Xin Chen, Feng Jiang, Yiqian Zhang, Hardy Chen, Shuo Yan, Wenya Xie, Min Yang, Shujian Huang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Transformer-based models have significantly advanced image captioning through self-attention mechanisms and parallel computation. However, existing methods typically adopt teacher-forcing strategies during training by conditioning the decoder exclusively on ground-truth tokens, whereas at inference, captions are generated autoregressively based solely on previously predicted tokens. Such discrepancy between training and inference conditions leads to a progressive accumulation of prediction errors, resulting in captions that deviate significantly from the visual content. While scheduled sampling strategies mitigate this issue, directly integrating them disrupts Transformer parallelism and overlooks visual saliency differences. To tackle these limitations, we present VMFormer, a multi-stage decoding Transformer featuring a Visual-Aware Scheduled Sampling (VASS) module that bridges this gap through two key innovations: 1) A two-stage decoding scheme where an initial self-study stage generates candidate tokens, followed by a hybrid stage dynamically blending ground-truth references with predictions via a visual clues controlled gate. 2) A cognitive-inspired prioritization mechanism that retains visual keywords (nouns/verbs/attributes) in early training phases before transitioning to linguistic refinements, mirroring human captioning patterns. Crucially, the VASS module preserves the parallel computational strengths of Transformer architectures and is designed as a plug-and-play component, readily adaptable to Transformer-based captioning models. Experiments on challenging MSCOCO achieve 142.2 CIDEr. We extend VMFormer to video captioning and demonstrate consistent improvements on MSRVTT and MSVD datasets.

In this paper, we propose a simple yet unified single object tracking (SOT) framework, dubbed SUTrack. It consolidates five SOT tasks (RGB-based, RGB-Depth, RGB-Thermal, RGB-Event, RGB-Language Tracking) into a unified model trained in a single session. Due to the distinct nature of the data, current methods typically design individual architectures and train separate models for each task. This fragmentation results in redundant training processes, repetitive technological innovations, and limited cross-modal knowledge sharing. In contrast, SUTrack demonstrates that a single model with a unified input representation can effectively handle various SOT tasks, eliminating the need for task-specific designs and separate training sessions. Additionally, we introduce a task-recognition training strategy and a soft token type embedding to further enhance SUTrack's performance with minimal overhead. Experiments show that SUTrack outperforms previous task-specific counterparts across 11 datasets spanning five SOT tasks. Moreover, we provide a range of models catering edge devices as well as high-performance GPUs, striking a good trade-off between speed and accuracy. We hope SUTrack could serve as a strong foundation for further compelling research into unified tracking models.

Efficient tracking has garnered attention for its ability to operate on resource-constrained platforms for real-world deployment beyond desktop GPUs. Current efficient trackers mainly follow precision-oriented trackers, adopting a one-stream framework with lightweight modules. However, blindly adhering to the one-stream paradigm may not be optimal, as incorporating template computation in every frame leads to redundancy, and pervasive semantic interaction between template and search region places stress on edge devices. In this work, we propose a novel asymmetric Siamese tracker named AsymTrack for efficient tracking. AsymTrack disentangles template and search streams into separate branches, with template computing only once during initialization to generate modulation signals. Building on this architecture, we devise an efficient template modulation mechanism to unidirectional inject crucial cues into the search features, and design an object perception enhancement module that integrates abstract semantics and local details to overcome the limited representation in lightweight tracker. Extensive experiments demonstrate that AsymTrack offers superior speed-precision trade-offs across different platforms compared to the current state-of-the-arts. For instance, AsymTrack-T achieves 60.8% AUC on LaSOT and 224/81/84 FPS on GPU/CPU/AGX, surpassing HiT-Tiny by 6.0% AUC with higher speeds.

Due to the challenges of processing temporal information, most trackers depend solely on visual discriminability and overlook the unique temporal coherence of video data. In this paper, we propose a lightweight and plug-and-play motion prompt tracking method. It can be easily integrated into existing vision-based trackers to build a joint tracking framework leveraging both motion and vision cues, thereby achieving robust tracking through efficient prompt learning. A motion encoder with three different positional encodings is proposed to encode the long-term motion trajectory into the visual embedding space, while a fusion decoder and an adaptive weight mechanism are designed to dynamically fuse visual and motion features. We integrate our motion module into three different trackers with five models in total. Experiments on seven challenging tracking benchmarks demonstrate that the proposed motion module significantly improves the robustness of vision-based trackers, with minimal training costs and negligible speed sacrifice. Code is available at https://github.com/zj5559/Motion-Prompt-Tracking.

Abstract Transformer-based visual trackers have demonstrated significant advancements due to their powerful modeling capabilities. However, their practicality is limited on resource-constrained devices because of their slow processing speeds. To address this challenge, we present HiT, a novel family of efficient tracking models that achieve high performance while maintaining fast operation across various devices. The core innovation of HiT lies in its Bridge Module, which connects lightweight transformers to the tracking framework, enhancing feature representation quality. Additionally, we introduce a dual-image position encoding approach to effectively encode spatial information. HiT achieves an impressive speed of 61 frames per second (fps) on the NVIDIA Jetson AGX platform, alongside a competitive AUC of 64.6% on the LaSOT benchmark, outperforming all previous efficient trackers. Building on HiT, we propose DyHiT, an efficient dynamic tracker that flexibly adapts to scene complexity by selecting routes with varying computational requirements. DyHiT uses search area features extracted by the backbone network and inputs them into an efficient dynamic router to classify tracking scenarios. Based on the classification, DyHiT applies a divide-and-conquer strategy, selecting appropriate routes to achieve a superior trade-off between accuracy and speed. The fastest version of DyHiT achieves 111 fps on NVIDIA Jetson AGX while maintaining an AUC of 62.4% on LaSOT. Furthermore, we introduce a training-free acceleration method based on the dynamic routing architecture of DyHiT. This method significantly improves the execution speed of various high-performance trackers without sacrificing accuracy. For instance, our acceleration method enables the state-of-the-art tracker SeqTrack-B256 to achieve a $$2.68\times $$ 2.68 × speedup on an NVIDIA GeForce RTX 2080 Ti GPU while maintaining the same AUC of 69.9% on the LaSOT. Codes, models, and results are available at https://github.com/kangben258/HiT .

