Structured pruning for large language models (LLMs) has garnered significant academic interest due to its ability to efficiently compress and accelerate LLMs by eliminating redundant weight groups at a coarse-grained granularity. Current structured pruning methods for LLMs typically depend on a singular granularity for assessing weight importance, resulting in notable performance degradation in downstream tasks. Intriguingly, our empirical investigations reveal that utilizing unstructured pruning, which achieves better performance retention by pruning weights at a finer granularity, \emph{i.e.}, individual weights, yields significantly varied sparse LLM structures when juxtaposed to structured pruning. This suggests that evaluating both holistic and individual assessments for weight importance are essential for LLM pruning. Building on this insight, we introduce the Hybrid-grained Weight Importance Assessment (HyWIA), a novel method that merges fine-grained and coarse-grained evaluations of weight importance for the pruning of LLMs. Leveraging an attention mechanism, HyWIA adaptively determines the optimal blend of granularity in weight importance assessments in an end-to-end pruning manner. Extensive experiments on LLaMA-V1/V2, Vicuna, Baichuan, and Bloom across various benchmarks demonstrate the effectiveness of HyWIA in pruning LLMs. For example, HyWIA surpasses the cutting-edge LLM-Pruner by an average margin of 2.82% in accuracy across seven downstream tasks when pruning LLaMA-7B by 50%.

In Computer Vision (CV), the deployment of Convolutional Neural Networks (CNNs) is often hindered by their substantial computational requirements and large labeled datasets. Self-supervised learning (SSL) serves as an effective approach to reducing the reliance on labeled data with the option of augmentation methods to infer and train CNNs. Excluding irrelevant features accelerates learning and improves optimization. We propose a Field-Programmable Gate Array (FPGA)-based hardware accelerator architecture tailored for SSL framework, leveraging its parallelism and reconfigurability to expedite block matching, optimize sparse convolutions, and manage data reuse, significantly improving resource and energy efficiency. The implementation and evaluation of our work on Xilinx ZCU102 FPGA working at 200 MHz confirm that the similarity finding part's FPGA accelerations with a low hardware overhead generates a latency of 0.0106 seconds, surpassing GPU and CPU, and in the sparse CNN's FPGA acceleration part, with the processing of VGG16 and ResNet50, compared with the related FPGA-based works, our design claims a maximum of 3.08× throughput improvement and 1.5× in energy efficiency.

Generative models have achieved impressive results in text to image tasks, significantly advancing visual content creation. However, this progress comes at a cost, as such models rely heavily on large-scale training data and may unintentionally replicate copyrighted elements, creating serious legal and ethical challenges for real-world deployment. To address these concerns, researchers have proposed various strategies to mitigate copyright risks, most of which are prompt based methods that filter or rewrite user inputs to prevent explicit infringement. While effective in handling obvious cases, these approaches often fall short in more subtle situations, where seemingly benign prompts can still lead to infringing outputs. To address these limitations, this paper introduces Assessing and Mitigating Copyright Risks (AMCR), a comprehensive framework which i) builds upon prompt-based strategies by systematically restructuring risky prompts into safe and non-sensitive forms, ii) detects partial infringements through attention-based similarity analysis, and iii) adaptively mitigates risks during generation to reduce copyright violations without compromising image quality. Extensive experiments validate the effectiveness of AMCR in revealing and mitigating latent copyright risks, offering practical insights and benchmarks for the safer deployment of generative models.

