns. One widely adopted scheme for privacy protection is differential privacy (DP) because of its clear interpretation and ease of implementation. Although numerous methods have been proposed under DP, it has not been extensively studied for multimodal EEG data due to the complexities of models and signal data considered there. In this paper, we propose a novel Differentially Private Multimodal Laplacian Dropout (DP-MLD) scheme for multimodal EEG learning. Our approach proposes a novel multimodal representative learning model that processes EEG data by language models as text and other modal data by vision transformers as images, incorporating well-designed cross-attention mechanisms to effectively extract and integrate cross-modal features. To achieve DP, we design a novel adaptive feature-level Laplacian dropout scheme, where randomness allocation and performance are dynamically optimized within given privacy budgets. In the experiment on an open-source multimodal dataset of Freezing of Gait (FoG) in Parkinson's Disease (PD), our proposed method demonstrates an approximate 4\% improvement in classification accuracy, and achieves state-of-the-art performance in multimodal EEG learning under DP.

Yang Xiang, Zhuo Chen, Nicolo Ronchi et al.

InGaZnO (IGZO) channel FeFETs have attracted notable interest thanks to their advances in endurance. This work evaluates the viability of NOR-type IGZO FeFETs for readcentric AI inference workloads via design-technology cooptimization (DTCO). We demonstrate the cross-node bitcell footprint scalability of back-end-of-line (BEOL) IGZO FeFETs capable of delivering 10-A SRAM-equivalent area (0.016 um2) with 7-nm ground rules and reaching sub-5 ns random access latency despite writability challenges. We further identify the sensing margin penalty in NOR FeFET arrays arising from sneak current associated with the negative program-state Vt, which requires positive-Vt engineering in order to eliminate the unwanted negative voltage read inhibition - for example, by ferroelectric layer thinning. Last but not least, we elucidate the read margin implications on 3D FeNOR for storage-class memories (SCMs), with the 3D stacking density limited by additional sneak current from neighbor channel shunting.

Nguyen Linh Bao Nguyen, Wanlun Ma, Viet Vo et al.

Retrieval-augmented generation (RAG) has become central to large language model (LLM) deployments, grounding responses in enterprise or proprietary data to reduce hallucinations. However, this design introduces a new privacy risk: model outputs may signal the presence of specific documents in the retrieval corpus, enabling membership inference attacks (MIAs) that leak sensitive information. Existing MIAs are feasible, but they often rely on easily detected templated queries or require many non-templated yet costly and repetitive queries, limiting practicality. We ask: Can an adversary launch a limited-budget, surrogate-free, stealthy, and defense-agnostic membership inference attack using non-templated queries? We present MEntA (Membership Entailment Attack), a query-efficient MIA that leverages natural-language entailment to maximize information gained per query. By asking low-cost, broad, information-seeking questions and measuring entailment between model responses and candidate documents, MEntA eliminates the need for costly shadow models and large query budgets. Across NFCorpus, SCIDOCS, and TREC-COVID, MEntA achieves up to 0.991 AUC with only 5 queries, outperforming prior methods by 0.20 to 0.50 AUC under equivalent conditions. It remains effective under state-of-the-art (SOTA) RAG defenses, while current detectors either miss MEntA or flag benign queries at high rates. Regarding cost, MEntA reduces total attack cost by up to 65 $\times$ lower compared to SOTA attacks under the same attack setting. Our findings expose the feasibility of realistic, low-cost privacy leakage in RAG systems and highlight the urgent need for privacy-aware retrieval and defense mechanisms.

Yanwu Yang, Xutao Guo, Zhikai Chang et al.

