Yi Tang

Sondos Mahmoud Bsharat, Jiacheng Liu, Xiaohan Zhao et al.

As AI writing assistants become increasingly integrated into real-world drafting and revision workflows, many documents are no longer purely human-written or AI-generated, but instead result from progressive human-AI co-editing. However, existing AI-text detection benchmarks largely focus on final outputs and provide limited understanding of how AI authorship signals emerge, accumulate, or disappear throughout the revision process. We introduce OpAI-Bench, an operation-guided benchmark for studying progressive human-to-AI text transformation across document, sentence, token, and span granularities. Starting from human-written documents, OpAI-Bench constructs nine sequentially revised versions for each sample under predefined AI coverage levels and five representative AI edit operations, covering four domains while preserving complete authorship provenance at multiple granularities. The benchmark supports comprehensive evaluation with 8 document-level detectors, 7 sentence-level detectors, and 2 fine-grained token/span-level detectors. Experiments reveal that AI-text detectability is governed not only by the proportion of AI-edited content, but also by edit operation, domain, and cumulative revision history. Interestingly, we notice that mixed-authorship intermediate versions are often harder to detect than both fully human and heavily AI-edited endpoints, exposing non-monotonic detection patterns missed by existing benchmarks. OpAI-Bench provides a controlled testbed for analyzing whether, when, and how AI-assisted writing becomes detectable under realistic progressive editing scenarios. Our code and benchmark are available at https://github.com/VILA-Lab/OpAI-Bench.

Yi Tang, Takafumi Iwaguchi, Hiroshi Kawasaki

In this paper, we present an approach to image enhancement with diffusion model in underwater scenes. Our method adapts conditional denoising diffusion probabilistic models to generate the corresponding enhanced images by using the underwater images and the Gaussian noise as the inputs. Additionally, in order to improve the efficiency of the reverse process in the diffusion model, we adopt two different ways. We firstly propose a lightweight transformer-based denoising network, which can effectively promote the time of network forward per iteration. On the other hand, we introduce a skip sampling strategy to reduce the number of iterations. Besides, based on the skip sampling strategy, we propose two different non-uniform sampling methods for the sequence of the time step, namely piecewise sampling and searching with the evolutionary algorithm. Both of them are effective and can further improve performance by using the same steps against the previous uniform sampling. In the end, we conduct a relative evaluation of the widely used underwater enhancement datasets between the recent state-of-the-art methods and the proposed approach. The experimental results prove that our approach can achieve both competitive performance and high efficiency. Our code is available at \href{mailto:https://github.com/piggy2009/DM_underwater}{\color{blue}{https://github.com/piggy2009/DM\_underwater}}.

Xinyi Shang, Yi Tang, Jiacheng Cui et al.

Existing tampering detection benchmarks largely rely on object masks, which severely misalign with the true edit signal: many pixels inside a mask are untouched or only trivially modified, while subtle yet consequential edits outside the mask are treated as natural. We reformulate VLM image tampering from coarse region labels to a pixel-grounded, meaning and language-aware task. First, we introduce a taxonomy spanning edit primitives (replace/remove/splice/inpaint/attribute/colorization, etc.) and their semantic class of tampered object, linking low-level changes to high-level understanding. Second, we release a new benchmark with per-pixel tamper maps and paired category supervision to evaluate detection and classification within a unified protocol. Third, we propose a training framework and evaluation metrics that quantify pixel-level correctness with localization to assess confidence or prediction on true edit intensity, and further measure tamper meaning understanding via semantics-aware classification and natural language descriptions for the predicted regions. We also re-evaluate the existing strong segmentation/localization baselines on recent strong tamper detectors and reveal substantial over- and under-scoring using mask-only metrics, and expose failure modes on micro-edits and off-mask changes. Our framework advances the field from masks to pixels, meanings and language descriptions, establishing a rigorous standard for tamper localization, semantic classification and description. Code and benchmark data are available at https://github.com/VILA-Lab/PIXAR.

Zhenjie Xu, Wenqing Chen, Yi Tang et al.