Fine-tuning helps large language models (LLM) recover degraded information and enhance task performance. Although Low-Rank Adaptation (LoRA) is widely used and effective for fine-tuning, we have observed that its scaling factor can limit or even reduce performance as the rank size increases. To address this issue, we propose RoRA (Rank-adaptive Reliability Optimization), a simple yet effective method for optimizing LoRA’s scaling factor. By replacing α/r with $\alpha /\sqrt r $, RoRA ensures improved performance as rank size increases. Moreover, RoRA enhances low-rank adaptation in fine-tuning uncompressed models and excels in the more challenging task of accuracy recovery when fine-tuning pruned models. Extensive experiments demonstrate the effectiveness of RoRA in fine-tuning both uncompressed and pruned models. RoRA surpasses the state-of-the-art (SOTA) in average accuracy and robustness on LLaMA-7B/13B, LLaMA2-7B, and LLaMA3-8B, specifically outperforming LoRA and DoRA by 6.5% and 2.9% on LLaMA-7B, respectively. In pruned model fine-tuning, RoRA shows significant advantages; for SHEARED-LLAMA-1.3, a LLaMA-7B with 81.4% pruning, RoRA achieves 5.7% higher average accuracy than LoRA and 3.9% higher than DoRA.

Zeroth-order (ZO) optimization is an emerging deep neural network (DNN) training paradigm that offers computational simplicity and memory savings. However, this seemingly promising approach faces a significant and long-ignored challenge. ZO requires generating a substantial number of Gaussian random numbers, which poses significant difficulties and even makes it infeasible for hardware platforms, such as FPGAs and ASICs. In this paper, we identify this critical issue, which arises from the mismatch between algorithm and hardware designers. To address this issue, we proposed PeZO, a perturbation-efficient ZO framework. Specifically, we design random number reuse strategies to significantly reduce the demand for random number generation and introduce a hardware-friendly adaptive scaling method to replace the costly Gaussian distribution with a uniform distribution. Our experiments show that PeZO reduces the required LUTs and FFs for random number generation by 48.6% and 12.7%, and saves at maximum 86% power consumption, all without compromising training performance, making ZO optimization feasible for on-device training. To the best of our knowledge, we are the first to explore the potential of on-device ZO optimization, providing valuable insights for future research.

Self-supervised learning (SSL) offers a compelling solution to the challenge of extensive labeled data requirements in traditional supervised learning. With the proven success of Vision Transformers (ViTs) in supervised tasks, there is increasing interest in adapting them for SSL frameworks. However, the high computational demands of SSL pose substantial challenges, particularly on resource-limited platforms like edge devices, despite its ability to achieve high accuracy without labeled data. Recent studies in supervised learning have shown that token pruning can reduce training costs by removing less informative tokens without compromising accuracy. However, SSL’s dual-branch encoders make traditional single-branch pruning strategies less effective, as they fail to account for the critical cross-branch similarity information, leading to reduced accuracy in SSL. To this end, we introduce SimPrune, a novel token pruning strategy designed for ViTs in SSL. SimPrune leverages cross-branch similarity information to efficiently prune tokens, retaining essential semantic information across dual branches. Additionally, we incorporate a difficulty-aware pruning strategy to further enhance SimPrune's effectiveness. Experimental results show that our proposed approach effectively reduces training computation while maintaining accuracy. Specifically, our approach offers 24\% savings in training costs compared to SSL baseline, without sacrificing accuracy.

Graph Neural Networks (GNNs) have excelled in diverse applications due to their outstanding predictive performance, yet they often overlook fairness considerations, prompting numerous recent efforts to address this societal concern. However, most fair GNNs assume complete demographics by design, which is impractical in most real-world socially sensitive applications due to privacy, legal, or regulatory restrictions. For example, the Consumer Financial Protection Bureau (CFPB) mandates that creditors ensure fairness without requesting or collecting information about an applicant’s race, religion, nationality, sex, or other demographics. To this end, this paper proposes fairGNN-WOD, a first-of-its-kind framework that considers mitigating unfairness in graph learning without using demographic information. In addition, this paper provides a theoretical perspective on analyzing bias in node representations and establishes the relationship between utility and fairness objectives. Experiments on three real-world graph datasets illustrate that fairGNN-WOD outperforms state-of-the-art baselines in achieving fairness but also maintains comparable prediction performance.

