The past decade has seen the great potential of applying deep neural network (DNN) based software to safety-critical scenarios, such as autonomous driving. Similar traditional software, DNNs could exhibit incorrect behaviors, caused by hidden defects, leading severe accidents and losses. In this paper, we propose DeepHunter, a coverage-guided fuzz testing framework for detecting defects general-purpose DNNs. To end, first metamorphic mutation strategy generate new semantically preserved...

FlowFormer [24] introduces a transformer architecture into optical flow estimation and achieves state-of-the-art performance. The core component of is the transformer-based cost-volume encoder. Inspired by recent success masked autoencoding (MAE) pretraining in unleashing transformers' capacity encoding visual representation, we propose Masked Cost Volume Autoencoding (MCVA) to enhance encoder with novel MAE scheme. Firstly, introduce block-sharing masking strategy prevent information...

We introduce VideoFlow, a novel optical flow estimation framework for videos. In contrast to previous methods that learn estimate from two frames, VideoFlow concurrently estimates bi-directional flows multiple frames are available in videos by sufficiently exploiting temporal cues.We first propose TRi-frame Optical Flow (TROF) module the center frame three-frame manner. The information of triplet is iteratively fused onto frame. To extend TROF handling more we further MOtion Propagation...

Video restoration, which aims to restore clear frames from degraded videos, has numerous important applications. The key video restoration depends on utilizing inter-frame information. However, existing deep learning methods often rely complicated network architectures, such as optical flow estimation, deformable convo-lution, and cross-frame self-attention layers, resulting in high computational costs. In this study, we propose a sim-ple yet effective framework for restoration. Our approach...

Abstract Predicting the blood–brain barrier (BBB) permeability of small-molecule compounds using a novel artificial intelligence platform is necessary for drug discovery. Machine learning and large language model on (AI) tools improve accuracy shorten time new development. The primary goal this research to develop computing models deep architectures capable predicting whether molecules can permeate human (BBB). in silico (computational) vitro (experimental) results were validated by Natural...

Radiology report generation (RRG) has gained increasing research attention because of its huge potential to mitigate medical resource shortages and aid the process disease decision making by radiologists. Recent advancements in Report Generation are largely driven improving a model's capabilities encoding single-modal feature representations, while few studies explicitly explore cross-modal alignment between image regions words. Radiologists typically focus first on abnormal before composing...

Distributed stochastic gradient descent (SGD) algorithms are widely deployed in training large-scale deep learning models, while the communication overhead among workers becomes new system bottleneck. Recently proposed sparsification techniques, especially Top-$k$ with error compensation (TopK-SGD), can significantly reduce traffic without an obvious impact on model accuracy. Some theoretical studies have been carried out to analyze convergence property of TopK-SGD. However, existing do not...

Continual Semantic Segmentation (CSS) extends static semantic segmentation by incrementally introducing new classes for training. To alleviate the catastrophic forgetting issue in CSS, a memory buffer that stores small number of samples from previous is constructed replay. However, existing methods select either randomly or based on single-factor-driven handcrafted strategy, which has no guarantee to be optimal. In this work, we propose novel sample selection mechanism selects informative...

An up-and-coming concept that seeks to transform how students learn about and study complex systems, as well industrial workers are trained, metaverse technology is characterized in this context by its use virtual simulation analysis. In work, a environment created duplicates real-world situations enables immersive interactive learning the educational metaverse. For purpose, we built digital twin of Nanyang Technological University (NTU) campus foundation, called NTUniverse. It designed an...

Wireless sensor networks have emerged as an exciting technology for a wide range of important applications that acquire and process information from the physical world. Grid computing has evolved standards-based approach coordinated resource sharing. Sensor grids combine these two promising technologies by extending grid paradigm to sharing resources in wireless networks. There are several issues challenges design grids. In this paper, we propose architecture, called scalable proxy-based...

Unified Memory is an emerging technology which supported by CUDA 6.X. Before 6.X, the existing programming model relies on programmers to explicitly manage data between CPU and GPU hence increases complexity. 6.X provides a new called as provide that defines memory space single coherent (imaging same common address space). The system manages access without explicit copy functions. This paper evaluate through different applications GPUs show users how use of efficiently. include Diffusion3D...