Natural language processing (NLP) has seen remarkable advancements with the development of large language models (LLMs). Despite these advancements, LLMs often produce socially biased outputs. Recent studies have mainly addressed this problem by prompting LLMs to behave ethically, but this approach results in unacceptable performance degradation. In this paper, we propose a multi-objective approach within a multi-agent framework (MOMA) to mitigate social bias in LLMs without significantly compromising their performance. The key idea of MOMA involves deploying multiple agents to perform causal interventions on bias-related contents of the input questions, breaking the shortcut connection between these contents and the corresponding answers. Unlike traditional debiasing techniques leading to performance degradation, MOMA substantially reduces bias while maintaining accuracy in downstream tasks. Our experiments conducted on two datasets and two models demonstrate that MOMA reduces bias scores by up to 87.7%, with only a marginal performance degradation of up to 6.8% in the BBQ dataset. Additionally, it significantly enhances the multi-objective metric icat in the StereoSet dataset by up to 58.1%. Code will be made available at https://github.com/Cortantse/MOMA.

Quanfeng Xu, Yi Tang, Yumei She

With the convenient availability of remote sensing data, how to make models to interpret complex remote sensing data attracts wide attention. In remote sensing data, hyperspectral images contain spectral information and LiDAR contains elevation information. Hence, more explorations are warranted to better fuse the features of different source data. In this paper, we introduce semantic understanding to dynamically fuse data from two different sources, extract features of HSI and LiDAR through different capsule network branches and improve self-supervised loss and random rigid rotation in Canonical Capsule to a high-dimensional situation. Canonical Capsule computes the capsule decomposition of objects by permutation-equivariant attention and the process is self-supervised by training pairs of randomly rotated objects. After fusing the features of HSI and LiDAR with semantic understanding, the unsupervised extraction of spectral-spatial-elevation fusion features is achieved. With two real-world examples of HSI and LiDAR fused, the experimental results show that the proposed multi-branch high-dimensional canonical capsule algorithm can be effective for semantic understanding of HSI and LiDAR. It indicates that the model can extract HSI and LiDAR data features effectively as opposed to existing models for unsupervised extraction of multi-source RS data.

Yuxi Liu, Yunfeng Ma, Yi Tang et al.

Industrial surface defect detection (SDD) is critical for ensuring product quality and manufacturing reliability. Due to the diverse shapes and sizes of surface defects, SDD faces two main challenges: intraclass difference and interclass similarity. Existing methods primarily utilize manually designed models, which require extensive trial and error and often struggle to address both challenges effectively. To overcome this, we propose AutoNAD, an automated neural architecture design framework for SDD that jointly searches over convolutions, transformers, and multi-layer perceptrons. This hybrid design enables the model to capture both fine-grained local variations and long-range semantic context, addressing the two key challenges while reducing the cost of manual network design. To support efficient training of such a diverse search space, AutoNAD introduces a cross weight sharing strategy, which accelerates supernet convergence and improves subnet performance. Additionally, a searchable multi-level feature aggregation module (MFAM) is integrated to enhance multi-scale feature learning. Beyond detection accuracy, runtime efficiency is essential for industrial deployment. To this end, AutoNAD incorporates a latency-aware prior to guide the selection of efficient architectures. The effectiveness of AutoNAD is validated on three industrial defect datasets and further applied within a defect imaging and detection platform. Code will be available at https://github.com/Yuxi104/AutoNAD.

Zejun Zhang, Jian Wang, Qingyun Yang et al.

Accurate project localization (e.g., files and functions) for issue resolution is a critical first step in software maintenance. However, existing benchmarks for issue localization, such as SWE-Bench and LocBench, are limited. They focus predominantly on pull-request issues and code locations, ignoring other evidence and non-code files such as commits, comments, configurations, and documentation. To address this gap, we introduce MULocBench, a comprehensive dataset of 1,100 issues from 46 popular GitHub Python projects. Comparing with existing benchmarks, MULocBench offers greater diversity in issue types, root causes, location scopes, and file types, providing a more realistic testbed for evaluation. Using this benchmark, we assess the performance of state-of-the-art localization methods and five LLM-based prompting strategies. Our results reveal significant limitations in current techniques: even at the file level, performance metrics (Acc@5, F1) remain below 40%. This underscores the challenge of generalizing to realistic, multi-faceted issue resolution. To enable future research on project localization for issue resolution, we publicly release MULocBench at https://huggingface.co/datasets/somethingone/MULocBench.

Taiming Wang, Yuxia Zhang, Lin Jiang et al.