Real‐world datasets usually contain multiple attributes, making it essential to ensure fairness across all of them simultaneously. However, different attributes may vary in difficulty, and no existing approaches have effectively addressed this issue. Consequently, an attribute‐adaptive strategy is needed to achieve fairness for all attributes. Multi‐task Learning (MTL) leverages shared information to optimize multiple tasks concurrently, while Sparsely‐Gated Mixture‐of‐Experts (SMoE) can dynamically allocate computational resources to the most needed tasks. In this work, we formulate multi‐attribute fairness issue as an MTL problem and employ SMoE to achieve desirable performance across all attributes simultaneously. We first analyze the feasibility and find the potentiality by formalizing multi-attribute fairness problem into a MTL problem and mitigating it by using SMoE. However, vanilla SMoE could lead to over-utilization problem which causes sub-optimal performance. We then proposed an innovative SMoE framework for multi-attribute fair image classification, which further improves multi-attribute fairness by redesigning the MoE layer and routing policy with fairness consideration. Extensive experiments demonstrated the effectiveness. Taking a DeiT-Small as the backbone, we achieve 77.25% and 86.01% accuracy on the ISIC2019 and CelebA dataset respectively with Multi-attribute Predictive Quality Disparity (PQD) score of 0.801 and 0.787, beating current state-of-the-art methods Muffin, InfoFair and MultiFair.

Large language models (LLMs) excel across various tasks, but standard first-order (FO) fine-tuning demands considerable memory, significantly limiting real-world deployment. Recently, zeroth-order (ZO) optimization stood out as a promising memory-efficient training paradigm, avoiding backward passes and relying solely on forward passes for gradient estimation, making it attractive for resource-constrained scenarios. However, ZO method lags far behind FO method in both convergence speed and accuracy. To bridge the gap, we introduce a novel layer-wise divergence analysis that uncovers the distinct update pattern of FO and ZO optimization. Aiming to resemble the learning capacity of FO method from the findings, we propose \textbf{Di}vergence-driven \textbf{Z}eroth-\textbf{O}rder (\textbf{DiZO}) optimization. DiZO conducts divergence-driven layer adaptation by incorporating projections to ZO updates, generating diverse-magnitude updates precisely scaled to layer-wise individual optimization needs. Our results demonstrate that DiZO significantly reduces the needed iterations for convergence without sacrificing throughput, cutting training GPU hours by up to 48\% on various datasets. Moreover, DiZO consistently outperforms the representative ZO baselines in fine-tuning RoBERTa-large, OPT-series, and Llama-series on downstream tasks and, in some cases, even surpasses memory-intensive FO fine-tuning. Our code is released at \url{https://github.com/Skilteee/DiZO}.

It is challenging to deploy 3D Convolutional Neural Networks (3D CNNs) on mobile devices, specifically if both real-time execution and high inference accuracy are in demand, because the increasingly large model size and complex model structure of 3D CNNs usually require tremendous computation and memory resources. Weight pruning is proposed to mitigate this challenge. However, existing pruning is either not compatible with modern parallel architectures, resulting in long inference latency or subject to significant accuracy degradation. This article proposes an end-to-end 3D CNN acceleration framework based on pruning/compilation co-design called Mobile-3DCNN that consists of two parts: a novel, fine-grained structured pruning enhanced by a prune/Winograd adaptive selection (that is mobile-hardware-friendly and can achieve high pruning accuracy), and a set of compiler optimization and code generation techniques enabled by our pruning (to fully transform the pruning benefit to real performance gains). The evaluation demonstrates that Mobile-3DCNN outperforms state-of-the-art end-to-end DNN acceleration frameworks that support 3D CNN execution on mobile devices, Alibaba Mobile Neural Networks and Pytorch-Mobile with speedup up to 34× with minor accuracy degradation, proving it is possible to execute high-accuracy large 3D CNNs on mobile devices in real-time (or even ultra-real-time).