Multi-modal neuroimaging technology has greatlly facilitated the efficiency and diagnosis accuracy, which provides complementary information in discovering objective disease biomarkers. Conventional deep learning methods, e.g. convolutional neural networks, overlook relationships between nodes and fail to capture topological properties in graphs. Graph neural networks have been proven to be of great importance in modeling brain connectome networks and relating disease-specific patterns. However, most existing graph methods explicitly require known graph structures, which are not available in the sophisticated brain system. Especially in heterogeneous multi-modal brain networks, there exists a great challenge to model interactions among brain regions in consideration of inter-modal dependencies. In this study, we propose a Multi-modal Dynamic Graph Convolution Network (MDGCN) for structural and functional brain network learning. Our method benefits from modeling inter-modal representations and relating attentive multi-model associations into dynamic graphs with a compositional correspondence matrix. Moreover, a bilateral graph convolution layer is proposed to aggregate multi-modal representations in terms of multi-modal associations. Extensive experiments on three datasets demonstrate the superiority of our proposed method in terms of disease classification, with the accuracy of 90.4%, 85.9% and 98.3% in predicting Mild Cognitive Impairment (MCI), Parkinson's disease (PD), and schizophrenia (SCHZ) respectively. Furthermore, our statistical evaluations on the correspondence matrix exhibit a high correspondence with previous evidence of biomarkers.

Jiaheng Wei, Yanjun Zhang, He Zhang et al.

Federated learning (FL) increasingly needs machine unlearning to comply with privacy regulations. However, existing federated unlearning approaches may overlook the overlapping information between the unlearning and remaining data, leading to ineffective unlearning and unfairness between clients. In this work, we revisit federated unlearning through the lens of memorization. We argue that unlearning should mainly remove the unique memorized information attributable to the data to be forgotten, while preserving overlapping patterns that are also supported by the remaining data. Specifically, we propose Grouped Memorization Evaluation, an example-level metric that separates memorized knowledge from overlapping knowledge. Building on this metric, we introduce Federated Memorization Pruning (FedMemPrune), a pruning-based unlearning approach that resets redundant parameters responsible for memorization. Extensive experiments show that FedMemPrune closely matches retraining-based unlearning baselines while more effectively eliminating memorization than existing federated unlearning algorithms, yielding strong unlearning performance without sacrificing the utility of retained knowledge.

Andong Chen, Lianzhang Lou, Kehai Chen et al.

Different from the traditional translation tasks, classical Chinese poetry translation requires both adequacy and fluency in translating culturally and historically significant content and linguistic poetic elegance. Large language models (LLMs) with impressive multilingual capabilities may bring a ray of hope to achieve this extreme translation demand. This paper first introduces a suitable benchmark (PoetMT) where each Chinese poetry has a recognized elegant translation. Meanwhile, we propose a new metric based on GPT-4 to evaluate the extent to which current LLMs can meet these demands. Our empirical evaluation reveals that the existing LLMs fall short in the challenging task. Hence, we propose a Retrieval-Augmented Machine Translation (RAT) method which incorporates knowledge related to classical poetry for advancing the translation of Chinese Poetry in LLMs. Experimental results show that RAT consistently outperforms all comparison methods regarding wildly used BLEU, COMET, BLEURT, our proposed metric, and human evaluation.

Yanwu Yang, Xutao Guo, Zhikai Chang et al.

The brain age has been proven to be a phenotype of relevance to cognitive performance and brain disease. Achieving accurate brain age prediction is an essential prerequisite for optimizing the predicted brain-age difference as a biomarker. As a comprehensive biological characteristic, the brain age is hard to be exploited accurately with models using feature engineering and local processing such as local convolution and recurrent operations that process one local neighborhood at a time. Instead, Vision Transformers learn global attentive interaction of patch tokens, introducing less inductive bias and modeling long-range dependencies. In terms of this, we proposed a novel network for learning brain age interpreting with global and local dependencies, where the corresponding representations are captured by Successive Permuted Transformer (SPT) and convolution blocks. The SPT brings computation efficiency and locates the 3D spatial information indirectly via continuously encoding 2D slices from different views. Finally, we collect a large cohort of 22645 subjects with ages ranging from 14 to 97 and our network performed the best among a series of deep learning methods, yielding a mean absolute error (MAE) of 2.855 in validation set, and 2.911 in an independent test set.