Concise and meaningful method names are crucial for program comprehension and maintenance. However, method names may become inconsistent with their corresponding implementations, causing confusion and errors. Several deep learning (DL)-based approaches have been proposed to identify such inconsistencies, with initial evaluations showing promising results. However, these evaluations typically use a balanced dataset, where the number of inconsistent and consistent names are equal. This setup, along with flawed dataset construction, leads to false positives, making reported performance less reliable in real-world scenarios, where most method names are consistent. In this paper, we present an empirical study that evaluates state-of-the-art DL-based methods for identifying inconsistent method names. We create a new benchmark by combining automatic identification from commit histories and manual developer inspections, reducing false positives. We evaluate five representative DL approaches (one retrieval-based and four generation-based) on this benchmark. Our results show that performance drops substantially when moving from the balanced dataset to the new benchmark. We further conduct quantitative and qualitative analyses to understand the strengths and weaknesses of the approaches. Retrieval-based methods perform well on simple methods and those with popular name sub-tokens but fail due to inefficient representation techniques. Generation-based methods struggle with inaccurate similarity calculations and immature name generation. Based on these findings, we propose improvements using contrastive learning and large language models (LLMs). Our study suggests that significant improvements are needed before these DL approaches can be effectively applied to real-world software systems.

Jialiang Wu, Yi Shen, Sijia Liu et al.

Despite their impressive capacities, Large language models (LLMs) often struggle with the hallucination issue of generating inaccurate or fabricated content even when they possess correct knowledge. In this paper, we extend the exploration of the correlation between hidden-state prediction changes and output factuality into a deeper, token-wise level. Based on the insights , we propose cross-layer Entropy eNhanced Decoding (END), a decoding method that mitigates hallucinations without requiring extra training. END leverages inner probability changes across layers to individually quantify the factual knowledge required for each candidate token, and adjusts the final predicting distribution to prioritize tokens with higher factuality. Experiments on both hallucination and QA benchmarks demonstrate that END significantly enhances the truthfulness and informativeness of generated content while maintaining robust QA accuracy. Moreover, our work provides a deeper perspective on understanding the correlations between inherent knowledge and output factuality.

Yi Tang, Peng Sun, Zhenglin Cheng et al.

Recent studies indicate that the denoising process in deep generative diffusion models implicitly learns and memorizes semantic information from the data distribution. These findings suggest that capturing more complex data distributions requires larger neural networks, leading to a substantial increase in computational demands, which in turn become the primary bottleneck in both training and inference of diffusion models. To this end, we introduce GMem: A Modular Approach for Ultra-Efficient Generative Models. Our approach GMem decouples the memory capacity from model and implements it as a separate, immutable memory set that preserves the essential semantic information in the data. The results are significant: GMem enhances both training, sampling efficiency, and diversity generation. This design on one hand reduces the reliance on network for memorize complex data distribution and thus enhancing both training and sampling efficiency. On ImageNet at $256 \times 256$ resolution, GMem achieves a $50\times$ training speedup compared to SiT, reaching FID $=7.66$ in fewer than $28$ epochs ($\sim 4$ hours training time), while SiT requires $1400$ epochs. Without classifier-free guidance, GMem achieves state-of-the-art (SoTA) performance FID $=1.53$ in $160$ epochs with only $\sim 20$ hours of training, outperforming LightningDiT which requires $800$ epochs and $\sim 95$ hours to attain FID $=2.17$.

Yi Tang, Kaini Wang, Yang Chen et al.

Developing general artificial intelligence (AI) systems to support endoscopic image diagnosis is an emerging research priority. Existing methods based on large-scale pretraining often lack unified coordination across tasks and struggle to handle the multi-step processes required in complex clinical workflows. While AI agents have shown promise in flexible instruction parsing and tool integration across domains, their potential in endoscopy remains underexplored. To address this gap, we propose EndoAgent, the first memory-guided agent for vision-to-decision endoscopic analysis that integrates iterative reasoning with adaptive tool selection and collaboration. Built on a dual-memory design, it enables sophisticated decision-making by ensuring logical coherence through short-term action tracking and progressively enhancing reasoning acuity through long-term experiential learning. To support diverse clinical tasks, EndoAgent integrates a suite of expert-designed tools within a unified reasoning loop. We further introduce EndoAgentBench, a benchmark of 5,709 visual question-answer pairs that assess visual understanding and language generation capabilities in realistic scenarios. Extensive experiments show that EndoAgent consistently outperforms both general and medical multimodal models, exhibiting its strong flexibility and reasoning capabilities.

Runkang Yang, Peng Sun, Xinyi Shang et al.