During software development, developers often need appropriate but unfamiliar APIs to implement a specific functionality. Under such circumstances, developers tend to leverage search tools to seek for the relevant APIs. However, there are always semantic gaps between query words and APIs, which negatively affects the performance of these tools. In this study, we introduce Glo-APIRec, a method that combines global semantics with local context to estimate the semantic relevance between query words and APIs to recommend APIs. In this method, the Transformer model is employed to obtain global semantics, while the Word2Vec model is utilized to capture local context using a fixed-size window. First, Glo-APIRec collects millions of Java projects from GitHub to construct the corpus. Afterward, a set of tuples consisting of words and APIs is built by extracting comments and API sequences from the source code files. Finally, Transformer is employed to capture long distance semantics about API sequences and code comments. Meanwhile, Word2Vec is used to generate word vectors to capture the local context by introducing the random shuffling strategy to break the positions of words and APIs in the tuples. We evaluate the performance of Glo-APIRec with 30 sentence-level queries. Experimental results show that Glo-APIRec can achieve 0.600 in terms of SuccessRate for top-1 recommendation and 0.900 for top-10 recommendation. When recommending 10 APIs, Glo-APIRec can achieve 0.480, 0.703 and 0.717 in terms of precision, Mean Reciprocal Rank ([Formula: see text]) and Normalized Discounted Cumulative Gain ([Formula: see text]), and outperforms the state-of-the-art method by 26.2%, 31.7% and 27.9%, respectively.

Abstract Detecting content generated by large language models (LLMs) is crucial for preventing misuse and building trustworthy AI systems. Although existing detection methods perform well, their robustness in out-of-distribution (OOD) scenarios is still lacking. In this paper, we hypothesize that, compared to features used by existing detection methods, the internal representations of LLMs contain more comprehensive and raw features that can more effectively capture and distinguish the statistical pattern differences between LLM-generated texts (LGT) and human-written texts (HWT). We validated this hypothesis across different LLMs and observed significant differences in neural activation patterns when processing these two types of texts. Based on this, we propose RepreGuard, an efficient statistics-based detection method. Specifically, we first employ a surrogate model to collect representation of LGT and HWT, and extract the distinct activation feature that can better identify LGT. We can classify the text by calculating the projection score of the text representations along this feature direction and comparing with a precomputed threshold. Experimental results show that RepreGuard outperforms all baselines with average 94.92% AUROC on both in-distribution and OOD scenarios, while also demonstrating robust resilience to various text sizes and mainstream attacks.1

Vision-language object tracking integrates advanced linguistic information, enhancing its robustness and accuracy in complex scenarios. Nevertheless, current methods are constrained by a lack of sufficient vision-language data, making it challenging for the model to learn generalized knowledge. To alleviate this issue, we propose a new prompt-based framework for vision-language tracking, named ProVLT. This framework casts language information as a prompt for pretrained visionbased tracking models, thereby leveraging the knowledge from extensive tracking data. Experiments demonstrate that ProVLT achieves competitive performance while training only a fraction of parameters (approximately 29% of modal parameters). For instance, ProVLT achieves competitive performance, attaining AUC of 59.8% on TNL2K benchmark. Furthermore, we augment five mainstream vision-only tracking benchmarks with language annotations, and find that the inclusion of linguistic information consistently improves tracking performance. On these benchmarks, the linguistic information improves the performance by an average of 2.9% compared with the vision-based tracker. We will release the code, models, and benchmarks for the community.

Due to the substantial gap between vision and language modalities, along with the mismatch problem between fixed language descriptions and dynamic visual information, existing vision-language tracking methods exhibit performance on par with or slightly worse than vision-only tracking. Effectively exploiting the rich semantics of language to enhance tracking robustness remains an open challenge. To address these issues, we propose a self-adaptive vision-language tracking framework that leverages the pre-trained multi-modal CLIP model to obtain well-aligned visual-language representations. A novel context-aware prompting mechanism is introduced to dynamically adapt linguistic cues based on the evolving visual context during tracking. Specifically, our context prompter extracts dynamic visual features from the current search image and integrates them into the text encoding process, enabling self-updating language embeddings. Furthermore, our framework employs a unified one-stream Transformer architecture, supporting joint training for both vision-only and vision-language tracking scenarios. Our method not only bridges the modality gap but also enhances robustness by allowing language features to evolve with visual context. Extensive experiments on four vision-language tracking benchmarks demonstrate that our method effectively leverages the advantages of language to enhance visual tracking. Our large model can obtain 55.0% AUC on $\text {LaSOT}_{\text {EXT}}$ and 69.0% AUC on TNL2K. Additionally, our language-only tracking model achieves performance comparable to that of state-of-the-art vision-only tracking methods on TNL2K. Code is available at https://github.com/zj5559/SAVLT.

Practical deployments, especially on resource-limited edge devices, necessitate high speed for visual object trackers. To meet this demand, we introduce a new efficient tracker with a Two-Stream architecture, named ToS. While the recent one-stream tracking framework, employing a unified backbone for simultaneous processing of both the template and search region, has demonstrated exceptional efficacy, we find the conventional two-stream tracking framework, which employs two separate backbones for the template and search region, offers inherent advantages. The two-stream tracking framework is more compatible with advanced lightweight backbones and can efficiently utilize benefits from large templates. We demonstrate that the two-stream setup can exceed the one-stream tracking model in both speed and accuracy through strategic designs. Our methodology rejuvenates the two-stream tracking paradigm with lightweight pre-trained backbones and the proposed three efficient strategies: 1) A feature-aggregation module that improves the representation capability of the backbone, 2) A channel-wise approach for feature fusion, presenting a more effective and lighter alternative to spatial concatenation techniques, and 3) An expanded template strategy to boost tracking accuracy with negligible additional computational cost. Extensive evaluations across multiple tracking benchmarks demonstrate that the proposed method sets a new state-of-the-art performance in efficient visual tracking.

