Yunhao Liu

Shen Xu, Xiangwen Zhuge, Zhe Xu et al.

LLMs often struggle with memory-constrained deployment on consumer-grade hardware due to their massive parameter sizes. While existing solutions such as model compression and offloading improve deployment feasibility, they often suffer from substantial accuracy degradation or severe throughput bottlenecks. Recent error compensation methods recover accuracy through auxiliary LoRA-style branches, and we observe that these branches are inherently amenable to offloading: they require substantial parameter storage but access only a small subset of compensation parameters during each inference step. Motivated by this opportunity, we propose HCInfer, a heterogeneous inference system that offloads residual compensation to the CPU while executing the compressed backbone on the GPU, and further introduces an asynchronous compensation pipeline and sensitivity-aware dynamic rank allocation to hide compensation overhead and maximize accuracy recovery. Experimental results show that HCInfer achieves a maximum accuracy improvement of 5.2% on downstream tasks compared to compression model and sustaining a maximum speedup of 10.4x compared to full-precision model.

Jialuo Du, Yidong Ren, Mi Zhang et al.

Low-Power Wide-Area Networks (LPWANs) are an emerging Internet-of-Things (IoT) paradigm marked by low-power and long-distance communication. Among them, LoRa is widely deployed for its unique characteristics and open-source technology. By adopting the Chirp Spread Spectrum (CSS) modulation, LoRa enables low signal-to-noise ratio (SNR) communication. The standard LoRa demodulation method accumulates the chirp power of the whole chirp into an energy peak in the frequency domain. In this way, it can support communication even when SNR is lower than -15 dB. Beyond that, we proposed NELoRa, a neural-enhanced decoder that exploits multi-dimensional information to achieve significant SNR gain. This paper presents the dataset used to train/test NELoRa, which includes 27,329 LoRa symbols with spreading factors from 7 to 10, for further improvement of neural-enhanced LoRa demodulation. The dataset shows that NELoRa can achieve 1.84-2.35 dB SNR gain over the standard LoRa decoder. The dataset and codes can be found at https://github.com/daibiaoxuwu/NeLoRa_Dataset.

Hao Wen, Yuanchun Li, Guohong Liu et al.

Mobile task automation is an attractive technique that aims to enable voice-based hands-free user interaction with smartphones. However, existing approaches suffer from poor scalability due to the limited language understanding ability and the non-trivial manual efforts required from developers or end-users. The recent advance of large language models (LLMs) in language understanding and reasoning inspires us to rethink the problem from a model-centric perspective, where task preparation, comprehension, and execution are handled by a unified language model. In this work, we introduce AutoDroid, a mobile task automation system capable of handling arbitrary tasks on any Android application without manual efforts. The key insight is to combine the commonsense knowledge of LLMs and domain-specific knowledge of apps through automated dynamic analysis. The main components include a functionality-aware UI representation method that bridges the UI with the LLM, exploration-based memory injection techniques that augment the app-specific domain knowledge of LLM, and a multi-granularity query optimization module that reduces the cost of model inference. We integrate AutoDroid with off-the-shelf LLMs including online GPT-4/GPT-3.5 and on-device Vicuna, and evaluate its performance on a new benchmark for memory-augmented Android task automation with 158 common tasks. The results demonstrated that AutoDroid is able to precisely generate actions with an accuracy of 90.9%, and complete tasks with a success rate of 71.3%, outperforming the GPT-4-powered baselines by 36.4% and 39.7%. The demo, benchmark suites, and source code of AutoDroid will be released at url{https://autodroid-sys.github.io/}.

Xinqi Jin, Zhui Zhu, Xikai Sun et al.

Neural enhancement through super-resolution (SR) deep neural networks (DNNs) opens up new possibilities for ultra-high-definition (UHD) live streaming over existing encoding and networking infrastructure. Yet, the heavy SR DNN inference overhead leads to severe deployment challenges. To reduce the overhead, existing systems propose to apply DNN-based SR only on carefully selected anchor frames while upscaling non-anchor frames via the lightweight reusing-based SR approach. However, frame-level scheduling is coarse-grained and fails to deliver optimal efficiency. In this work, we propose Palantir, the first neural-enhanced UHD live streaming system with fine-grained patch-level scheduling. Two novel techniques are incorporated into Palantir to select the most beneficial anchor patches and support latency-sensitive UHD live streaming applications. Firstly, under the guidance of our pioneering and theoretical analysis, Palantir constructs a directed acyclic graph (DAG) for lightweight yet accurate SR quality estimation under any possible anchor patch set. Secondly, to further optimize the scheduling latency, Palantir improves parallelizability by refactoring the computation subprocedure of the estimation process into a sparse matrix-matrix multiplication operation. The evaluation results suggest that Palantir incurs a negligible scheduling latency accounting for less than 5.7% of the end-to-end latency requirement. When compared to the naive method of applying DNN-based SR on all the frames, Palantir can reduce the SR DNN inference overhead by 20 times (or 60 times) while preserving 54.0-82.6% (or 32.8-64.0%) of the quality gain. When compared to the state-of-the-art real-time frame-level scheduling strategy, Palantir can reduce the SR DNN inference overhead by 80.1% at most (and 38.4% on average) without sacrificing the video quality.