Zelin Li, Kehai Chen, Lemao Liu et al.

With the great advancements in large language models (LLMs), adversarial attacks against LLMs have recently attracted increasing attention. We found that pre-existing adversarial attack methodologies exhibit limited transferability and are notably inefficient, particularly when applied to LLMs. In this paper, we analyze the core mechanisms of previous predominant adversarial attack methods, revealing that 1) the distributions of importance score differ markedly among victim models, restricting the transferability; 2) the sequential attack processes induces substantial time overheads. Based on the above two insights, we introduce a new scheme, named TF-Attack, for Transferable and Fast adversarial attacks on LLMs. TF-Attack employs an external LLM as a third-party overseer rather than the victim model to identify critical units within sentences. Moreover, TF-Attack introduces the concept of Importance Level, which allows for parallel substitutions of attacks. We conduct extensive experiments on 6 widely adopted benchmarks, evaluating the proposed method through both automatic and human metrics. Results show that our method consistently surpasses previous methods in transferability and delivers significant speed improvements, up to 20 times faster than earlier attack strategies.

Hongbin Zhang, Chaozheng Wang, Kehai Chen et al.

On-policy self-distillation (OPSD) is an emerging LLM post-training paradigm in which the model serves as its own teacher: conditioned on privileged information such as a reference trace or hint, the same policy provides dense token-level supervision on its own rollouts. However, recent studies show that OPSD degrades complex reasoning by suppressing predictive uncertainty, which supports exploration and hypothesis revision. Our token-level analysis shows that this failure arises from applying a uniform direction of teacher supervision across tokens with different uncertainty levels: conformity to the privileged self-teacher suppresses exploration at high entropy, while deviation from the teacher degrades step accuracy at low entropy. Accordingly, we propose \textbf{Direction-Adaptive Self-Distillation} (\textbf{DASD}), which reframes privileged self-distillation from uniform teacher imitation into entropy-routed directional supervision: high-entropy tokens are pushed away from the privileged teacher to preserve exploration, while low-entropy tokens are pulled toward the teacher to stabilize step-level execution. Across six mathematical reasoning benchmarks, DASD achieves the best macro Avg@16 over strong RLVR and self-distillation baselines. Pass@$k$, reasoning-health, and generalization analyses show that these average gains come from preserving exploration without sacrificing step-level execution.

Yahong Yang, Yang Xiang

In this paper, we establish a neural network to approximate functionals, which are maps from infinite dimensional spaces to finite dimensional spaces. The approximation error of the neural network is $O(1/\sqrt{m})$ where $m$ is the size of networks, which overcomes the curse of dimensionality. The key idea of the approximation is to define a Barron spectral space of functionals.

Yu Zhang, Chuyang Sun, Kehai Chen et al.

Large Vision-Language Models (LVLMs) still struggle with vision hallucination, where generated responses are inconsistent with the visual input. Existing methods either rely on large-scale annotated data for fine-tuning, which incurs massive computational overhead, or employ static post-hoc strategies that overlook the dynamic nature of hallucination emergence. To address these, we introduce a new self-rewarding framework, enabling dynamic hallucination mitigation at inference time without external supervision. On the empirical side, we reveal that visual hallucination exhibits phase-wise dynamic patterns, peaking at the onset of each semantic phase. Drawing on these insights, we propose \textbf{PSRD} (\textbf{Phase-wise \textbf{S}elf-\textbf{R}eward \textbf{D}ecoding) for online hallucination correction guided by phase-wise self-reward signals. To reduce the cost of repeated self-evaluation during decoding, we distill the hallucination guidance signal from LVLMs into a lightweight reward model. The reward model subsequently provides on-the-fly guidance for targeted intervention during the decoding process, enabling precise hallucination suppression. The proposed PSRD significantly reduces the hallucination rate of LLaVA-1.5-7B by 50.0% and consistently outperforms existing post-hoc methods across five hallucination evaluation benchmarks for four LVLMs. Further analysis confirms that PSRD effectively mitigates hallucination propagation and achieves a highly controllable trade-off between strong performance and inference efficiency.