The speed-precision tradeoff is a critical problem in visual object tracking, as it typically requires low latency and is deployed on resource-constrained platforms. Existing solutions for efficient tracking primarily focus on lightweight backbones or modules, which, however, come at a sacrifice in precision. In this article, inspired by dynamic network routing, we propose DyTrack, a dynamic transformer framework for efficient tracking. Real-world tracking scenarios exhibit varying levels of complexity. We argue that a simple network is sufficient for easy video frames, while more computational resources should be assigned to difficult ones. DyTrack automatically learns to configure proper reasoning routes for different inputs, thereby improving the utilization of the available computational budget and achieving higher performance at the same running speed. We formulate instance-specific tracking as a sequential decision problem and incorporate terminating branches to intermediate layers of the model. Furthermore, we propose a feature recycling mechanism to maximize computational efficiency by reusing the outputs of predecessors. Additionally, a target-aware self-distillation strategy is designed to enhance the discriminating capabilities of early-stage predictions by mimicking the representation patterns of the deep model. Extensive experiments demonstrate that DyTrack achieves promising speed-precision tradeoffs with only a single model. For instance, DyTrack obtains 64.9% area under the curve (AUC) on LaSOT with a speed of 256 fps.

In recent years, researchers have proposed various approaches to detect code comments that explicitly acknowledge Technical Debt (TD), which are referred to as Self-Admitted Technical Debt (SATD) comments. Previous studies have proven that hidden patterns can be learned from SATD code snippets to predict whether the code snippets hold TD without the aid of comments. In this study, we refer to such TD as unadmitted TD, i.e., TD whose code snippets exhibit patterns similar to those of SATD, but are not annotated with comments indicating the existence of TD. Given that current unadmitted TD datasets are limited to method-level and conditional-statement-level code snippets, we construct the world’s most comprehensive dataset of code snippets and their corresponding comments, which includes 18 popular Java open-source projects and covers code snippets at the file, class, method and block levels. Around this dataset, we have conducted four key research activities.Firstly, we propose an automated framework for data collection and annotation, which extracts commented code snippets of varying granularity from projects and assigns SATD labels using three state-of-the-art SATD detection approaches. Secondly, we conduct a rigorous evaluation process, including the validity test, reliability test and manual verification, to ensure the accuracy and consistency of the dataset before further analysis and utilization. Additionally, we propose a metric-based detection approach named LiteM that detects unadmitted TD solely based on code metrics. As for the real-world scenarios where training data is scarce, we further introduce LiteMC, which generates pseudo-labels for commented code snippets and then employs LiteM to train a model on these pseudo-labeled data, enabling the detection of unadmitted TD in uncommented code snippets. The experimental results demonstrate the effectiveness and efficiency of both LiteM and LiteMC. The dataset and the code are available athttps://github.com/HduDBSI/Dataset4TD.

Existing binary vulnerability detection methods are mainly divided into static analysis and dynamic analysis. Compared with dynamic analysis, static analysis has higher code coverage and detection efficiency but is limited by accuracy. Recently, deep learning has achieved a significant improvement in static analysis for binary vulnerability detection. However, static analysis still suffers from two limitations: (i) the identification of function vulnerabilities relies on their patches; (ii) feature extraction either loses contextual dependency or cannot effectively extract local features. In this paper, we propose a new binary vulnerability detection method named VulDet-BC by combining bidirectional gated recurrent units (BiGRU) and convolutional neural networks (CNN). First, VulDet-BC preprocesses binary code text to make each function have the same text structure. Then, for each instruction, we perform BiGRU with word-attention to generate the vector representation which contains contextual dependency between words. Finally, another BiGRU and CNN are combined to train the prediction model and the eventual model is adopted to forecast whether target binary functions contain vulnerabilities or not. VulDet-BC not only preserves long-distance information but also effectively extracts local features related to vulnerabilities. We have applied VulDet- BC to the dataset with 15,954 non-vulnerable functions and 16,327 vulnerable functions. The experimental results show that VulDet-BC achieves 86.7% in terms of accuracy, 97.6% in terms of recall, 80.8% in terms of precision, 88.3% in terms of F1-score, and 86.3% in terms of area under curve (AUC) on average, and outperforms the state-of-the-art baseline.

This paper introduces a Safety-First Tracker (SF-Tracker) designed for omnidirectional autonomous tracking robots. The position and orientation of omnidirectional robots are decoupled for stepwise planning to ensure trajectory safety and maintain target visibility. SF-Tracker puts the trajectory safety in the first place. First, a collision-free and occlusion-free reference path is efficiently initialized by constructing a directed weighted graph. By building upon this path, safe trajectory optimization is implemented to ensure safe movement. Finally, an orientation planner is developed to achieve target visibility based on the safe trajectory. Extensive experimental evaluations in simulated environments and the real world demonstrate that the SF-Tracker outperforms state-of-the-art methods in terms trajectory safety and target visibility. Ablation experiments further demonstrate the significance of each step of the SF-Tracker. The source code and demonstration video can be found at https://github.com/Yue-0/SF-Tracker.

Pathogenic chromosome abnormalities are very common among the general population. While numerical chromosome abnormalities can be quickly and precisely detected, structural chromosome abnormalities are far more complex and typically require considerable efforts by human experts for identification. This paper focuses on investigating the modeling of chromosome features and the identification of chromosomes with structural abnormalities. Most existing data-driven methods concentrate on a single chromosome and consider each chromosome independently, overlooking the crucial aspect of homologous chromosomes. In normal cases, homologous chromosomes share identical structures, with the exception that one of them is abnormal. Therefore, we propose an adaptive method to align homologous chromosomes and diagnose structural abnormalities through homologous similarity. Inspired by the process of human expert diagnosis, we incorporate information from multiple pairs of homologous chromosomes simultaneously, aiming to reduce noise disturbance and improve prediction performance. Extensive experiments on real-world datasets validate the effectiveness of our model compared to baselines.