Haoyang Wang, Ruishan Guo, Pengtao Ma et al.

With the increasing complexity of mobile device applications, these devices are evolving toward high agility. This shift imposes new demands on mobile sensing, particularly in achieving high-accuracy and low-latency. Event-based vision has emerged as a disruptive paradigm, offering high temporal resolution and low latency, making it well-suited for high-accuracy and low-latency sensing tasks on high-agility platforms. However, the presence of substantial noisy events, lack of stable, persistent semantic information, and large data volume pose challenges for event-based data processing on resource-constrained mobile devices. This paper surveys the literature from 2014 to 2025 and presents a comprehensive overview of event-based mobile sensing, encompassing its fundamental principles, event \textit{abstraction} methods, \textit{algorithm} advancements, and both hardware and software \textit{acceleration} strategies. We discuss key \textit{applications} of event cameras in mobile sensing, including visual odometry, object tracking, optical flow, and 3D reconstruction, while highlighting challenges associated with event data processing, sensor fusion, and real-time deployment. Furthermore, we outline future research directions, such as improving the event camera with advanced optics, leveraging neuromorphic computing for efficient processing, and integrating bio-inspired algorithms. To support ongoing research, we provide an open-source \textit{Online Sheet} with recent developments. We hope this survey serves as a reference, facilitating the adoption of event-based vision across diverse applications.

Liuchuan Yu, Erdem Murat, Beichen Wang et al.

Egocentric video is increasingly used as a data source for robot learning, activity understanding, and embodied AI research, but collecting it at scale remains fragmented in practice: each candidate host device, such as an Android phone, iPhone, iPad, smart glasses, or extended reality (XR) headset, exposes a different SDK, a different policy on raw camera access, and different limitations on external USB cameras and on-device tracking. Synchronized ego-view and wrist-view capture is therefore typically obtained by either committing to a single proprietary platform or building one-off rigs that do not transfer across devices. To address this gap, we present EgoKit, a toolkit that exposes the same egocentric recording workflow across six heterogeneous host devices. Across all supported devices, EgoKit presents the same recording interaction and produces locally stored video with a uniform log format; on XR headsets, it additionally logs head pose and OpenXR-standard 26-joint hand tracking aligned to the video streams. The companion accessories, including two wrist cameras with mounts, a head strap, and a USB-C hub, add wrist-view capture to any supported host without custom hardware fabrication. EgoKit is available at \url{https://egokit.chuange.org/}.

Nan Zhou, Zuxin Li, Fanhang Man et al.

This paper addresses the challenge of achieving optimal Quality of Information (QoI) in non-dedicated vehicular mobile crowdsensing (NVMCS) systems. The key obstacles are the interrelated issues of sensing coverage, sensing reliability, and the dynamic participation of vehicles. To tackle these, we propose QUIDS, a QUality-informed Incentive-driven multi-agent Dispatching System, which ensures high sensing coverage and reliability under budget constraints. QUIDS introduces a novel metric, Aggregated Sensing Quality (ASQ), to quantitatively capture QoI by integrating both coverage and reliability. We also develop a Mutually Assisted Belief-aware Vehicle Dispatching algorithm that estimates sensing reliability and allocates incentives under uncertainty, further improving ASQ. Evaluation using real-world data from a metropolitan NVMCS deployment shows QUIDS improves ASQ by 38% over non-dispatching scenarios and by 10% over state-of-the-art methods. It also reduces reconstruction map errors by 39-74% across algorithms. By jointly optimizing coverage and reliability via a quality-informed incentive mechanism, QUIDS enables low-cost, high-quality urban monitoring without dedicated infrastructure, applicable to smart-city scenarios like traffic and environmental sensing.

Yunhao Liu, Yu-Ju Tsai, Kelvin C. K. Chan et al.

In this paper, we tackle the challenge of face recognition in the wild, where images often suffer from low quality and real-world distortions. Traditional heuristic approaches-either training models directly on these degraded images or their enhanced counterparts using face restoration techniques-have proven ineffective, primarily due to the degradation of facial features and the discrepancy in image domains. To overcome these issues, we propose an effective adapter for augmenting existing face recognition models trained on high-quality facial datasets. The key of our adapter is to process both the unrefined and enhanced images using two similar structures, one fixed and the other trainable. Such design can confer two benefits. First, the dual-input system minimizes the domain gap while providing varied perspectives for the face recognition model, where the enhanced image can be regarded as a complex non-linear transformation of the original one by the restoration model. Second, both two similar structures can be initialized by the pre-trained models without dropping the past knowledge. The extensive experiments in zero-shot settings show the effectiveness of our method by surpassing baselines of about 3%, 4%, and 7% in three datasets. Our code will be publicly available.

Yunhao Liu, Zian Jia, Xinyu Gao et al.