Autonomous driving platforms encounter diverse driving scenarios, each with varying hardware resources and precision requirements. Given the computational limitations of embedded devices, it is crucial to consider computing costs when deploying on target platforms like the DRIVE PX 2. Our objective is to customize the semantic segmentation network according to the computing power and specific scenarios of autonomous driving hardware. We implement dynamic adaptability through a three-tier control mechanism—width multiplier, classifier depth, and classifier kernel—allowing fine-grained control over model components based on hardware constraints and task requirements. This adaptability facilitates broad model scaling, targeted refinement of the final layers, and scenario-specific optimization of kernel sizes, leading to improved resource allocation and performance. Additionally, we leverage Bayesian Optimization with surrogate modeling to efficiently explore hyperparameter spaces under tight computational budgets. Our approach addresses scenario-specific and task-specific requirements through automatic parameter search, accommodating the unique computational complexity and accuracy needs of autonomous driving. It scales its multiply-accumulate operations (MACs) for task-specific learning adaptation (TSLA), resulting in alternative configurations tailored to diverse self-driving tasks. These TSLA customizations maximize computational capacity and model accuracy, optimizing hardware utilization.

Vision-language foundational models have achieved commendable results on related tasks. However, their application to medical tasks is still limited due to issues arising from data biases. Currently, leveraging existing foundational models to improve medical tasks remains a challenge. To this end, this paper proposes a strong multimodal representation learning method based on cross-domain distillation handling structural Magnetic Resonance Imaging (sMRI), Positron Emission Computed Tomograph (PET) images, and mini-mental state examination (MMSE) score for Alzheimer’s disease (AD) classification. Specifically, we establish a text-to-image cross-domain distillation learning framework, enabling a text encoder pre-trained on general visual recognition tasks to guide the training of sMRI and PET image feature extractors. Simultaneously, positional encoding is used to extract the magnitude features of MMSE scores. Based on the multimodal representations extracted from sMRI, PET images, and MMSE scores, we perform a self-attention operation equipped with a gating mechanism for multimodal feature fusion. This mechanism controls the contribution of each modality representation to the classification decision, dynamically strengthening or weakening specific modality representations and helping construct stronger fused features for AD classification. Our method undergoes 5-fold cross-validation on the widely used ADNI dataset, and comparative experimental results demonstrate that our method achieves advanced performance in two AD-related binary classification tasks.

Virtual try-on system that transfers clothes onto the target person has attracted rapidly. Previous works use the affine or Thin Plate Spline (TPS) transformation for clothes warping and directly learn a composition mask to fuse the warped clothes and the person image, which usually causes rough shape and blurry details due to the poor warping mechanism and lack of human structure. In this paper, we propose a novel Dense Flow guided Virtual Try-On Network (DF-VTON), which contains a progressive warping network for clothes deformation, and a personalized fitting network for the fusion of the clothes and the person image. Specifically, given a target clothes image and a reference person image, the progressive warping network generates the dense flow in a progressive way and multi-scale views. The personalized fitting network aims to fuse warped clothes and person image seemly by using multi-scale composition masks. Extensive experiments on two challenging benchmarks demonstrate the superiority of our proposed DF-VTON over existing strong baselines with realistic high-resolution try-on results.

From monocular RGB images captured in the wild, detecting multi-body 3D meshes in physical sizes and locations is notoriously difficult due to the diverse visual ambiguity and lack of explicit depth measurement. Modern DNN approaches made numerous advances based on either two-stage Region-of-Interests(RoI)-Align or single-stage fixed Field-of-View (FoV) detector frameworks for two main subtasks: local pelvis-centered mesh regression and global body-to-camera translation regression. However, sub-meter-level physical-space monocular mesh detection is still out of reach by existing solutions. In this paper, we recognize two common drawbacks: (1) The local meshes are usually estimated without explicitly aligning body features under image-space scaling, occlusion, and truncation; (2) The global translations are estimated based on a weak-perspective assumption, which tricks the network into prioritizing image-space (front-view) mesh alignment and leads to inaccurate mesh depth. We introduce Physical-space Multi-body Mesh Detection (PMMD), in which (1) Locally, we preserve the body aspect ratio, align the body-to-RoI layout, and densely refine the person-wise RoI features for robustness; (2) Globally, we learn dense-depth-guided features to amend the body-wise local feature for physical depth estimation. With the cleaned local features and explicit local-global associations, PMMD achieves the best centimeter-level local mesh metrics and the first sub-meter-level global mesh metrics from monocular images in 3DPW and AGORA datasets.