The dominant trackers generate a fixed-size rectangular region based on the previous prediction or initial bounding box as the model input, i.e., search region. While this manner obtains promising tracking efficiency, a fixed-size search region lacks flexibility and is likely to fail in some cases, e.g., fast motion and distractor interference. Trackers tend to lose the target object due to the limited search region or experience interference from distractors due to the excessive search region. Drawing inspiration from the pattern humans track an object, we propose a novel tracking paradigm, called Search Region Regulation Tracking (SRRT) that applies a small eyereach when the target is captured and zooms out the search field when the target is about to be lost. SRRT applies a proposed search region regulator to estimate an optimal search region dynamically for each frame, by which the tracker can flexibly respond to transient changes in the location of object occurrences. To adapt the object’s appearance variation during online tracking, we further propose a locking-state determined updating strategy for reference frame updating. The proposed SRRT is concise without bells and whistles, yet achieves evident improvements and competitive results with other state-of-the-art trackers on eight benchmarks. On the large-scale LaSOT benchmark, SRRT improves SiamRPN++ and TransT with absolute gains of 4.6% and 3.1% in terms of AUC. The code and models will be released.

Familiar analysis for malware plays an important role in comprehending the diversity of malicious behaviors and identifying the emerging security threats. Existing studies mainly focus on classifying malware into known families by supervised learning. However, these methods face two main challenges, 1) the lack of a large amount of labeled data and 2) the poor effectiveness in identifying unknown families of malware. To overcome these challenges, we propose a new method called multiple features (MulFC) based on unsupervised learning. In the method, we first leverage a decompiling tool to extract multiple features, including manifest features, application programming interface (API) features, and opcode features. Then, the opcode features are preprocessed to filter out the redundant ones to reduce the calculation cost. After that, we adopt the Jaccard index to calculate the similarities between malware and construct a malware network. Finally, InfoMap is applied to perform the clustering on the basis of the malware network. Overall, MulFC does not require the use of labeled data and can identify unknown families of malware. Experiments are conducted on two datasets for the performance evaluation of MulFC. The experimental results show that MulFC achieves 0.810 in terms of normalized mutual information, 0.576 in terms of adjusted rand index, 0.620 in terms of the Fowlkes–Mallows index, and 0.805 in terms of V-measure on average, and outperforms the state-of-the-art baseline method by 0.060, 0.054, 0.038, and 0.065, respectively.

Convolutional Neural Networks (CNNs) are widely deployed on the Graphics Processing Unit (GPU) to support Deep Learning (DL) based services. Popular DL frameworks usually ignore the inter-operator parallelism when executing the inference of CNNs, which results in high inference latency. Although some inter-operator scheduling methods have been proposed, there remains a critical trade-off issue between inference latency (effectiveness) and scheduling time (efficiency). In this article, we propose LIOS, a novel latency-based heuristic inter-operator scheduling method to balance inference latency and scheduling time. In LIOS, a CNN latency model is built based on the given CNN and GPU. Then every operator is assigned a priority value to represent its importance. During each iteration of the scheduling process, LIOS identifies the current data-independent operators, selects the operator with the highest priority value, and assigns it to the GPU stream with the smallest finish time. Extensive experimental results have demonstrated the effectiveness and efficiency of LIOS. For the effectiveness, LIOS can speed up the inference of normal-size and large-size CNNs by 1.13$\sim 1.59 \times$compared to sequential scheduling. This result is comparable to IOS, the latest state-of-the-art scheduling method. For the efficiency, LIOS can speed up the scheduling process by 7$\sim 9210\times$compared to IOS.

Automated software fault localization has become one of the hot spots on which researchers have focused in recent years. Existing studies have shown that learning-based techniques can effectively localize faults leveraging various information. However, there exist two problems in these techniques. The first is that they simply represent various information without caring the contribution of different information. The second is that the data imbalance problem is not considered in these techniques. Thus, their effectiveness is limited in practice. In this paper, we propose HetFL, a novel heterogeneous graph-based software fault localization technique to aggregate different information into a heterogeneous graph in which program entities and test cases are regarded as nodes, and coverage, change histories, and call relationships are viewed as edges. HetFL first extracts textual and structure information from source code as attributes of nodes and integrates them to form an attribute vector. Then, for a given node, HetFL finds its neighbor nodes based on the types of edges and aggregates corresponding neighbor nodes to form type vectors. After that, the attribute vector and all the type vectors of each node are aggregated to generate the final vector representation by an attention mechanism. Finally, we leverage a convolution neural network (CNN) to obtain the suspicious score of each method. To validate the effectiveness of HetFL, experiments are conducted on the widely used dataset Defects4J (v1.2.0). The experimental results show that HetFL can localize 217 faults within Top-1 that is 25 higher than the state-of-the-art technique DeepFL, and achieve 6.37 and 5.58 in terms of MAR and MFR which improve DeepFL by 9.0% and 5.6%, respectively. In addition, we also perform experiments on the latest version of Defects4J (v2.0.0). The experimental results show that HetFL has better performance than the baseline methods.

In this paper, we present a new sequence-to-sequence learning framework for visual tracking, dubbed SeqTrack. It casts visual tracking as a sequence generation problem, which predicts object bounding boxes in an autoregressive fashion. This is different from prior Siamese trackers and transformer trackers, which rely on designing complicated head networks, such as classification and regression heads. SeqTrack only adopts a simple encoder-decoder transformer architecture. The encoder extracts visual features with a bidirectional transformer, while the decoder generates a sequence of bounding box values autoregressively with a causal transformer. The loss function is a plain cross-entropy. Such a sequence learning paradigm not only simplifies tracking framework, but also achieves competitive performance on benchmarks. For instance, SeqTrack gets 72.5% AUC on LaSOT, establishing a new state-of-the-art performance. Code and models are available at https://github.com/microsoft/VideoX.