Chuqi Chen, Yue Wu, Yang Xiang

In this paper, we propose a novel approach to generative modeling using a loss function based on elastic interaction energy (EIE), which is inspired by the elastic interaction between defects in crystals. The utilization of the EIE-based metric presents several advantages, including its long range property that enables consideration of global information in the distribution. Moreover, its inclusion of a self-interaction term helps to prevent mode collapse and captures all modes of distribution. To overcome the difficulty of the relatively scattered distribution of high-dimensional data, we first map the data into a latent feature space and approximate the feature distribution instead of the data distribution. We adopt the GAN framework and replace the discriminator with a feature transformation network to map the data into a latent space. We also add a stabilizing term to the loss of the feature transformation network, which effectively addresses the issue of unstable training in GAN-based algorithms. Experimental results on popular datasets, such as MNIST, FashionMNIST, CIFAR-10, and CelebA, demonstrate that our EIEG GAN model can mitigate mode collapse, enhance stability, and improve model performance.

Xuan Tan, Xun Gong, Yang Xiang

Contrastive Language-Image Pre-Training (CLIP) model excels in traditional person re-identification (ReID) tasks due to its inherent advantage in generating textual descriptions for pedestrian images. However, applying CLIP directly to intra-camera supervised person re-identification (ICS ReID) presents challenges. ICS ReID requires independent identity labeling within each camera, without associations across cameras. This limits the effectiveness of text-based enhancements. To address this, we propose a novel framework called CLIP-based Camera-Agnostic Feature Learning (CCAFL) for ICS ReID. Accordingly, two custom modules are designed to guide the model to actively learn camera-agnostic pedestrian features: Intra-Camera Discriminative Learning (ICDL) and Inter-Camera Adversarial Learning (ICAL). Specifically, we first establish learnable textual prompts for intra-camera pedestrian images to obtain crucial semantic supervision signals for subsequent intra- and inter-camera learning. Then, we design ICDL to increase inter-class variation by considering the hard positive and hard negative samples within each camera, thereby learning intra-camera finer-grained pedestrian features. Additionally, we propose ICAL to reduce inter-camera pedestrian feature discrepancies by penalizing the model's ability to predict the camera from which a pedestrian image originates, thus enhancing the model's capability to recognize pedestrians from different viewpoints. Extensive experiments on popular ReID datasets demonstrate the effectiveness of our approach. Especially, on the challenging MSMT17 dataset, we arrive at 58.9\% in terms of mAP accuracy, surpassing state-of-the-art methods by 7.6\%. Code will be available at: https://github.com/Trangle12/CCAFL.

Yang Xiang, Yixin Ji, Ruotao Xu et al.

Large reasoning models (LRMs) have achieved remarkable performance in complex reasoning tasks, driven by their powerful inference-time scaling capability. However, LRMs often suffer from overthinking, which results in substantial computational redundancy and significantly reduces efficiency. Early-exit methods aim to mitigate this issue by terminating reasoning once sufficient evidence has been generated, yet existing approaches mostly rely on handcrafted or empirical indicators that are unreliable and impractical. In this work, we introduce Dynamic Thought Sufficiency in Reasoning (DTSR), a novel framework for efficient reasoning that enables the model to dynamically assess the sufficiency of its chain-of-thought (CoT) and determine the optimal point for early exit. Inspired by human metacognition, DTSR operates in two stages: (1) Reflection Signal Monitoring, which identifies reflection signals as potential cues for early exit, and (2) Thought Sufficiency Check, which evaluates whether the current CoT is sufficient to derive the final answer. Experimental results on the Qwen3 models show that DTSR reduces reasoning length by 28.9%-34.9% with minimal performance loss, effectively mitigating overthinking. We further discuss overconfidence in LRMs and self-evaluation paradigms, providing valuable insights for early-exit reasoning.