Noninvasively and accurately predicting the epidermal growth factor receptor (EGFR) mutation status is a clinically vital problem. Moreover, further identifying the most suspicious area related to the EGFR mutation status can guide the biopsy to avoid false negatives. Deep learning methods based on computed tomography (CT) images may improve the noninvasive prediction of EGFR mutation status and potentially help clinicians guide biopsies by visual methods. Inspired by the potential inherent links between EGFR mutation status and invasiveness information, we hypothesized that the predictive performance of a deep learning network can be improved through extra utilization of the invasiveness information. Here, we created a novel explainable transformer network for EGFR classification named gated multiple instance learning transformer (GMILT) by integrating multi-instance learning and discriminative weakly supervised feature learning. Pathological invasiveness information was first introduced into the multitask model as embeddings. GMILT was trained and validated on a total of 512 patients with adenocarcinoma and tested on three datasets (the internal test dataset, the external test dataset, and The Cancer Imaging Archive (TCIA) public dataset). The performance (area under the curve (AUC) =0.772 on the internal test dataset) of GMILT exceeded that of previously published methods and radiomics-based methods (i.e., random forest and support vector machine) and attained a preferable generalization ability (AUC =0.856 in the TCIA test dataset and AUC =0.756 in the external dataset). A diameter-based subgroup analysis further verified the efficiency of our model (most of the AUCs exceeded 0.772) to noninvasively predict EGFR mutation status from computed tomography (CT) images. In addition, because our method also identified the "core area" of the most suspicious area related to the EGFR mutation status, it has the potential ability to guide biopsies.

The purposes are to explore the safety performance of the Maritime Transportation System (MTS) based on Digital Twins (DTs) Internet of Things (IoT) and develop maritime transportation towards intelligence and digitalization. Because the comprehensive operational security of modern MTS is not yet mature, historical transportation data of the Maritime Silk Road are acquired and preprocessed. Afterward, DTs are introduced, and relay nodes are added to data transmission paths to construct a maritime transportation DTs model based on relay cooperation IoT. Eventually, this model's security performance is validated through simulation experiments. Relay security analysis suggests that interference information is a vital guarantee to assist in information non-disclosure, from which the constructed model can harvest energy to increase the data transmission power, thereby improving communication performance and secrecy rate. Outage probability analysis reveals that the simulated and the theoretical results are almost the same; moreover, given the system's multi-hop paths in the same environment, the more the relays and the greater the fading index, the better the system performance and the lower the outage probability. Once the iterations reach a particular number, the node secrecy rate becomes optimal and cannot cause excessive burden to the system; besides, the power distribution can establish a new equilibrium when the nodes are in different locations, so that system security performance gets improved. The simulated value is closest to the actual result under 100% successful transmission probability and 0.01~0.05 λ value. To sum up, the constructed maritime transportation DTs model presents extraordinary transmission and security performance, providing an experimental basis for intelligent and secure maritime transportation in the future.

Accurate and rapid diagnosis of coronavirus disease 2019 (COVID-19) from chest CT scans is of great importance and urgency during the worldwide outbreak. However, radiologists have to distinguish COVID-19 pneumonia from other pneumonia in a large number of CT scans, which is tedious and inefficient. Thus, it is urgently and clinically needed to develop an efficient and accurate diagnostic tool to help radiologists to fulfill the difficult task. In this study, we proposed a deep supervised autoencoder (DSAE) framework to automatically identify COVID-19 using multi-view features extracted from CT images. To fully explore features characterizing CT images from different frequency domains, DSAE was proposed to learn the latent representation by multi-task learning. The proposal was designed to both encode valuable information from different frequency features and construct a compact class structure for separability. To achieve this, we designed a multi-task loss function, which consists of a supervised loss and a reconstruction loss. Our proposed method was evaluated on a newly collected dataset of 787 subjects including COVID-19 pneumonia patients, other pneumonia patients, and normal subjects without abnormal CT findings. Extensive experimental results demonstrated that our proposed method achieved encouraging diagnostic performance and may have potential clinical application for the diagnosis of COVID-19.