Visible-modal object tracking gives rise to a series of downstream multi-modal tracking tributaries. To inherit the powerful representations of the foundation model, a natural modus operandi for multi-modal tracking is full fine-tuning on the RGB-based parameters. Albeit effective, this manner is not optimal due to the scarcity of downstream data and poor transferability, etc. In this paper, inspired by the recent success of the prompt learning in language models, we develop Visual Prompt multi-modal Tracking (ViPT), which learns the modal-relevant prompts to adapt the frozen pre-trained foundation model to various downstream multi-modal tracking tasks. ViPT finds a better way to stimulate the knowledge of the RGB-based model that is pre-trained at scale, meanwhile only introducing a few trainable parameters (less than 1% of model parameters). ViPT outperforms the full fine-tuning paradigm on multiple downstream tracking tasks including RGB+Depth, RGB+Thermal, and RGB+Event tracking. Extensive experiments show the potential of visual prompt learning for multi-modal tracking, and ViPT can achieve state-of-the-art performance while satisfying parameter efficiency. Code and models are available at https://github.com/jiawen-zhu/ViPT.

Transformer-based visual trackers have demonstrated significant progress owing to their superior modeling capabilities. However, existing trackers are hampered by low speed, limiting their applicability on devices with limited computational power. To alleviate this problem, we propose HiT, a new family of efficient tracking models that can run at high speed on different devices while retaining high performance. The central idea of HiT is the Bridge Module, which bridges the gap between modern lightweight transformers and the tracking framework. The Bridge Module incorporates the high-level information of deep features into the shallow large-resolution features. In this way, it produces better features for the tracking head. We also propose a novel dual-image position encoding technique that simultaneously encodes the position information of both the search region and template images. The HiT model achieves promising speed with competitive performance. For instance, it runs at 61 frames per second (fps) on the Nvidia Jetson AGX edge device. Furthermore, HiT attains 64.6% AUC on the LaSOT benchmark, surpassing all previous efficient trackers. Code and models are available at https://github.com/kangben258/HiT.

Correlation has a critical role in the tracking field, especially in recent popular Siamese-based trackers. The correlation operation is a simple fusion method that considers the similarity between the template and the search region. However, the correlation operation is a local linear matching process, losing semantic information and easily falling into a local optimum, which may be the bottleneck in designing high-accuracy tracking algorithms. In this work, to determine whether a better feature fusion method exists than correlation, a novel attention-based feature fusion network, inspired by the transformer, is presented. This network effectively combines the template and search region features using attention mechanism. Specifically, the proposed method includes an ego-context augment module based on self-attention and a cross-feature augment module based on cross-attention. First, we present a transformer tracking (named TransT) method based on the Siamese-like feature extraction backbone, the designed attention-based fusion mechanism, and the classification and regression heads. Based on the TransT baseline, we also design a segmentation branch to generate the accurate mask. Finally, we propose a stronger version of TransT by extending it with a multi-template scheme and an IoU prediction head, named TransT-M. Experiments show that our TransT and TransT-M methods achieve promising results on seven popular benchmarks. Code and models are available at https://github.com/chenxin-dlut/TransT-M.

Neural Architecture Search (NAS) aims to find efficient models for multiple tasks. Beyond seeking solutions for a single task, there are surging interests in transferring network design knowledge across multiple tasks. In this line of research, effectively modeling task correlations is vital yet highly neglected. Therefore, we propose Arch-Graph, a transferable NAS method that predicts task-specific optimal architectures with respect to given task embeddings. It leverages correlations across multiple tasks by using their embeddings as a part of the predictor's input for fast adaptation. We also formulate NAS as an architecture relation graph prediction problem, with the relational graph constructed by treating candidate architectures as nodes and their pairwise relations as edges. To enforce some basic properties such as acyclicity in the relational graph, we add additional constraints to the optimization process, converting NAS into the problem of finding a Maximal Weighted Acyclic Subgraph (MWAS). Our algorithm then strives to eliminate cycles and only establish edges in the graph if the rank results can be trusted. Through MWAS, Arch-Graph can effectively rank candidate models for each task with only a small budget to finetune the predictor. With extensive experiments on TransNAS-Bench-101, we show Arch-Graph's transferability and high sample efficiency across numerous tasks, beating many NAS methods designed for both single-task and multi-task search. It is able to find top 0.16% and 0.29% architectures on average on two search spaces under the budget of only 50 models.11Code: https://github.com/Centaurus982034/Arch-Graph

As one of the most common code smells, feature envy reduces the cohesion of classes and increases the coupling between classes, thus leading to difficulty of software maintainability. Though many studies have made good achievements on feature envy detection, they often despise or even ignore the inherent calling relationships between methods, causing unimpressive detection efficiency. To address this problem, we propose a Graph Neural Network (GNN) based approach towards feature envy detection. We first collect code metrics and calling relationships, and then convert them to the form of a graph, where the node represents the code metrics of a method and the edge represents the calling relationship between methods. Particularly, considering the unbalance of positive and negative samples, we introduce a graph augmenter to obtain an enhanced graph. Finally, we feed the enhanced graph into a GNN model for training and predicting. We conducted extensive experiments on a dataset containing five open-source software projects. The result shows that our approach achieves 78.90% in terms of average F1-score, which is 37.98% absolutely higher than the best comparison approach. Besides, we propose a refactoring recommendation approach based on calling strength. It achieves 61.44% of accuracy, which is 5.13% absolutely higher than the best competitive. Our code and datasets are available at https://github.com/HduDBSI/Feature-Envy-Detection.