Chuqi Chen, Yue Wu, Yang Xiang

In this paper, we investigate the training process of generative networks that use a type of probability density distance named particle-based distance as the objective function, e.g. MMD GAN, Cramér GAN, EIEG GAN. However, these GANs often suffer from the problem of unstable training. In this paper, we analyze the stability of the training process of these GANs from the perspective of probability density dynamics. In our framework, we regard the discriminator $D$ in these GANs as a feature transformation mapping that maps high dimensional data into a feature space, while the generator $G$ maps random variables to samples that resemble real data in terms of feature space. This perspective enables us to perform stability analysis for the training of GANs using the Wasserstein gradient flow of the probability density function. We find that the training process of the discriminator is usually unstable due to the formulation of $\min_G \max_D E(G, D)$ in GANs. To address this issue, we add a stabilizing term in the discriminator loss function. We conduct experiments to validate our stability analysis and stabilizing method.

Yu Zhang, Mufan Xu, Xuefeng Bai et al.

Modality following serves as the capacity of multimodal large language models (MLLMs) to selectively utilize multimodal contexts based on user instructions. It is fundamental to ensuring safety and reliability in real-world deployments. However, the underlying mechanisms governing this decision-making process remain poorly understood. In this paper, we investigate its working mechanism through an information flow lens. Our findings reveal that instruction tokens function as structural anchors for modality arbitration: Shallow attention layers perform non-selective information transfer, routing multimodal cues to these anchors as a latent buffer; Modality competition is resolved within deep attention layers guided by the instruction intent, while MLP layers exhibit semantic inertia, acting as an adversarial force. Furthermore, we identify a sparse set of specialized attention heads that drive this arbitration. Causal interventions demonstrate that manipulating a mere $5\%$ of these critical heads can decrease the modality-following ratio by $60\%$ through blocking, or increase it by $60\%$ through targeted amplification of failed samples. Our work provides a substantial step toward model transparency and offers a principled framework for the orchestration of multimodal information in MLLMs.

Yixin Ji, Yang Xiang, Juntao Li et al.

As large language models (LLMs) are widely applied across various fields, model compression has become increasingly crucial for reducing costs and improving inference efficiency. Post-training pruning is a promising method that does not require resource-intensive iterative training and only needs a small amount of calibration data to assess the importance of parameters. Recent research has enhanced post-training pruning from different aspects but few of them systematically explore the effects of calibration data, and it is unclear if there exist better calibration data construction strategies. We fill this blank and surprisingly observe that calibration data is also crucial to post-training pruning, especially for high sparsity. Through controlled experiments on important influence factors of calibration data, including the pruning settings, the amount of data, and its similarity with pre-training data, we observe that a small size of data is adequate, and more similar data to its pre-training stage can yield better performance. As pre-training data is usually inaccessible for advanced LLMs, we further provide a self-generating calibration data synthesis strategy to construct feasible calibration data. Experimental results on recent strong open-source LLMs (e.g., DCLM, and LLaMA-3) show that the proposed strategy can enhance the performance of strong pruning methods (e.g., Wanda, DSnoT, OWL) by a large margin (up to $2.68\%$). Code is available at https://github.com/Dereck0602/calibration_data.

Shuoran Jiang, Qingcai Chen, Yang Xiang et al.