Retrieval-Augmented Generation (RAG) effectively grounds Large Language Models (LLMs) with external knowledge and is widely applied to Web-related tasks. However, its scalability is hindered by excessive context length and redundant retrievals. Recent research on soft context compression aims to address this by encoding long documents into compact embeddings, yet they often underperform non-compressed RAG due to their reliance on auto-encoder-like full-compression that forces the encoder to compress all document information regardless of relevance to the input query. In this work, we conduct an analysis on this paradigm and reveal two fundamental limitations: (I) Infeasibility, full-compression conflicts with the LLM's downstream generation behavior; and (II) Non-necessity: full-compression is unnecessary and dilutes task-relevant information density. Motivated by these insights, we introduce SeleCom, a selector-based soft compression framework for RAG that redefines the encoder's role as query-conditioned information selector. The selector is decoder-only and is trained with a massive, diverse and difficulty-graded synthetic QA dataset with curriculum learning. Extensive experiments show that SeleCom significantly outperforms existing soft compression approaches and achieves competitive or superior performance to non-compression baselines, while reducing computation and latency by 33.8%~84.6%.

Guoxuan Chi, Zheng Yang, Chenshu Wu et al. · tsinghua

Along with AIGC shines in CV and NLP, its potential in the wireless domain has also emerged in recent years. Yet, existing RF-oriented generative solutions are ill-suited for generating high-quality, time-series RF data due to limited representation capabilities. In this work, inspired by the stellar achievements of the diffusion model in CV and NLP, we adapt it to the RF domain and propose RF-Diffusion. To accommodate the unique characteristics of RF signals, we first introduce a novel Time-Frequency Diffusion theory to enhance the original diffusion model, enabling it to tap into the information within the time, frequency, and complex-valued domains of RF signals. On this basis, we propose a Hierarchical Diffusion Transformer to translate the theory into a practical generative DNN through elaborated design spanning network architecture, functional block, and complex-valued operator, making RF-Diffusion a versatile solution to generate diverse, high-quality, and time-series RF data. Performance comparison with three prevalent generative models demonstrates the RF-Diffusion's superior performance in synthesizing Wi-Fi and FMCW signals. We also showcase the versatility of RF-Diffusion in boosting Wi-Fi sensing systems and performing channel estimation in 5G networks.

Ke Ma, Jiaqi Tang, Bin Guo et al.

Proactive streaming video understanding requires Video-LLMs to decide when to respond as a video unfolds, a task where existing methods often fall short due to their implicit, query-agnostic modeling of visual evidence. We introduce Response-G1, a novel framework that establishes explicit, structured alignment between the accumulated video evidence and the query's expected response conditions via scene graphs. The framework operates in three fine-tuning-free stages: (1) online query-guided scene graph generation from streaming clips; (2) memory-based retrieval of the most semantically relevant historical scene graphs; and (3) retrieval-augmented trigger prompting for per-frame "silence/response" decisions.By grounding both evidence and conditions in a shared graph representation, Response-G1 achieves more interpretable and accurate response timing decisions. Experimental results on established benchmarks demonstrate the superiority of our method in both proactive and reactive tasks, validating the advantage of explicit scene graph modeling and retrieval in streaming video understanding.

Mu Yuan, Liekang Zeng, Guoliang Xing et al.

Autoregressive and diffusion models represent two complementary generative paradigms. Autoregressive models excel at sequential planning and constraint composition, yet struggle with tasks that require explicit spatial or physical grounding. Diffusion models, in contrast, capture rich spatial structure through high-dimensional generation, but lack the stepwise logical control needed to satisfy complex, multi-stage constraints or to reliably identify and correct errors. We introduce Collaborative Thoughts, a unified collaborative framework that enables autoregressive and diffusion models to reason and generate jointly through a closed-loop interaction. In Collaborative Thoughts, autoregressive models perform structured planning and constraint management, diffusion models instantiate these constraints as intermediate visual thoughts, and a vision-based critic module evaluates whether the visual thoughts satisfy the intended structural and physical requirements. This feedback is then used to iteratively refine subsequent planning and generation steps, mitigating error propagation across modalities. Importantly, Collaborative Thoughts uses the same collaborative loop regardless of whether the task is autoregressive question answering or diffusion-based visual generation. Through representative examples, we illustrate how Collaborative Thoughts can improve the reliability of spatial reasoning and the controllability of generation.

Haoyang Wang, Jingao Xu, Xinyu Luo et al.

For precise, efficient, and safe drone landings, ground platforms should real-time, accurately locate descending drones and guide them to designated spots. While mmWave sensing combined with cameras improves localization accuracy, the lower sampling frequency of traditional frame cameras compared to mmWave radar creates bottlenecks in system throughput. In this work, we replace the traditional frame camera with event camera, a novel sensor that harmonizes in sampling frequency with mmWave radar within the ground platform setup, and introduce mmE-Loc, a high-precision, low-latency ground localization system designed for drone landings. To fully leverage the \textit{temporal consistency} and \textit{spatial complementarity} between these modalities, we propose two innovative modules, \textit{consistency-instructed collaborative tracking} and \textit{graph-informed adaptive joint optimization}, for accurate drone measurement extraction and efficient sensor fusion. Extensive real-world experiments in landing scenarios from a leading drone delivery company demonstrate that mmE-Loc outperforms state-of-the-art methods in both localization accuracy and latency.

Hao Wen, Yifan Su, Feifei Zhang et al.