Efficient deployment of Deep Neural Networks (DNNs) on edge devices (i.e., FPGAs and mobile platforms) is very challenging, especially under a recent witness of the increasing DNN model size and complexity. Model compression strategies, including weight quantization and pruning, are widely recognized as effective approaches to significantly reduce computation and memory intensities, and have been implemented in many DNNs on edge devices. However, most state-of-the-art works focus on ad hoc optimizations, and there lacks a thorough study to comprehensively reveal the potentials and constraints of different edge devices when considering different compression strategies. In this article, we qualitatively and quantitatively compare the energy efficiency of FPGA-based and mobile-based DNN executions using mobile GPU and provide a detailed analysis. Based on the observations obtained from the analysis, we propose a unified optimization framework using block-based pruning to reduce the weight storage and accelerate the inference speed on mobile devices and FPGAs, achieving high hardware performance and energy-efficiency gain while maintaining accuracy.

The Internet of Things (IoT) is the integration of all information and Internet technology in the information age, which can realize the collection and transmission of intelligent information. A large number of sensors are producing and collecting data involving various industries every day. The amount of stream data generated is huge, and a large number of abnormal data are also generated in the process. Due to the demands of business and life quality improvement, the application of IoT technology to real-time monitoring and correction of massive stream data, especially the correction of abnormal data, is a very valuable research direction, and also the key to ensure the credibility and fidelity of IoT data. This article proposes a recurrent neural network model based on long- and short-term memory network (LSTM) and LSTM+. LSTM+ model not only reduces the regression error compared with the traditional LSTM model, but also can detect abnormal data collected by IoT terminal nodes, and can correct the abnormal data in real time, so as to ensure that the network prediction can have good stability and robustness.

Effective fusion of multimodal magnetic resonance imaging (MRI) is of great significance to boost the accuracy of glioma grading thanks to the complementary information provided by different imaging modalities. However, how to extract the common and distinctive information from MRI to achieve complementarity is still an open problem in information fusion research. In this study, we propose a deep neural network model termed as multimodal disentangled variational autoencoder (MMD-VAE) for glioma grading based on radiomics features extracted from preoperative multimodal MRI images. Specifically, the radiomics features are quantized and extracted from the region of interest for each modality. Then, the latent representations of variational autoencoder for these features are disentangled into common and distinctive representations to obtain the shared and complementary data among modalities. Afterwards, cross-modality reconstruction loss and common-distinctive loss are designed to ensure the effectiveness of the disentangled representations. Finally, the disentangled common and distinctive representations are fused to predict the glioma grades, and SHapley Additive exPlanations (SHAP) is adopted to quantitatively interpret and analyze the contribution of the important features to grading. Experimental results on two benchmark datasets demonstrate that the proposed MMD-VAE model achieves encouraging predictive performance (AUC:0.9939) on a public dataset, and good generalization performance (AUC:0.9611) on a cross-institutional private dataset. These quantitative results and interpretations may help radiologists understand gliomas better and make better treatment decisions for improving clinical outcomes.