The excellent generalization, contextual learning, and emergence abilities in the pre-trained large models (PLMs) handle specific tasks without direct training data, making them the better foundation models in the adversarial domain adaptation (ADA) methods to transfer knowledge learned from the source domain to target domains. However, existing ADA methods fail to account for the confounder properly, which is the root cause of the source data distribution that differs from the target domains. This study proposes an adversarial domain adaptation with confounder balancing for PLMs fine-tuning (ADA-CBF). The ADA-CBF includes a PLM as the foundation model for a feature extractor, a domain classifier and a confounder classifier, and they are jointly trained with an adversarial loss. This loss is designed to improve the domain-invariant representation learning by diluting the discrimination in the domain classifier. At the same time, the adversarial loss also balances the confounder distribution among source and unmeasured domains in training. Compared to existing ADA methods, ADA-CBF can correctly identify confounders in domain-invariant features, thereby eliminating the confounder biases in the extracted features from PLMs. The confounder classifier in ADA-CBF is designed as a plug-and-play and can be applied in the confounder measurable, unmeasurable, or partially measurable environments. Empirical results on natural language processing and computer vision downstream tasks show that ADA-CBF outperforms the newest GPT-4, LLaMA2, ViT and ADA methods.

Tianyu Jin, Georg Maierhofer, Katharina Schratz et al.

The use of implicit time-stepping schemes for the numerical approximation of solutions to stiff nonlinear time-evolution equations brings well-known advantages including, typically, better stability behaviour and corresponding support of larger time steps, and better structure preservation properties. However, this comes at the price of having to solve a nonlinear equation at every time step of the numerical scheme. In this work, we propose a novel deep learning based hybrid Newton's method to accelerate this solution of the nonlinear time step system for stiff time-evolution nonlinear equations. We propose a targeted learning strategy which facilitates robust unsupervised learning in an offline phase and provides a highly efficient initialisation for the Newton iteration leading to consistent acceleration of Newton's method. A quantifiable rate of improvement in Newton's method achieved by improved initialisation is provided and we analyse the upper bound of the generalisation error of our unsupervised learning strategy. These theoretical results are supported by extensive numerical results, demonstrating the efficiency of our proposed neural hybrid solver both in one- and two-dimensional cases.

Yang Xiang, Martin Cole, Zhengwu Zhang

Cortical surface registration is often driven by local geometric descriptors (e.g., sulcal depth and curvature). While this approach achieves geometric correspondence, it neglects the long-range wiring constraints imposed by white-matter anatomy. Diffusion MRI tractography offers these crucial constraints; however, prior connectivity-informed pipelines typically align precomputed connectivity matrices, making the optimization highly sensitive to connectivity estimation and its resolution. In this paper, we introduce a novel connectivity-based surface registration method that aligns cortical surfaces by operating directly on white-matter fiber-tract endpoints. We model tract endpoints as a point cloud on the product manifold $Ω\times Ω$, where $Ω$ represents the spherical domain of the inflated cortical hemispheres. Our alignment method iteratively (i) computes a small diffeomorphic warp for $Ω$ by minimizing connectivity mismatch, and (ii) updates the endpoints based on this warp. The method relies on a geometric framework that ensures output warps are diffeomorphisms and has a final goal that optimizes the matching of well-known fiber bundles. Experiments on Human Connectome Project (HCP) data demonstrate improved tract-level correspondence, achieving higher connectivity-level overlap coefficients on major fiber bundles and stronger robustness across grid resolutions for $Ω$ compared to state-of-the-art methods such as ENCORE and MSMAll.

Cong Wang, Changfeng Gao, Yang Xiang et al.

Differentiable reinforcement learning (RL) frameworks like DiffRO offer a powerful approach for controllable text-to-speech (TTS), but are vulnerable to reward hacking, particularly for nuanced tasks like emotion control. The policy model can exploit a vanilla Reward Model (RM) by generating acoustic artifacts to achieve spurious rewards, but at the cost of degrading perceptual quality. To address this, we propose Robust Reward Policy Optimization (RRPO), a novel framework that employs a hybrid regularization scheme. This scheme develops a robust RM whose reward signal is more reliably aligned with human perception, compelling the policy to abandon detrimental shortcuts and instead learn the complex features of genuine emotions. Our ablation study confirms the enhanced robustness of our RM, as evidenced by its strong cross-lingual generalization. The subjective evaluation demonstrates that this robust RM effectively mitigates reward hacking, leading to significant improvements in both emotional expressiveness and naturalness over all baselines. Demo page: https://lrwinr.github.io/RRPO-CosyVoice.