Recent advances in Large Language Models (LLMs) have been driven by test-time compute scaling - a strategy that improves reasoning by generating longer, sequential thought processes. While effective, this approach encounters a significant bottleneck as computation increases, where further computation offers only marginal performance gains. We argue this ceiling is not an inherent limit of the model's capability but a flaw in the scaling strategy itself, a phenomenon we term "Tunnel Vision", where a model's imperfect initial steps lock it into a suboptimal reasoning path. To overcome this, we introduce a new scaling paradigm: native thought parallelism. We present ParaThinker, an end-to-end framework that trains an LLM to generate multiple, diverse reasoning paths in parallel and synthesize them into a superior final answer. By exploring different lines of thoughts simultaneously, ParaThinker effectively sidesteps the Tunnel Vision issue and unlocks the model's latent reasoning potential. Our approach demonstrates that scaling compute in parallel (width) is a more effective and efficient way to superior reasoning than simply scaling sequentially (depth). On challenging reasoning benchmarks, ParaThinker achieves substantial accuracy improvements over sequential LLMs (12.3% for 1.5B and 7.5% for 7B models on average with 8 parallel paths), while adding only negligible latency overhead (7.1%). This enables smaller models to surpass much larger counterparts and establishes parallel thinking as a critical, efficient dimension for scaling future LLMs.

Hao Wen, Xinrui Wu, Yi Sun et al.

Recent advancements in Large Language Models (LLMs) have leveraged increased test-time computation to enhance reasoning capabilities, a strategy that, while effective, incurs significant latency and resource costs, limiting their applicability in real-world time-constrained or cost-sensitive scenarios. This paper introduces BudgetThinker, a novel framework designed to empower LLMs with budget-aware reasoning, enabling precise control over the length of their thought processes. We propose a methodology that periodically inserts special control tokens during inference to continuously inform the model of its remaining token budget. This approach is coupled with a comprehensive two-stage training pipeline, beginning with Supervised Fine-Tuning (SFT) to familiarize the model with budget constraints, followed by a curriculum-based Reinforcement Learning (RL) phase that utilizes a length-aware reward function to optimize for both accuracy and budget adherence. We demonstrate that BudgetThinker significantly surpasses strong baselines in maintaining performance across a variety of reasoning budgets on challenging mathematical benchmarks. Our method provides a scalable and effective solution for developing efficient and controllable LLM reasoning, making advanced models more practical for deployment in resource-constrained and real-time environments.

Yunhao Liu, Suyang Xi, Shiqi Liu et al.

Medical image analysis increasingly relies on the integration of multiple imaging modalities to capture complementary anatomical and functional information, enabling more accurate diagnosis and treatment planning. Achieving aligned feature representations across these diverse modalities is therefore important for effective multimodal analysis. While contrastive language-image pre-training (CLIP) and its variant have enabled image-text alignments, they require explicitly paired data between arbitrary two modalities, which is difficult to acquire in medical contexts. To address the gap, we present Multimodal Medical Image Binding with Text (M\textsuperscript{3}Bind), a novel pre-training framework that enables seamless alignment of multiple medical imaging modalities through a shared text representation space without requiring explicit paired data between any two medical image modalities. Specifically, based on the insight that different images can naturally bind with text, M\textsuperscript{3}Bind first fine-tunes pre-trained CLIP-like image-text models to align their modality-specific text embedding space while preserving their original image-text alignments. Subsequently, we distill these modality-specific text encoders into a unified model, creating a shared text embedding space. Experiments on X-ray, CT, retina, ECG, and pathological images on multiple downstream tasks demonstrate that M\textsuperscript{3}Bind achieves state-of-the-art performance in zero-shot, few-shot classification and cross-modal retrieval tasks compared to its CLIP-like counterparts. These results validate M\textsuperscript{3}Bind's effectiveness in achieving cross-image-modal alignment for medical analysis.

Ke Ma, Jiaqi Tang, Bin Guo et al.

Despite the growing integration of deep models into mobile terminals, the accuracy of these models declines significantly due to various deployment interferences. Test-time adaptation (TTA) has emerged to improve the performance of deep models by adapting them to unlabeled target data online. Yet, the significant memory cost, particularly in resource-constrained terminals, impedes the effective deployment of most backward-propagation-based TTA methods. To tackle memory constraints, we introduce SURGEON, a method that substantially reduces memory cost while preserving comparable accuracy improvements during fully test-time adaptation (FTTA) without relying on specific network architectures or modifications to the original training procedure. Specifically, we propose a novel dynamic activation sparsity strategy that directly prunes activations at layer-specific dynamic ratios during adaptation, allowing for flexible control of learning ability and memory cost in a data-sensitive manner. Among this, two metrics, Gradient Importance and Layer Activation Memory, are considered to determine the layer-wise pruning ratios, reflecting accuracy contribution and memory efficiency, respectively. Experimentally, our method surpasses the baselines by not only reducing memory usage but also achieving superior accuracy, delivering SOTA performance across diverse datasets, architectures, and tasks.

Le Yu, Zhengyue Zhao, Yawen Zheng et al.

