RESUMEN
The evolution of super-resolution (SR) technology has seen significant advancements through the adoption of deep learning methods. However, the deployment of such models by resource-constrained devices necessitates models that not only perform efficiently, but also conserve computational resources. Binary neural networks (BNNs) offer a promising solution by minimizing the data precision to binary levels, thus reducing the computational complexity and memory requirements. However, for BNNs, an effective architecture is essential due to their inherent limitations in representing information. Designing such architectures traditionally requires extensive computational resources and time. With the advancement in neural architecture search (NAS), differentiable NAS has emerged as an attractive solution for efficiently crafting network structures. In this paper, we introduce a novel and efficient binary network search method tailored for image super-resolution tasks. We adapt the search space specifically for super resolution to ensure it is optimally suited for the requirements of such tasks. Furthermore, we incorporate Libra Parameter Binarization (Libra-PB) to maximize information retention during forward propagation. Our experimental results demonstrate that the network structures generated by our method require only a third of the parameters, compared to conventional methods, and yet deliver comparable performance.
RESUMEN
Binary neural networks (BNNs) are variations of artificial/deep neural network (ANN/DNN) architectures that constrain the real values of weights to the binary set of numbers {-1,1}. By using binary values, BNNs can convert matrix multiplications into bitwise operations, which accelerates both training and inference and reduces hardware complexity and model sizes for implementation. Compared to traditional deep learning architectures, BNNs are a good choice for implementation in resource-constrained devices like FPGAs and ASICs. However, BNNs have the disadvantage of reduced performance and accuracy because of the tradeoff due to binarization. Over the years, this has attracted the attention of the research community to overcome the performance gap of BNNs, and several architectures have been proposed. In this paper, we provide a comprehensive review of BNNs for implementation in FPGA hardware. The survey covers different aspects, such as BNN architectures and variants, design and tool flows for FPGAs, and various applications for BNNs. The final part of the paper gives some benchmark works and design tools for implementing BNNs in FPGAs based on established datasets used by the research community.
RESUMEN
Recent studies have shown the efficacy of mobile elements in optimizing the energy consumption of sensor nodes. Current data collection approaches for waste management applications focus on exploiting IoT-enabled technologies. However, these techniques are no longer sustainable in the context of smart city (SC) waste management applications due to the emergence of large-scale wireless sensor networks (LS-WSNs) in smart cities with sensor-based big data architectures. This paper proposes an energy-efficient swarm intelligence (SI) Internet of Vehicles (IoV)-based technique for opportunistic data collection and traffic engineering for SC waste management strategies. This is a novel IoV-based architecture exploiting the potential of vehicular networks for SC waste management strategies. The proposed technique involves deploying multiple data collector vehicles (DCVs) traversing the entire network for data gathering via a single-hop transmission. However, employing multiple DCVs comes with additional challenges including costs and network complexity. Thus, this paper proposes analytical-based methods to investigate critical tradeoffs in optimizing energy consumption for big data collection and transmission in an LS-WSN such as (1) finding the optimal number of data collector vehicles (DCVs) required in the network and (2) determining the optimal number of data collection points (DCPs) for the DCVs. These critical issues affect efficient SC waste management and have been overlooked by previous studies exploring waste management strategies. Simulation-based experiments using SI-based routing protocols validate the efficacy of the proposed method in terms of the evaluation metrics.
RESUMEN
This paper investigates multimodal sensor architectures with deep learning for audio-visual speech recognition, focusing on in-the-wild scenarios. The term "in the wild" is used to describe AVSR for unconstrained natural-language audio streams and video-stream modalities. Audio-visual speech recognition (AVSR) is a speech-recognition task that leverages both an audio input of a human voice and an aligned visual input of lip motions. However, since in-the-wild scenarios can include more noise, AVSR's performance is affected. Here, we propose new improvements for AVSR models by incorporating data-augmentation techniques to generate more data samples for building the classification models. For the data-augmentation techniques, we utilized a combination of conventional approaches (e.g., flips and rotations), as well as newer approaches, such as generative adversarial networks (GANs). To validate the approaches, we used augmented data from well-known datasets (LRS2-Lip Reading Sentences 2 and LRS3) in the training process and testing was performed using the original data. The study and experimental results indicated that the proposed AVSR model and framework, combined with the augmentation approach, enhanced the performance of the AVSR framework in the wild for noisy datasets. Furthermore, in this study, we discuss the domains of automatic speech recognition (ASR) architectures and audio-visual speech recognition (AVSR) architectures and give a concise summary of the AVSR models that have been proposed.
Asunto(s)
Aprendizaje Profundo , Percepción del Habla , Humanos , Habla , LenguajeRESUMEN
This paper investigated the utility of drone-based environmental monitoring to assist with forest inventory in Queensland private native forests (PNF). The research aimed to build capabilities to carry out forest inventory more efficiently without the need to rely on laborious field assessments. The use of drone-derived images and the subsequent application of digital photogrammetry to obtain information about PNFs are underinvestigated in southeast Queensland vegetation types. In this study, we used image processing to separate individual trees and digital photogrammetry to derive a canopy height model (CHM). The study was supported with tree height data collected in the field for one site. The paper addressed the research question "How well do drone-derived point clouds estimate the height of trees in PNF ecosystems?" The study indicated that a drone with a basic RGB camera can estimate tree height with good confidence. The results can potentially be applied across multiple land tenures and similar forest types. This informs the development of drone-based and remote-sensing image-processing methods, which will lead to improved forest inventories, thereby providing forest managers with recent, accurate, and efficient information on forest resources.