Yingjie Zhu, Xuefeng Bai, Kehai Chen et al.

Reasoning over table images remains challenging for Large Vision-Language Models (LVLMs) due to complex layouts and tightly coupled structure-content information. Existing solutions often depend on expensive supervised training, reinforcement learning, or external tools, limiting efficiency and scalability. This work addresses a key question: how to adapt LVLMs to table reasoning with minimal annotation and no external tools? Specifically, we first introduce DiSCo, a Disentangled Structure-Content alignment framework that explicitly separates structural abstraction from semantic grounding during multimodal alignment, efficiently adapting LVLMs to tables structures. Building on DiSCo, we further present Table-GLS, a Global-to-Local Structure-guided reasoning framework that performs table reasoning via structured exploration and evidence-grounded inference. Extensive experiments across diverse benchmarks demonstrate that our framework efficiently enhances LVLM's table understanding and reasoning capabilities, particularly generalizing to unseen table structures.

Le Liu, Zhiming Li, Jianzhi Yan et al.

Despite its success, existing in-context learning (ICL) relies on in-domain expert demonstrations, limiting its applicability when expert annotations are scarce. We posit that different domains may share underlying reasoning structures, enabling source-domain demonstrations to improve target-domain inference despite semantic mismatch. To test this hypothesis, we conduct a comprehensive empirical study of different retrieval methods to validate the feasibility of achieving cross-domain knowledge transfer under the in-context learning setting. Our results demonstrate conditional positive transfer in cross-domain ICL. We identify a clear example absorption threshold: beyond it, positive transfer becomes more likely, and additional demonstrations yield larger gains. Further analysis suggests that these gains stem from reasoning structure repair by retrieved cross-domain examples, rather than semantic cues. Overall, our study validates the feasibility of leveraging cross-domain knowledge transfer to improve cross-domain ICL performance, motivating the community to explore designing more effective retrieval approaches for this novel direction.\footnote{Our implementation is available at https://github.com/littlelaska/ICL-TF4LR}

Jianzhi Yan, Zhiming Li, Le Liu et al.

Large language models (LLMs) have made notable progress in logical reasoning, yet still fall short of human-level performance. Current boosting strategies rely on expert-crafted in-domain demonstrations, limiting their applicability in expertise-scarce domains, such as specialized mathematical reasoning, formal logic, or legal analysis. In this work, we demonstrate the feasibility of leveraging cross-domain demonstrating examples to boost the LLMs' reasoning performance. Despite substantial domain differences, many reusable implicit logical structures are shared across domains. In order to effectively retrieve cross-domain examples for unseen domains under investigation, in this work, we further propose an effective retrieval method, called domain-invariant neurons-based retrieval (\textbf{DIN-Retrieval}). Concisely, DIN-Retrieval first summarizes a hidden representation that is universal across different domains. Then, during the inference stage, we use the DIN vector to retrieve structurally compatible cross-domain demonstrations for the in-context learning. Experimental results in multiple settings for the transfer of mathematical and logical reasoning demonstrate that our method achieves an average improvement of 1.8 over the state-of-the-art methods \footnote{Our implementation is available at https://github.com/Leon221220/DIN-Retrieval}.

Yaxin Feng, Yuan Lan, Luchan Zhang et al.