Reasoning-augmented Vision-Language Models (RVLMs) rely on safety alignment to prevent harmful behavior, yet their exposed chain-of-thought (CoT) traces introduce new attack surfaces. In this work, we find that the safety alignment of RVLMs can be easily break through a novel attack method termed \textbf{Stealth Fine-Tuning}. Our method elicits harmful reasoning traces through \textbf{segment-level interference} and reuses the self-generated outputs as supervised fine-tuning data. Through a \textbf{turn-based weighted} loss design, yielding a lightweight, distribution-consistent finetuning method. In our experiment, with only 499 samples and under 3 hours on a single A100 (QLoRA), Stealth Fine-Tuning outperforms IDEATOR by 38.52\% ASR while preserving general reasoning ability, as the tuned model retains the original representation distribution. Experiments on AdvBench and several general benchmarks demonstrate that Stealth Fine-Tuning is a low-cost and highly effective way to bypass alignment defenses. \textcolor{red}{\textbf{Disclaimer: This paper contains content that may be disturbing or offensive.}}

Suyang Xi, Chenxi Yang, Hong Ding et al.

Multimodal large language models (MLLMs) often fail in fine-grained visual question answering, producing hallucinations about object identities, positions, and relations because textual queries are not explicitly anchored to visual referents. Retrieval-augmented generation (RAG) alleviates some errors, but it fails to align with human-like processing at both the retrieval and augmentation levels. Specifically, it focuses only on global-level image information but lacks local detail and limits reasoning about fine-grained interactions. To overcome this limitation, we present Human-Like Retrieval-Augmented Generation (HuLiRAG), a framework that stages multimodal reasoning as a ``what--where--reweight'' cascade. Queries are first anchored to candidate referents via open-vocabulary detection (what), then spatially resolved with SAM-derived masks to recover fine-grained precision (where), and adaptively prioritized through the trade-off between local and global alignment (reweight). Mask-guided fine-tuning further injects spatial evidence into the generation process, transforming grounding from a passive bias into an explicit constraint on answer formulation. Extensive experiments demonstrate that this human-like cascade improves grounding fidelity and factual consistency while reducing hallucinations, advancing multimodal question answering toward trustworthy reasoning.

Zheng Yang, Guoxuan Chi, Chenshu Wu et al. · tsinghua

Generative Artificial Intelligence (GenAI) has made significant advancements in fields such as computer vision (CV) and natural language processing (NLP), demonstrating its capability to synthesize high-fidelity data and improve generalization. Recently, there has been growing interest in integrating GenAI into wireless sensing systems. By leveraging generative techniques such as data augmentation, domain adaptation, and denoising, wireless sensing applications, including device localization, human activity recognition, and environmental monitoring, can be significantly improved. This survey investigates the convergence of GenAI and wireless sensing from two complementary perspectives. First, we explore how GenAI can be integrated into wireless sensing pipelines, focusing on two modes of integration: as a plugin to augment task-specific models and as a solver to directly address sensing tasks. Second, we analyze the characteristics of mainstream generative models, such as Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), and diffusion models, and discuss their applicability and unique advantages across various wireless sensing tasks. We further identify key challenges in applying GenAI to wireless sensing and outline a future direction toward a wireless foundation model: a unified, pre-trained design capable of scalable, adaptable, and efficient signal understanding across diverse sensing tasks.

Weichen Zhang, Kebin Liu, Fan Dang et al.

Semantic segmentation in open-vocabulary scenarios presents significant challenges due to the wide range and granularity of semantic categories. Existing weakly-supervised methods often rely on category-specific supervision and ill-suited feature construction methods for contrastive learning, leading to semantic misalignment and poor performance. In this work, we propose a novel weakly-supervised approach, SynSeg, to address the challenges. SynSeg performs Multi-Category Contrastive Learning (MCCL) as a stronger training signal with a new feature reconstruction framework named Feature Synergy Structure (FSS). Specifically, MCCL strategy robustly combines both intra- and inter-category alignment and separation in order to make the model learn the knowledge of correlations from different categories within the same image. Moreover, FSS reconstructs discriminative features for contrastive learning through prior fusion and semantic-activation-map enhancement, effectively avoiding the foreground bias introduced by the visual encoder. Furthermore, SynSeg is a lightweight end-to-end solution without using any mid-term output from large-scale pretrained models and capable for real-time inference. In general, SynSeg effectively improves the abilities in semantic localization and discrimination under weak supervision in an efficient manner. Extensive experiments on benchmarks demonstrate that our method outperforms state-of-the-art (SOTA) performance. Particularly, SynSeg achieves higher accuracy than SOTA baselines with a ratio from 6.9\% up to 26.2\%.

Weichen Zhang, Zhui Zhu, Ningbo Li et al.

Vision-language models (VLMs) have achieved impressive performance on multimodal reasoning tasks such as visual question answering (VQA), but their inference cost remains a significant challenge due to the large number of vision tokens processed during the prefill stage. Existing pruning methods often rely on directly using the attention patterns or static text prompt guidance, failing to exploit the dynamic internal signals generated during inference. To address these issues, we propose AdaptInfer, a plug-and-play framework for adaptive vision token pruning in VLMs. First, we introduce a fine-grained, dynamic text-guided pruning mechanism that reuses layer-wise text-to-text attention maps to construct soft priors over text-token importance, allowing more informed scoring of vision tokens at each stage. Second, we perform an offline analysis of cross-modal attention shifts and identify consistent inflection locations in inference, which inspire us to propose a more principled and efficient pruning schedule. Our method is lightweight and plug-and-play, also generalizable across multi-modal tasks. Experimental results have verified the effectiveness of the proposed method. For example, it reduces CUDA latency by 61.3\% while maintaining an average accuracy of 92.9\% on vanilla LLaVA-1.5-7B. Under the same token budget, AdaptInfer surpasses SOTA in accuracy.