Data inherently possesses dual attributes: samples and targets. For targets, knowledge distillation has been widely employed to accelerate model convergence, primarily relying on teacher-generated soft target supervision. Conversely, recent advancements in data-efficient learning have emphasized sample optimization techniques, such as dataset distillation, while neglected the critical role of target. This dichotomy motivates our investigation into understanding how both sample and target collectively influence training dynamic. To address this gap, we first establish a taxonomy of existing paradigms through the lens of sample-target interactions, categorizing them into distinct sample-to-target mapping strategies. Building upon this foundation, we then propose a novel unified loss framework to assess their impact on training efficiency. Through extensive empirical studies on our proposed strategies, we comprehensively analyze how variations in target and sample types, quantities, and qualities influence model training, providing six key insights to enhance training efficacy.

Huck Bennett, Chris Peikert, Yi Tang

We show improved fine-grained hardness of two key lattice problems in the $\ell_p$ norm: Bounded Distance Decoding to within an $α$ factor of the minimum distance ($\mathrm{BDD}_{p, α}$) and the (decisional) $γ$-approximate Shortest Vector Problem ($\mathrm{SVP}_{p,γ}$), assuming variants of the Gap (Strong) Exponential Time Hypothesis (Gap-(S)ETH). Specifically, we show: 1. For all $p \in [1, \infty)$, there is no $2^{o(n)}$-time algorithm for $\mathrm{BDD}_{p, α}$ for any constant $α> α_\mathsf{kn}$, where $α_\mathsf{kn} = 2^{-c_\mathsf{kn}}$ and $c_\mathsf{kn}$ is the $\ell_2$ kissing-number constant, assuming $c_\mathsf{kn} > 0$ and that non-uniform Gap-ETH holds. 2. For all $p \in [1, \infty)$, there is no $2^{o(n)}$-time algorithm for $\mathrm{BDD}_{p, α}$ for any constant $α> α^\ddagger_p$, where $α^\ddagger_p$ is explicit and satisfies $α^\ddagger_p = 1$ for $1 \leq p \leq 2$, $α^\ddagger_p < 1$ for all $p > 2$, and $α^\ddagger_p \to 1/2$ as $p \to \infty$, unless randomized Gap-ETH is false. 3. For all $p \in [1, \infty) \setminus 2 \mathbb{Z}$ and all $C > 1$, there is no $2^{n/C}$-time algorithm for $\mathrm{BDD}_{p, α}$ for any constant $α> α^\dagger_{p, C}$, where $α^\dagger_{p, C}$ is explicit and satisfies $α^\dagger_{p, C} \to 1$ as $C \to \infty$ for any fixed $p \in [1, \infty)$, assuming $c_\mathsf{kn} > 0$ and that non-uniform Gap-SETH holds. 4. For all $p > p_0 \approx 2.1397$, $p \notin 2\mathbb{Z}$, and all $C > C_p$, there is no $2^{n/C}$-time algorithm for $\mathrm{SVP}_{p, γ}$ for some constant $γ> 1$, where $C_p > 1$ is explicit and satisfies $C_p \to 1$ as $p \to \infty$, unless randomized Gap-SETH is false.

Yi Tang, Yuanman Li, Guoliang Xing

Video salient object detection (VSOD) is an important task in many vision applications. Reliable VSOD requires to simultaneously exploit the information from both the spatial domain and the temporal domain. Most of the existing algorithms merely utilize simple fusion strategies, such as addition and concatenation, to merge the information from different domains. Despite their simplicity, such fusion strategies may introduce feature redundancy, and also fail to fully exploit the relationship between multi-level features extracted from both spatial and temporal domains. In this paper, we suggest an adaptive local-global refinement framework for VSOD. Different from previous approaches, we propose a local refinement architecture and a global one to refine the simply fused features with different scopes, which can fully explore the local dependence and the global dependence of multi-level features. In addition, to emphasize the effective information and suppress the useless one, an adaptive weighting mechanism is designed based on graph convolutional neural network (GCN). We show that our weighting methodology can further exploit the feature correlations, thus driving the network to learn more discriminative feature representation. Extensive experimental results on public video datasets demonstrate the superiority of our method over the existing ones.