The present work aims to improve the communication security of Internet of Vehicles (IoV) nodes in intelligent transportation through studying the safety of IoV in smart transportation based on Blockchain (BC). An IoV DTs model is built by combining big data with Digital Twins (DTs). Then, regarding the current IoV communication security issues, a secure communication architecture for the IoV system is proposed based on the immutable and trackable BC data. Besides, Wasserstein Distance Based Generative Adversarial Network (WaGAN) model constructs the IoV node risk forecast model. Because the WaGAN model calculates the loss function through Wasserstein distance, the learning rate of the model accelerates remarkably. After ten iterations, the loss rate of the WaGAN model is close to zero. Massive in-vehicle devices in IoV are connected simultaneously to the base station, causing network channel congestion. Therefore, a Group Authentication and Privacy-preserving (GAP) scheme is put forward. As users increase during authentication, the GAP scheme performs better than other authentication access schemes. In summary, the Intelligent Transportation System driven by DTs can promote intelligent transportation management. Besides, introducing BC into IoV can improve access control’s accuracy and response efficiency. The research reported here has significant value for improving the security of the information sharing of the IoV.

Early and accurate severity assessment of Coronavirus disease 2019 (COVID-19) based on computed tomography (CT) images offers a great help to the estimation of intensive care unit event and the clinical decision of treatment planning. To augment the labeled data and improve the generalization ability of the classification model, it is necessary to aggregate data from multiple sites. This task faces several challenges including class imbalance between mild and severe infections, domain distribution discrepancy between sites, and presence of heterogeneous features. In this paper, we propose a novel domain adaptation (DA) method with two components to address these problems. The first component is a stochastic class-balanced boosting sampling strategy that overcomes the imbalanced learning problem and improves the classification performance on poorly-predicted classes. The second component is a representation learning that guarantees three properties: 1) domain-transferability by prototype triplet loss, 2) discriminant by conditional maximum mean discrepancy loss, and 3) completeness by multi-view reconstruction loss. Particularly, we propose a domain translator and align the heterogeneous data to the estimated class prototypes (i.e., class centers) in a hyper-sphere manifold. Experiments on cross-site severity assessment of COVID-19 from CT images show that the proposed method can effectively tackle the imbalanced learning problem and outperform recent DA approaches.

Efficient deployment of Deep Neural Networks (DNNs) on edge devices (i.e., FPGAs and mobile platforms) is very challenging, especially under a recent witness of the increasing DNN model size and complexity. Although various optimization approaches have been proven to be effective in many DNNs on edge devices, most state-of-the-art work focuses on ad-hoc optimizations, and there lacks a thorough study to comprehensively reveal the potentials and constraints of different edge devices when considering different optimizations. In this paper, we qualitatively and quantitatively compare the energyefficiency of FPGA-based and mobile-based DNN executions, and provide detailed analysis.

The coronavirus disease (COVID-19) is rapidly spreading all over the world, and has infected more than 1,436,000 people in more than 200 countries and territories as of April 9, 2020. Detecting COVID-19 at early stage is essential to deliver proper healthcare to the patients and also to protect the uninfected population. To this end, we develop a dual-sampling attention network to automatically diagnose COVID-19 from the community acquired pneumonia (CAP) in chest computed tomography (CT). In particular, we propose a novel online attention module with a 3D convolutional network (CNN) to focus on the infection regions in lungs when making decisions of diagnoses. Note that there exists imbalanced distribution of the sizes of the infection regions between COVID-19 and CAP, partially due to fast progress of COVID-19 after symptom onset. Therefore, we develop a dual-sampling strategy to mitigate the imbalanced learning. Our method is evaluated (to our best knowledge) upon the largest multi-center CT data for COVID-19 from 8 hospitals. In the training-validation stage, we collect 2186 CT scans from 1588 patients for a 5-fold cross-validation. In the testing stage, we employ another independent large-scale testing dataset including 2796 CT scans from 2057 patients. Results show that our algorithm can identify the COVID-19 images with the area under the receiver operating characteristic curve (AUC) value of 0.944, accuracy of 87.5%, sensitivity of 86.9%, specificity of 90.1%, and F1-score of 82.0%. With this performance, the proposed algorithm could potentially aid radiologists with COVID-19 diagnosis from CAP, especially in the early stage of the COVID-19 outbreak.