Xikai Sun, Fan Dang, Shiqi Jiang et al.

Conformance testing is essential for ensuring that protocol implementations comply with their specifications. However, traditional testing approaches involve manually creating numerous test cases and scripts, making the process labor-intensive and inefficient. Recently, Large Language Models (LLMs) have demonstrated impressive text comprehension and code generation abilities, providing promising opportunities for automation. In this paper, we propose iPanda, the first framework that leverages LLMs to automate protocol conformance testing. Given a protocol specification document and its implementation, iPanda first employs a keyword-based method to automatically generate comprehensive test cases. Then, it utilizes retrieval-augmented generation and customized CoT strategy to effectively interpret the implementation and produce executable test programs. To further enhance programs' quality, iPanda incorporates an iterative optimization mechanism to refine generated test scripts interactively. Finally, by executing and analyzing the generated tests, iPanda systematically verifies compliance between implementations and protocol specifications. Comprehensive experiments on various protocols show that iPanda significantly outperforms pure LLM-based approaches, improving the success rate (Pass@1) of test-program generation by factors ranging from 4.675 times to 10.751 times.

Chong Zhong, Yang Li, Jinfeng Xu et al.

We aim to assist image-based myopia screening by resolving two longstanding problems, "how to integrate the information of ocular images of a pair of eyes" and "how to incorporate the inherent dependence among high-myopia status and axial length for both eyes." The classification-regression task is modeled as a novel 4-dimensional muti-response regression, where discrete responses are allowed, that relates to two dependent 3rd-order tensors (3D ultrawide-field fundus images). We present a Vision Transformer-based bi-channel architecture, named CeViT, where the common features of a pair of eyes are extracted via a shared Transformer encoder, and the interocular asymmetries are modeled through separated multilayer perceptron heads. Statistically, we model the conditional dependence among mixture of discrete-continuous responses given the image covariates by a so-called copula loss. We establish a new theoretical framework regarding fine-tuning on CeViT based on latent representations, allowing the black-box fine-tuning procedure interpretable and guaranteeing higher relative efficiency of fine-tuning weight estimation in the asymptotic setting. We apply CeViT to an annotated ultrawide-field fundus image dataset collected by Shanghai Eye \& ENT Hospital, demonstrating that CeViT enhances the baseline model in both accuracy of classifying high-myopia and prediction of AL on both eyes.

Sirui Chen, Yunhao Liu, Jialong Li et al.

Dynamic state representation learning is an important task in robot learning. Latent space that can capture dynamics related information has wide application in areas such as accelerating model free reinforcement learning, closing the simulation to reality gap, as well as reducing the motion planning complexity. However, current dynamic state representation learning methods scale poorly on complex dynamic systems such as deformable objects, and cannot directly embed well defined simulation function into the training pipeline. We propose DiffSRL, a dynamic state representation learning pipeline utilizing differentiable simulation that can embed complex dynamics models as part of the end-to-end training. We also integrate differentiable dynamic constraints as part of the pipeline which provide incentives for the latent state to be aware of dynamical constraints. We further establish a state representation learning benchmark on a soft-body simulation system, PlasticineLab, and our model demonstrates superior performance in terms of capturing long-term dynamics as well as reward prediction.

Haochen Liu, Yiqi Wang, Wenqi Fan et al.

In the past few decades, artificial intelligence (AI) technology has experienced swift developments, changing everyone's daily life and profoundly altering the course of human society. The intention of developing AI is to benefit humans, by reducing human labor, bringing everyday convenience to human lives, and promoting social good. However, recent research and AI applications show that AI can cause unintentional harm to humans, such as making unreliable decisions in safety-critical scenarios or undermining fairness by inadvertently discriminating against one group. Thus, trustworthy AI has attracted immense attention recently, which requires careful consideration to avoid the adverse effects that AI may bring to humans, so that humans can fully trust and live in harmony with AI technologies. Recent years have witnessed a tremendous amount of research on trustworthy AI. In this survey, we present a comprehensive survey of trustworthy AI from a computational perspective, to help readers understand the latest technologies for achieving trustworthy AI. Trustworthy AI is a large and complex area, involving various dimensions. In this work, we focus on six of the most crucial dimensions in achieving trustworthy AI: (i) Safety & Robustness, (ii) Non-discrimination & Fairness, (iii) Explainability, (iv) Privacy, (v) Accountability & Auditability, and (vi) Environmental Well-Being. For each dimension, we review the recent related technologies according to a taxonomy and summarize their applications in real-world systems. We also discuss the accordant and conflicting interactions among different dimensions and discuss potential aspects for trustworthy AI to investigate in the future.

Kaixuan Chen, Dalin Zhang, Lina Yao et al.