Asunto(s)
Ecosistema , Dispositivos Aéreos No Tripulados , Bosques , Árboles , Monitoreo del Ambiente/métodos , Tecnología de Sensores Remotos/métodosRESUMEN
This paper proposes an approach for modeling and mining curriculum Big data from real-world education datasets crawled online from university websites in Australia. It addresses the scenario to give a student a study plan to complete a course by accumulating credits on top of subjects he or she has completed. One challenge to be addressed is that subjects with similar bs from different universities may put barriers for setting up a reasonable, time-saving learning path because the student may be unable to distinguish them before an intensive research on all subjects related to the degree from the universities. We used concept graph-based learning techniques and discuss data representations and techniques which are more suited for large datasets. We created ground truth of subjects relations and subject's description with Bag of Words representations based on natural language processing. The generated ground truth was used to train a model, which summarizes a subject network and a concepts graph, where the concepts are automatically extracted from the subject descriptions across all the universities. The practical challenges to collect and extract the data from the university websites are also discussed in the paper. The work was validated on nineteen real-world education datasets crawled online from university websites in Australia and showed good performance.
Asunto(s)
Curriculum , Universidades , Australia , Humanos , Aprendizaje , EstudiantesRESUMEN
Application Specific Internet of Things (ASIoTs) has recently been proposed to address specific requirements for IoT. The objective of this paper is to serve as a framework for the design of ASIoTs using biometrics as the application. This paper provides comprehensive discussions for an ASIoT architecture considering the requirements for biometrics-based security, multimedia content and Big data applications. A comprehensive architecture for Biometrics-based IoT (BiometricIoT) and Big data applications needs to address three challenges: 1) IoT devices are hardware-constrained and cannot afford resource-demanding cryptographic protocols; 2) Biometrics devices introduce multimedia data content due to different biometric traits; and 3) The rapid growth of biometrics-based IoT devices and content creates large amounts of data for computational processing. The proposed BiometricIoT architecture consists of seven layers which have been designed to handle the challenges for biometrics applications and decision making. The latter part of the paper gives discussions for design factors for the BiometricIoT from four perspectives: 1) parallel divide-and-conquer (D&C) computation; 2) computational complexity; 3) device security; and 4) algorithm efficacies. Experimental results are given to validate the effectiveness of the D&C approach. The paper motivates the further research towards the research and development of ASIoTs for biometrics applications.
Asunto(s)
Internet de las Cosas , Algoritmos , Macrodatos , Seguridad Computacional , Computadores , InternetRESUMEN
In machine learning, the nature of the dataset itself such as convexity of the data point sets affects the right choice of clustering algorithm to give good performance. This brief paper first focuses on how data convexity influences the clustering performance on biomedical datasets. Then it addresses the main challenges of two well-known clustering groups which are centroid-based and density-based clustering. These techniques typically require a set of parameters to be provided by the user before the algorithms can perform well in terms of good clustering and give the optimal number of clusters. Two parameter independent clustering techniques utilizing unique neighborhood sets (UNSs) called Parameter Independent Convex Centroid-based Clustering (ConvexClust) for convex-dominated datasets and Parameter Independent Non-Convex Density-based Clustering (NonConvexClust) for nonconvex-dominated datasets are introduced. The ConvexClust and NonConvex Clust algorithms are extensively evaluated on real-world biomedical datasets. Their performances are also compared with other clustering algorithms using evaluation criteria such as SSE, entropy and purity. The results have revealed the good performance of the proposed parameter-independent clustering techniques and also shown that most of the biomedical datasets in the experiments demonstrated their tendency towards convex-dominated data point sets.
Asunto(s)
Investigación Biomédica , Análisis por Conglomerados , Bases de Datos Factuales , Algoritmos , Biología Computacional , Humanos , Neoplasias/genética , Neoplasias/metabolismo , TranscriptomaRESUMEN
This paper presents an investigation of natural inspired intelligent computing and its corresponding application towards visual information processing systems for viticulture. The paper has three contributions: (1) a review of visual information processing applications for viticulture; (2) the development of natural inspired computing algorithms based on artificial immune system (AIS) techniques for grape berry detection; and (3) the application of the developed algorithms towards real-world grape berry images captured in natural conditions from vineyards in Australia. The AIS algorithms in (2) were developed based on a nature-inspired clonal selection algorithm (CSA) which is able to detect the arcs in the berry images with precision, based on a fitness model. The arcs detected are then extended to perform the multiple arcs and ring detectors information processing for the berry detection application. The performance of the developed algorithms were compared with traditional image processing algorithms like the circular Hough transform (CHT) and other well-known circle detection methods. The proposed AIS approach gave a Fscore of 0.71 compared with Fscores of 0.28 and 0.30 for the CHT and a parameter-free circle detection technique (RPCD) respectively.