Vehicular ad hoc network (VANET) has become an accessible technology for improving road safety and driving experience, the problems of heterogeneity and lack of resources it faces have also attracted widespread attention. With the development of software-defined networking (SDN) and multiaccess edge computing (MEC), a variety of resource allocation strategies in MEC-enabled software-defined networking-based VANET (SDVN) have been proposed to solve these problems. However, we note that few of these work involves the situation where SDVN is under Distributed Denial of Service (DDoS) attacks. Actually, Internet of Things (IoT) devices are extremely easy to be compromised by malicious users, and compromised IoT devices may be used to launch edge DDoS attacks against the MEC servers in MEC-enabled SDVN at any time. In this article, we propose a graph neural network (GNN)-based collaborative deep reinforcement learning (GCDRL) model to generate the resource provisioning and mitigating strategy. The model evaluates the trust value of the vehicles, formulates mitigation of edge DDoS attacks and resource provisioning strategies to ensure that the MEC servers can work normally under edge DDoS attacks. In addition, GNN is adopted in the DRL model to extract the structure feature of the graph composed of MEC servers, and help transfer computing tasks between MEC servers to alleviate the problem of resources imbalance between them. Experimental results show that the method of estimating the vehicular trust value is effective, and our method can make the average throughput of edge nodes more stable and lower down the average delay and the average energy consumption under the edge DDoS attack. Also, a real-world case study is conducted to verify our conclusion.

In software development, due to the demands from users or the limitations of time and resources, developers tend to adopt suboptimal solutions to achieve quick software development. In such a way, the released software usually involves not-quite-right code that is called technical debt, which will significantly decrease the quality of software and increase the maintenance cost. Recently, the concept of self-admitted technical debt (SATD) is proposed and refers to technical debt that is self-admitted by developers in code comments. Existing studies mainly focus on detecting technical debt by classifying code comments into either “SATD” or “non-SATD.” However, different types of SATD has different impacts on software maintenance and needs to be handled by different developers. Therefore, the detected SATD should be further classified so that developers can understand and remove technical debt better. In this article, we propose a new method based on eXtreme Gradient Boosting (XGBoost) to classify SATD into multiple classes. In our approach, we first preprocess the original code comments and adopt the easy data augmentation strategy to overcome the class unbalance problem. Then, chi-square is leveraged to select representative features from the textual feature set. Finally, we apply XGBoost to train a classifier and use the trained classifier to partition each comment into the corresponding class. We experimentally investigate the effectiveness of our approach on a public dataset, including 62 566 code comments from 10 open-source projects. Experimental results show that our approach achieves 56.66% in terms of macroaveraged precision, 59.07% in terms of macroaveraged recall, and 55.77% in terms of macroaveraged F-measure on average, and outperforms the natural language processing based method by 4.98%, 5.32%, and 3.17%, respectively. In addition, the experimental results also demonstrate that the data augmentation strategy is effective in improving the effectiveness of our approach.

Compiling source code effectively to meet various criteria is a critical task in software engineering. Especially, code size reduction has attracted much attention from both industry and academia due to the requirement of resource utilization. Generally, developers rely on compiler optimization passes to realize code size reduction. However, it is impractical to select a desirable optimization sequence manually since a wide variety of optimization passes are integrated into a compiler. Evolutionary algorithms offer an impressive way to alleviate this problem. Nevertheless, previous approaches fail to balance the exploitation and exploration of the search space. Moreover, the expensive fitness evaluation requires actual compilation, which makes the evolution rather time-consuming. To tackle the challenges, we propose a novel approach SMARTEST, which characterizes the systematic exploitation of a huge volume of historical compilation information. Specifically, SMARTEST comprises two components: 1) a local search operator to enhance the solution quality; and 2) a data-driven surrogate model to avoid expensive fitness evaluation. We evaluate the effectiveness of SMARTEST over the cBench benchmark suite. Experimental results indicate that SMARTEST outperforms the standard level -Os by 2.17% on average, and achieves 1.2 times code size reduction compared with the genetic algorithm. Furthermore, experimental results over the benchmark suite evidently show that SMARTEST gets a better result and takes less actual fitness evaluations than its variants, which demonstrates the contribution of the local search and the surrogate model.

Correlation acts as a critical role in the tracking field, especially in recent popular Siamese-based trackers. The correlation operation is a simple fusion manner to consider the similarity between the template and the search region. However, the correlation operation itself is a local linear matching process, leading to lose semantic information and fall into local optimum easily, which may be the bottleneck of designing high-accuracy tracking algorithms. Is there any better feature fusion method than correlation? To address this issue, inspired by Transformer, this work presents a novel attention-based feature fusion network, which effectively combines the template and search region features solely using attention. Specifically, the proposed method includes an ego-context augment module based on self-attention and a cross-feature augment module based on cross-attention. Finally, we present a Transformer tracking (named TransT) method based on the Siamese-like feature extraction backbone, the designed attention-based fusion mechanism, and the classification and regression head. Experiments show that our TransT achieves very promising results on six challenging datasets, especially on large-scale LaSOT, TrackingNet, and GOT-10k benchmarks. Our tracker runs at approximatively 50 fps on GPU. Code and models are available at https://github.com/chenxin-dlut/TransT.

Recent breakthroughs of Neural Architecture Search (NAS) extend the field’s research scope towards a broader range of vision tasks and more diversified search spaces. While existing NAS methods mostly design architectures on a single task, algorithms that look beyond single-task search are surging to pursue a more efficient and universal solution across various tasks. Many of them leverage transfer learning and seek to preserve, reuse, and refine network design knowledge to achieve higher efficiency in future tasks. However, the enormous computational cost and experiment complexity of cross-task NAS are imposing barriers for valuable research in this direction. Existing NAS benchmarks all focus on one type of vision task, i.e., classification. In this work, we propose TransNAS-Bench-101, a benchmark dataset containing network performance across seven tasks, covering classification, regression, pixel-level prediction, and self-supervised tasks. This diversity provides opportunities to transfer NAS methods among tasks and allows for more complex transfer schemes to evolve. We explore two fundamentally different types of search space: cell-level search space and macro-level search space. With 7,352 backbones evaluated on seven tasks, 51,464 trained models with detailed training information are provided. With TransNAS-Bench-101, we hope to encourage the advent of exceptional NAS algorithms that raise cross-task search efficiency and generalizability to the next level. Our dataset and code will be available at Mindspore1and VEGA2.