The vast proliferation of sensor devices and Internet of Things enables the applications of sensor-based activity recognition. However, there exist substantial challenges that could influence the performance of the recognition system in practical scenarios. Recently, as deep learning has demonstrated its effectiveness in many areas, plenty of deep methods have been investigated to address the challenges in activity recognition. In this study, we present a survey of the state-of-the-art deep learning methods for sensor-based human activity recognition. We first introduce the multi-modality of the sensory data and provide information for public datasets that can be used for evaluation in different challenge tasks. We then propose a new taxonomy to structure the deep methods by challenges. Challenges and challenge-related deep methods are summarized and analyzed to form an overview of the current research progress. At the end of this work, we discuss the open issues and provide some insights for future directions.

Xiang Zhang, Lina Yao, Shuai Zhang et al.

A Brain-Computer Interface (BCI) acquires brain signals, analyzes and translates them into commands that are relayed to actuation devices for carrying out desired actions. With the widespread connectivity of everyday devices realized by the advent of the Internet of Things (IoT), BCI can empower individuals to directly control objects such as smart home appliances or assistive robots, directly via their thoughts. However, realization of this vision is faced with a number of challenges, most importantly being the issue of accurately interpreting the intent of the individual from the raw brain signals that are often of low fidelity and subject to noise. Moreover, pre-processing brain signals and the subsequent feature engineering are both time-consuming and highly reliant on human domain expertise. To address the aforementioned issues, in this paper, we propose a unified deep learning based framework that enables effective human-thing cognitive interactivity in order to bridge individuals and IoT objects. We design a reinforcement learning based Selective Attention Mechanism (SAM) to discover the distinctive features from the input brain signals. In addition, we propose a modified Long Short-Term Memory (LSTM) to distinguish the inter-dimensional information forwarded from the SAM. To evaluate the efficiency of the proposed framework, we conduct extensive real-world experiments and demonstrate that our model outperforms a number of competitive state-of-the-art baselines. Two practical real-time human-thing cognitive interaction applications are presented to validate the feasibility of our approach.

Xiang Zhang, Lina Yao, Xianzhi Wang et al.

Electroencephalography (EEG) signals reflect activities on certain brain areas. Effective classification of time-varying EEG signals is still challenging. First, EEG signal processing and feature engineering are time-consuming and highly rely on expert knowledge. In addition, most existing studies focus on domain-specific classification algorithms which may not be applicable to other domains. Moreover, the EEG signal usually has a low signal-to-noise ratio and can be easily corrupted. In this regard, we propose a generic EEG signal classification framework that accommodates a wide range of applications to address the aforementioned issues. The proposed framework develops a reinforced selective attention model to automatically choose the distinctive information among the raw EEG signals. A convolutional mapping operation is employed to dynamically transform the selected information to an over-complete feature space, wherein implicit spatial dependency of EEG samples distribution is able to be uncovered. We demonstrate the effectiveness of the proposed framework using three representative scenarios: intention recognition with motor imagery EEG, person identification, and neurological diagnosis. Three widely used public datasets and a local dataset are used for our evaluation. The experiments show that our framework outperforms the state-of-the-art baselines and achieves the accuracy of more than 97% on all the datasets with low latency and good resilience of handling complex EEG signals across various domains. These results confirm the suitability of the proposed generic approach for a range of problems in the realm of Brain-Computer Interface applications.

Xiang Zhang, Lina Yao, Salil S. Kanhere et al.

Person identification technology recognizes individuals by exploiting their unique, measurable physiological and behavioral characteristics. However, the state-of-the-art person identification systems have been shown to be vulnerable, e.g., contact lenses can trick iris recognition and fingerprint films can deceive fingerprint sensors. EEG (Electroencephalography)-based identification, which utilizes the users' brainwave signals for identification and offers a more resilient solution, draw a lot of attention recently. However, the accuracy still requires improvement and very little work is focusing on the robustness and adaptability of the identification system. We propose MindID, an EEG-based biometric identification approach, achieves higher accuracy and better characteristics. At first, the EEG data patterns are analyzed and the results show that the Delta pattern contains the most distinctive information for user identification. Then the decomposed Delta pattern is fed into an attention-based Encoder-Decoder RNNs (Recurrent Neural Networks) structure which assigns various attention weights to different EEG channels based on the importance of channels. The discriminative representations learned from the attention-based RNN are used to recognize the user identification through a boosting classifier. The proposed approach is evaluated over 3 datasets (two local and one public). One local dataset (EID-M) is used for performance assessment and the result illustrates that our model achieves an accuracy of 0.982 which outperforms the baselines and the state-of-the-art. Another local dataset (EID-S) and a public dataset (EEG-S) are utilized to demonstrate the robustness and adaptability, respectively. The results indicate that the proposed approach has the potential to be largely deployed in the practice environment.

Xiang Zhang, Lina Yao, Chaoran Huang et al.