Pengwei Zhan, Liming Wang, Yi Tang

Cryptographic protocols have been widely used to protect the user's privacy and avoid exposing private information. QUIC (Quick UDP Internet Connections), including the version originally designed by Google (GQUIC) and the version standardized by IETF (IQUIC), as alternatives to the traditional HTTP, demonstrate their unique transmission characteristics: based on UDP for encrypted resource transmitting, accelerating web page rendering. However, existing encrypted transmission schemes based on TCP are vulnerable to website fingerprinting (WFP) attacks, allowing adversaries to infer the users' visited websites by eavesdropping on the transmission channel. Whether GQUIC and IQUIC can effectively resist such attacks is worth investigating. In this paper, we study the vulnerabilities of GQUIC, IQUIC, and HTTPS to WFP attacks from the perspective of traffic analysis. Extensive experiments show that, in the early traffic scenario, GQUIC is the most vulnerable to WFP attacks among GQUIC, IQUIC, and HTTPS, while IQUIC is more vulnerable than HTTPS, but the vulnerability of the three protocols is similar in the normal full traffic scenario. Features transferring analysis shows that most features are transferable between protocols when on normal full traffic scenario. However, combining with the qualitative analysis of latent feature representation, we find that the transferring is inefficient when on early traffic, as GQUIC, IQUIC, and HTTPS show the significantly different magnitude of variation in the traffic distribution on early traffic. By upgrading the one-time WFP attacks to multiple WFP Top-a attacks, we find that the attack accuracy on GQUIC and IQUIC reach 95.4% and 95.5%, respectively, with only 40 packets and just using simple features, whereas reach only 60.7% when on HTTPS. We also demonstrate that the vulnerability of IQUIC is only slightly dependent on the network environment.

Rongqin Liang, Yuanman Li, Xia Li et al.

Predicting human motion behavior in a crowd is important for many applications, ranging from the natural navigation of autonomous vehicles to intelligent security systems of video surveillance. All the previous works model and predict the trajectory with a single resolution, which is rather inefficient and difficult to simultaneously exploit the long-range information (e.g., the destination of the trajectory), and the short-range information (e.g., the walking direction and speed at a certain time) of the motion behavior. In this paper, we propose a temporal pyramid network for pedestrian trajectory prediction through a squeeze modulation and a dilation modulation. Our hierarchical framework builds a feature pyramid with increasingly richer temporal information from top to bottom, which can better capture the motion behavior at various tempos. Furthermore, we propose a coarse-to-fine fusion strategy with multi-supervision. By progressively merging the top coarse features of global context to the bottom fine features of rich local context, our method can fully exploit both the long-range and short-range information of the trajectory. Experimental results on several benchmarks demonstrate the superiority of our method.

Yi Tang, Yuanman Li, Wenbin Zou

Since the wide employment of deep learning frameworks in video salient object detection, the accuracy of the recent approaches has made stunning progress. These approaches mainly adopt the sequential modules, based on optical flow or recurrent neural network (RNN), to learn robust spatiotemporal features. These modules are effective but significantly increase the computational burden of the corresponding deep models. In this paper, to simplify the network and maintain the accuracy, we present a lightweight network tailored for video salient object detection through the spatiotemporal knowledge distillation. Specifically, in the spatial aspect, we combine a saliency guidance feature embedding structure and spatial knowledge distillation to refine the spatial features. In the temporal aspect, we propose a temporal knowledge distillation strategy, which allows the network to learn the robust temporal features through the infer-frame feature encoding and distilling information from adjacent frames. The experiments on widely used video datasets (e.g., DAVIS, DAVSOD, SegTrack-V2) prove that our approach achieves competitive performance. Furthermore, without the employment of the complex sequential modules, the proposed network can obtain high efficiency with 0.01s per frame.

Joan Bruna, Oded Regev, Min Jae Song et al.

We introduce a continuous analogue of the Learning with Errors (LWE) problem, which we name CLWE. We give a polynomial-time quantum reduction from worst-case lattice problems to CLWE, showing that CLWE enjoys similar hardness guarantees to those of LWE. Alternatively, our result can also be seen as opening new avenues of (quantum) attacks on lattice problems. Our work resolves an open problem regarding the computational complexity of learning mixtures of Gaussians without separability assumptions (Diakonikolas 2016, Moitra 2018). As an additional motivation, (a slight variant of) CLWE was considered in the context of robust machine learning (Diakonikolas et al.~FOCS 2017), where hardness in the statistical query (SQ) model was shown; our work addresses the open question regarding its computational hardness (Bubeck et al.~ICML 2019).