Current 3D object detection paradigms highly rely on extensive annotation efforts, which makes them not practical in many real-world industrial applications. Inspired by that a human driver can keep accumulating experiences from self-exploring the roads without any tutor’s guidance, we first step forwards to explore a simple yet effective self-supervised learning framework tailored for LiDAR-based 3D object detection. Although the self-supervised pipeline has achieved great success in 2D domain, the characteristic challenges (e.g., complex geometry structure and various 3D object views) encountered in the 3D domain hinder the direct adoption of existing techniques that often contrast the 2D augmented data or cluster single-view features. Here we present a novel self-supervised 3D Object detection framework that seamlessly integrates the geometry-aware contrast and clustering harmonization to lift the unsupervised 3D representation learning, named GCC-3D. First, GCC-3D introduces a Geometric-Aware Contrastive objective to learn spatial-sensitive local structure representation. This objective enforces the spatially close voxels to have high feature similarity. Second, a Pseudo-Instance Clustering harmonization mechanism is proposed to encourage that different views of pseudo-instances should have consistent similarities to clustering prototype centers. This module endows our model semantic discriminative capacity. Extensive experiments demonstrate our GCC-3D achieves significant performance improvement on data-efficient 3D object detection benchmarks (nuScenes and Waymo). Moreover, our GCC-3D framework can achieve state-of-the art performances on all popular 3D object detection benchmarks.

In software maintainence, defect report severity prediction is an important task which can effectively help developer judge the urgency of defects. However, manually classifying the severity of defect reports is very time consuming and tedious. Recently, researchers have proposed many advanced methods to automate the severity prediction of defect reports. But there is still room for improvement in performance. Therefore, in this paper, we propose a distinctive method for automatically identifying the severity of software defect reports using convolutional neural network (CNN). We first preprocess defect reports and select textual features by genetic algorithms (GA). Then, the Word2Vec model is employed to generate word vectors for the selected features in each defect report. Finally we train the classifier based on CNN and leverage the trained classifier to predict the severity of defect reports. To validate the performance of our method, we experiment with defect reports from four open source projects and compare our method with three common machine learning methods. The experimental results show that our method achieves 77.38% in terms of precision, 62.09% in terms of recall and 68.76% in terms of Fl-score on average, and outperforms the best baseline method by 11.61%, 7.23% and 9.32%, respectively.

Summary Workflow scheduling with minimum cost is one of the most challenging problems for the users who need to execute a large‐scale scientific application on a cloud platform. However, traditional methods are hard to cover the highly complex applications and ignore the billing model of the public clouds. In this paper, we address the problem of scheduling a scientific application on cloud platform from the perspective of users. First, we propose a Satisfiability Modulo Theories (SMT) based algorithm to schedule a scientific application on cloud platform, the SMT algorithm constructs the scheduling problem to first‐order logic expressions and checks the expressions by solvers, which minimizes the number of Virtual Machine instances (VMs) allocated to the application. Furthermore, due to the hourly payment of cloud, we develop a heuristic algorithm called Multiple Strategies Algorithm (MSA) which determines the minimum instance hours of a scientific application deployed on VMs. At last, we combine the proposed SMT based algorithm and the MSA to a framework named SMT‐MSA, and compare it with other outstanding algorithms in experiments, the results show that, in most of cases, our algorithms reduce more cost than the other three methods which are HEFT, MSMD and IC‐PCPD2.

Developers increasingly rely on text matching tools to analyze the relation between natural language words and APIs. However, semantic gaps, namely textual mismatches between words and APIs, negatively affect these tools. Previous studies have transformed words or APIs into low-dimensional vectors for matching; however, inaccurate results were obtained due to the failure of modeling words and APIs simultaneously. To resolve this problem, two main challenges are to be addressed: the acquisition of massive words and APIs for mining and the alignment of words and APIs for modeling. Therefore, this study proposes Word2API to effectively estimate relatedness of words and APIs. Word2API collects millions of commonly used words and APIs from code repositories to address the acquisition challenge. Then, a shuffling strategy is used to transform related words and APIs into tuples to address the alignment challenge. Using these tuples, Word2API models words and APIs simultaneously. Word2API outperforms baselines by 10-49.6 percent of relatedness estimation in terms of precision and NDCG. Word2API is also effective on solving typical software tasks, e.g., query expansion and API documents linking. A simple system with Word2API-expanded queries recommends up to 21.4 percent more related APIs for developers. Meanwhile, Word2API improves comparison algorithms by 7.9-17.4 percent in linking questions in Question&Answer communities to API documents.

In this paper, we focus on investigating function based code clone detection and leveraging the structural information to measure the similarity of code fragments in the function level. The method first combines a variant of Abstract Syntax Tree(AST) to achieve more abstract code representations by using defined node types instead of the original node representations, and then adopts a local comparison algorithm, namely Smith Waterman, to calculate the similarity scores of pairs of code fragments in the function level. Experiments conducted over the five open-source datasets show that our method can achieve 92.46% in precision on average, and outperform the comparative algorithms by up to 10.94% and 4.02%, respectively. Meanwhile, experimental results show that our method can achieve 90.73% in precision on average in code clone detection over cross-projects.