Urban segmentation and lane detection are two important tasks for traffic scene perception. Accuracy and fast inference speed of visual perception are crucial for autonomous driving safety. Fine and complex geometric objects are the most challenging but important recognition targets in traffic scene, such as pedestrians, traffic signs and lanes. In this paper, a simple and efficient topology-aware energy loss function-based network training strategy named EIEGSeg is proposed. EIEGSeg is designed for multi-class segmentation on real-time traffic scene perception. To be specific, the convolutional neural network (CNN) extracts image features and produces multiple outputs, and the elastic interaction energy loss function (EIEL) drives the predictions moving toward the ground truth until they are completely overlapped. Our strategy performs well especially on fine-scale structure, \textit{i.e.} small or irregularly shaped objects can be identified more accurately, and discontinuity issues on slender objects can be improved. We quantitatively and qualitatively analyze our method on three traffic datasets, including urban scene segmentation data Cityscapes and lane detection data TuSimple and CULane. Our results demonstrate that EIEGSeg consistently improves the performance, especially on real-time, lightweight networks that are better suited for autonomous driving.

Yue Wu, Tianyu Jin, Chuqi Chen et al.

We propose an energy stable network (EStable-Net) for solving gradient flow equations. The EStable-Net enables decreasing of a discrete energy along the neural network, which is consistent with the property of the gradient flow equation. The architecture of the neural network EStable-Net is based on the block network structure (Autoflow) in which output of each block can be interpreted as an intermediate state of the evolution process of the equation, and the energy stable property is incorporated in each block, which is easily generalized to include other physical and/or numerical properties. Our EStable-Net is a supervised learning network approach for solving evolution equations which does not depend on the convergence of time step goes to 0, and can be applied generally even when only data is available but the equation is unknown. We also propose a training strategy for supervised learning that employs data of the evolution stages with different nature. The EStable-Net is validated by numerical experimental results based on the Allen-Cahn equation and the Cahn-Hilliard equation in two dimensions.

Yexing Du, Kaiyuan Liu, Youcheng Pan et al.

As Large Language Models (LLMs) are increasingly popularized in the multilingual world, ensuring hallucination-free factuality becomes markedly crucial. However, existing benchmarks for evaluating the reliability of Multimodal Large Language Models (MLLMs) predominantly focus on textual or visual modalities with a primary emphasis on English, which creates a gap in evaluation when processing multilingual input, especially in speech. To bridge this gap, we propose a novel \textbf{C}ross-lingual and \textbf{C}ross-modal \textbf{F}actuality benchmark (\textbf{CCFQA}). Specifically, the CCFQA benchmark contains parallel speech-text factual questions across 8 languages, designed to systematically evaluate MLLMs' cross-lingual and cross-modal factuality capabilities. Our experimental results demonstrate that current MLLMs still face substantial challenges on the CCFQA benchmark. Furthermore, we propose a few-shot transfer learning strategy that effectively transfers the Question Answering (QA) capabilities of LLMs in English to multilingual Spoken Question Answering (SQA) tasks, achieving competitive performance with GPT-4o-mini-Audio using just 5-shot training. We release CCFQA as a foundational research resource to promote the development of MLLMs with more robust and reliable speech understanding capabilities. Our code and dataset are available at https://github.com/yxduir/ccfqa.

Ningyu Yan, Shuai Wang, Xing Shen et al.

Visuotactile sensors are indispensable for contact-rich robotic manipulation tasks. However, policy learning with tactile feedback in simulation, especially for online reinforcement learning (RL), remains a critical challenge, as it demands a delicate balance between physics fidelity and computational efficiency. To address this challenge, we present Tac2Real, a lightweight visuotactile simulation framework designed to enable efficient online RL training. Tac2Real integrates the Preconditioned Nonlinear Conjugate Gradient Incremental Potential Contact (PNCG-IPC) method with a multi-node, multi-GPU high-throughput parallel simulation architecture, which can generate marker displacement fields at interactive rates. Meanwhile, we propose a systematic approach, TacAlign, to narrow both structured and stochastic sources of domain gap, ensuring a reliable zero-shot sim-to-real transfer. We further evaluate Tac2Real on the contact-rich peg insertion task. The zero-shot transfer results achieve a high success rate in the real-world scenario, verifying the effectiveness and robustness of our framework. The project page is: https://ningyurichard.github.io/tac2real-project-page/