Biometric authentication involves various technologies to identify individuals by exploiting their unique, measurable physiological and behavioral characteristics. However, traditional biometric authentication systems (e.g., face recognition, iris, retina, voice, and fingerprint) are facing an increasing risk of being tricked by biometric tools such as anti-surveillance masks, contact lenses, vocoder, or fingerprint films. In this paper, we design a multimodal biometric authentication system named Deepkey, which uses both Electroencephalography (EEG) and gait signals to better protect against such risk. Deepkey consists of two key components: an Invalid ID Filter Model to block unauthorized subjects and an identification model based on attention-based Recurrent Neural Network (RNN) to identify a subject`s EEG IDs and gait IDs in parallel. The subject can only be granted access while all the components produce consistent evidence to match the user`s proclaimed identity. We implement Deepkey with a live deployment in our university and conduct extensive empirical experiments to study its technical feasibility in practice. DeepKey achieves the False Acceptance Rate (FAR) and the False Rejection Rate (FRR) of 0 and 1.0%, respectively. The preliminary results demonstrate that Deepkey is feasible, show consistent superior performance compared to a set of methods, and has the potential to be applied to the authentication deployment in real world settings.

Lan Zhang, Taeho Jung, Puchun Feng et al.

High-resolution cameras produce huge volume of high quality images everyday. It is extremely challenging to store, share and especially search those huge images, for which increasing number of cloud services are presented to support such functionalities. However, images tend to contain rich sensitive information (\eg, people, location and event), and people's privacy concerns hinder their readily participation into the services provided by untrusted third parties. In this work, we introduce PIC: a Privacy-preserving large-scale Image search system on Cloud. Our system enables efficient yet secure content-based image search with fine-grained access control, and it also provides privacy-preserving image storage and sharing among users. Users can specify who can/cannot search on their images when using the system, and they can search on others' images if they satisfy the condition specified by the image owners. Majority of the computationally intensive jobs are outsourced to the cloud side, and users only need to submit the query and receive the result throughout the entire image search. Specially, to deal with massive images, we design our system suitable for distributed and parallel computation and introduce several optimizations to further expedite the search process. We implement a prototype of PIC including both cloud side and client side. The cloud side is a cluster of computers with distributed file system (Hadoop HDFS) and MapReduce architecture (Hadoop MapReduce). The client side is built for both Windows OS laptops and Android phones. We evaluate the prototype system with large sets of real-life photos. Our security analysis and evaluation results show that PIC successfully protect the image privacy at a low cost of computation and communication.

Lan Zhang, Taeho Jung, Cihang Liu et al.

With the proliferation of mobile devices, cloud-based photo sharing and searching services are becoming common due to the mobile devices' resource constrains. Meanwhile, there is also increasing concern about privacy in photos. In this work, we present a framework \ourprotocolNSP, which enables cloud servers to provide privacy-preserving photo sharing and search as a service to mobile device users. Privacy-seeking users can share their photos via our framework to allow only their authorized friends to browse and search their photos using resource-bounded mobile devices. This is achieved by our carefully designed architecture and novel outsourced privacy-preserving computation protocols, through which no information about the outsourced photos or even the search contents (including the results) would be revealed to the cloud servers. Our framework is compatible with most of the existing image search technologies, and it requires few changes to the existing cloud systems. The evaluation of our prototype system with 31,772 real-life images shows the communication and computation efficiency of our system.

Lan Zhang, Kebin Liu, Xiang-Yang Li et al.

The wide adoption of wearable smart devices with onboard cameras greatly increases people's concern on privacy infringement. Here we explore the possibility of easing persons from photos captured by smart devices according to their privacy protection requirements. To make this work, we need to address two challenges: 1) how to let users explicitly express their privacy protection intention, and 2) how to associate the privacy requirements with persons in captured photos accurately and efficiently. Furthermore, the association process itself should not cause portrait information leakage and should be accomplished in a privacy-preserving way. In this work, we design, develop, and evaluate a protocol, that enables a user to flexibly express her privacy requirement and empowers the photo service provider (or image taker) to exert the privacy protection policy.Leveraging the visual distinguishability of people in the field-of-view and the dimension-order-independent property of vector similarity measurement, we achieves high accuracy and low overhead. We implement a prototype system, and our evaluation results on both the trace-driven and real-life experiments confirm the feasibility and efficiency of our system.

Jizhong Zhao, Wei Xi, Jinsong Han et al.

Generating keys and keeping them secret is critical in secure communications. Due to the "open-air" nature, key distribution is more susceptible to attacks in wireless communications. An ingenious solution is to generate common secret keys by two communicating parties separately without the need of key exchange or distribution, and regenerate them on needs. Recently, it is promising to extract keys by measuring the random variation in wireless channels, e.g., RSS. In this paper, we propose an efficient Secret Key Extraction protocol without Chasing down Errors, SKECE. It establishes common cryptographic keys for two communicating parties in wireless networks via the realtime measurement of Channel State Information (CSI). It outperforms RSS-based approaches for key generation in terms of multiple subcarriers measurement, perfect symmetry in channel, rapid decorrelation with distance, and high sensitivity towards environments. In the SKECE design, we also propose effective mechanisms such as the adaptive key stream generation, leakage resilient consistence validation, and weighted key recombination, to fully exploit the excellent properties of CSI. We implement SKECE on off-the-shelf 802.11n devices and evaluate its performance via extensive experiments. The results demonstrate that SKECE achieves a more than 3x throughput gain in the key generation from one subcarrier in static scenarios, and due to its high efficiency, a 50% reduction on the communication overhead compared to the state-of-the-art RSS based approaches.