In crowdsourced mobile application testing, crowd workers help developers perform testing and submit test reports for unexpected behaviors. These submitted test reports usually provide critical information for developers to understand and reproduce the bugs. However, due to the poor performance of workers and the inconvenience of editing on mobile devices, the quality of test reports may vary sharply. At times developers have to spend a significant portion of their available resources to handle the low-quality test reports, thus heavily decreasing their efficiency. In this paper, to help developers predict whether a test report should be selected for inspection within limited resources, we propose a new framework named TERQAF to automatically model the quality of test reports. TERQAF defines a series of quantifiable indicators to measure the desirable properties of test reports and aggregates the numerical values of all indicators to determine the quality of test reports by using step transformation functions. Experiments conducted over five crowdsourced test report datasets of mobile applications show that TERQAF can correctly predict the quality of test reports with accuracy of up to 88.06% and outperform baselines by up to 23.06%. Meanwhile, the experimental results also demonstrate that the four categories of measurable indicators have positive impacts on TERQAF in evaluating the quality of test reports.

In software maintenance process, it is a fairly important activity to predict the severity of bug reports. However, manually identifying the severity of bug reports is a tedious and time-consuming task. So developing automatic judgment methods for predicting the severity of bug reports has become an urgent demand. In general, a bug report contains a lot of descriptive natural language texts, thus resulting in a high-dimensional feature set which poses serious challenges to traditionally automatic methods. Therefore, we attempt to use automatic feature selection methods to improve the performance of the severity prediction of bug reports. In this paper, we introduce a ranking-based strategy to improve existing feature selection algorithms and propose an ensemble feature selection algorithm by combining existing ones. In order to verify the performance of our method, we run experiments over the bug reports of Eclipse and Mozilla and conduct comparisons with eight commonly used feature selection methods. The experiment results show that the ranking-based strategy can effectively improve the performance of the severity prediction of bug reports by up to 54.76% on average in terms of [Formula: see text]-measure, and it also can significantly reduce the dimension of the feature set. Meanwhile, the ensemble feature selection method can get better results than a single feature selection algorithm.

DevOps is a new approach to drive a seamless Application (App) cycle from development to delivery. As a critical part to promote the successful implementation of DevOps, testing can significantly improve team productivity and reliably deliver user experience. However, it is difficult to use traditional testing to cover diverse mobile phones, network environments, operating systems, and so on. Hence, many large companies crowdsource their App testing tasks to workers from open platforms. In crowdsourced testing, test reports submitted by workers may be highly redundant, and their quality may vary sharply. Meanwhile, multi-bug test reports may be submitted, and their root causes are hard to diagnose. Hence, it is a time-consuming and tedious task for developers to manually inspect these test reports. To help developers address the above challenges, we issue the new problem of Fuzzy Clustering Test Reports (FULTER). Aiming to resolve FULTER, a series of barriers need to be overcome. In this study, we propose a new framework named Test Report Fuzzy Clustering Framework (TERFUR) by aggregating redundant and multi-bug test reports into clusters to reduce the number of inspected test reports. First, we construct a filter to remove invalid test reports to break through the invalid barrier . Then, a preprocessor is built to enhance the descriptions of short test reports to break through the uneven barrier . Last, a two-phase merging algorithm is proposed to partition redundant and multi-bug test reports into clusters that can break through the multi-bug barrier . Experimental results over 1,728 test reports from five industrial Apps show that TERFUR can cluster test reports by up to 78.15% in terms of AverageP , 78.41% in terms of AverageR , and 75.82% in terms of AverageF1 and outperform comparative methods by up to 31.69%, 33.06%, and 24.55%, respectively. In addition, the effectiveness of TERFUR is validated in prioritizing test reports for manual inspection.

General-Purpose Graphic Processing Units (GPGPU) have been widely used in high performance computing as application accelerators due to their massive parallelism and high throughput. A GPGPU generally contains two layers of schedulers, a cooperative-thread-array (CTA) scheduler and a warp scheduler, which administer the thread level parallelism (TLP). Previous research shows the maximized TLP does not always deliver the optimal performance. Unfortunately, existing warp scheduling schemes do not optimize TLP at runtime, which is impossible to fit various access patterns for diverse applications. Dynamic TLP optimization in the warp scheduler remains a challenge to exploit the GPGPU highly-parallel compute power.

Nowadays, dependability is of paramount importance in modern distributed storage systems. A challenging issue to deploy a storage system with certain dependability requirements or improve existing systems' dependability is how to comprehensively and efficiently characterize the dependability of those systems. In this paper, we present a two-layer Hidden Markov Model (HMM) to characterize the dependability of a distributed storage system, focusing on the layer of parallel file system. By training the model with observable measurements under faulty scenarios, such as I/O performance, we quantify the system dependability via a tuple of state transition probability, service degradation, and fault latency under those scenarios. Our experimental results on a distributed storage system with PVFS (Parallel Virtual File System) demonstrate the effectiveness of our HMM-based approach, which efficiently captures the behavior patterns of the target system under disk faults and memory overusage.

Correlated failures have recently gained more attention in the research of failures in large scale systems. Recent studies have pointed out the negative effect of ignoring such failures when designing a fault tolerant scheme for large scale systems. In this paper, we explore the behaviors of temporal correlated failures arising from cyclic dependency among task nodes via an abstract model. Using this model, we find that fast failure propagation and slow recovery from failures are two dominant factors which make recovering from such failures much difficult. To efficiently stop failure propagation and shorten the total recovering time, we propose a recovery protocol called GCCTS (group-based coordinated checkpointing and task suspending) against temporal correlated failures.

Re-texturing is an image editing method which changes the material appearance of an object. In this paper, a new method for re-texturing by extracting object interactively from an image is presented. Rather than relying on tedium of manual object selection, a new re-texturing strategy which combines image matting is proposed. Given images or photographs with object to be re-textured, only several scribbles which indicate the foreground object and the background image are needed. The alpha matte outlined the object is automatically produced by optimizing an improved cost function of image matting. According to the alpha matte, re-textured result can be obtained by selection of new texture. Comparing with related work, the method demonstrated in this paper requires considerably less manual effort and can obtain satisfied results.