Purpose Recently, various Twitter Sentiment Analysis (TSA) techniques have been developed, but little has paid attention to the microblogging feature – emojis, and few works conducted on multi-class sentiment analysis of tweets. The purpose this paper is consider popularity emojis investigate feasibility an emoji training heuristic for classification Tweets from “2016 Orlando nightclub shooting” were used as a source study. Besides, study also aims demonstrate how mapping can contribute...

Purpose Emoji has become an essential component of any digital communication and its importance can be attested to by sustained popularity widespread use. However, research in Emojis is rarely seen due the lack data at a greater scale. The purpose this paper systematically analyse compare usage cross-cultural manner. Design/methodology/approach This conducted empirical analysis using large-scale, cross-regional emoji set from Twitter, platform where limited 140 characters allowance made it...

Brain age is an imaging-based biomarker with excellent feasibility for characterizing individual brain health and may serve as a single quantitative index clinical domain-specific usage. has been successfully estimated using extensive neuroimaging data from healthy participants various feature extraction conventional machine learning (ML) approaches. Recently, several end-to-end deep (DL) analytical frameworks have proposed alternative approaches to predict higher accuracy. However, the...

Extracting and using class-discriminative features is critical for fine-grained recognition. Existing works have demonstrated the possibility of applying deep CNNs to exploit that distinguish similar classes. However, suffer from problems including frequency bias loss detailed local information, which restricts performance recognizing categories. To address challenge, we propose a novel texture branch as complimentary CNN feature extraction. We innovatively utilize Gabor filters powerful...

As a representative cyber-physical system (CPS), robotic manipulators have been widely adopted in various academic research and industrial processes, indicating their potential to act as universal interface between the cyber physical worlds. Recent studies robotics manipulation started employing artificial intelligence (AI) approaches controllers achieve better adaptability performance. However, inherent challenge of explaining AI components introduces uncertainty unreliability these...

In the realm of machine learning in healthcare, federated (FL) is often recognized as a practical solution for addressing issues related to data privacy and distribution. However, many real-world datasets are not identically independently distributed (non-IID). That is, characteristics differ from one institute another. Non-IID poses challenges convergence FL models, such client drifting, where model weights drift towards local optima instead global optimum. As solution, leveraging synthetic...

Bayesian inverse problems arise in various scientific and engineering domains, solving them can be computationally demanding. This is especially the case for governed by partial differential equations, where repeated evaluation of forward operator extremely expensive. Recent advances Deep Learning (DL)-based surrogate models have shown promising potential to accelerate solution such problems. However, despite their ability learn from complex data, DL-based generally cannot match accuracy...

Abstract We present an efficient tensor-network-based approach for simulating large-scale quantum circuits exemplified by Quantum Support Vector Machines (QSVMs). Experimentally, leveraging the cuTensorNet library on multiple GPUs, our method effectively reduces exponential runtime growth to near-quadratic scaling with respect number of qubits in practical scenarios. Traditional state-vector simulations become computationally infeasible beyond approximately 50 qubits; contrast, simulator...

Online continual learning for image classification is crucial models to adapt new data while retaining knowledge of previously learned tasks. This capability essential address real-world challenges involving dynamic environments and evolving distributions. Traditional approaches predominantly employ Convolutional Neural Networks, which are limited processing images as grids primarily capture local patterns rather than relational information. Although the emergence transformer architectures...

Modern large language models (LLMs) employ various forms of logical inference, both implicitly and explicitly, when addressing reasoning tasks. Understanding how to optimally leverage these inference paradigms is critical for advancing LLMs' capabilities. This paper adopts an exploratory approach by introducing a controlled evaluation environment analogical -- fundamental cognitive task that systematically parameterized across three dimensions: modality (textual, visual, symbolic),...

Accurate and efficient climate simulations are crucial for understanding Earth's evolving climate. However, current general circulation models (GCMs) face challenges in capturing unresolved physical processes, such as cloud convection. A common solution is to adopt resolving models, that provide more accurate results than the standard subgrid parametrisation schemes typically used GCMs. also referred super paramtetrizations, remain computationally prohibitive. Hybrid modeling, which...