The increasing application of Artificial Intelligence (AI) in health and medicine has attracted a great deal research interest recent decades. This study aims to provide global historical picture concerning AI medicine. A total 27,451 papers that were published between 1977 2018 (84.6% dated 2008⁻2018) retrieved from the Web Science platform. descriptive analysis examined publication volume, authors countries collaboration. network authors' keywords content related scientific literature...

Mobile phones have evolved into powerful image and video processing devices equipped with high-resolution cameras, color displays, hardware-accelerated graphics. They are also increasingly a global positioning system connected to broadband wireless networks. All this enables new class of applications that use the camera phone initiate search queries about objects in visual proximity user (Figure 1). Such can be used, e.g., for identifying products, comparison shopping, finding information...

Recent successes in Generative Adversarial Networks (GAN) have affirmed the importance of using more data GAN training. Yet it is expensive to collect many domains such as medical applications. Data Augmentation (DA) has been applied these In this work, we first argue that classical DA approach could mislead generator learn distribution augmented data, which be different from original data. We then propose a principled framework, termed Optimized for (DAG), enable use training improve...

In recent years Deep Neural Networks (DNNs) have been rapidly developed in various applications, together with increasingly complex architectures. The performance gain of these DNNs generally comes high computational costs and large memory consumption, which may not be affordable for mobile platforms. model quantization can used reducing the computation DNNs, deploying on equipment. this work, we propose an optimization framework deep quantization. First, a measurement to estimate effect...

We survey popular data sets used in computer vision literature and point out their limitations for mobile visual search applications. To overcome many of the limitations, we propose Stanford Mobile Visual Search set. The set contains camera-phone images products, CDs, books, outdoor landmarks, business cards, text documents, museum paintings video clips. has several key characteristics lacking existing sets: rigid objects, widely varying lighting conditions, perspective distortion,...

We propose DGG: Deep clustering via a Gaussian-mixture variational autoencoder (VAE) with Graph embedding. To facilitate clustering, we apply Gaussian mixture model (GMM) as the prior in VAE. handle data complex spread, graph Our idea is that information which captures local structures an excellent complement to deep GMM. Combining them facilitates network learn powerful representations follow global and structural constraints. Therefore, our method unifies model-based similarity-based...

Nowadays, the occurrence of skin cancer cases has grown worldwide due to extended exposure harmful radiation from Sun. Most common approach detect malignancy moles is by visual inspection performed an expert dermatologist, using a set specific clinical rules. Computer-aided diagnosis, based on mole imaging, another concurrent method which experienced major advancements improvement imaging sensors and processing power. However, these schemes use hand-crafted features are difficult tune...

Recently, the introduction of generative adversarial network (GAN) and its variants has enabled generation realistic synthetic samples, which been used for enlarging training sets. Previous work primarily focused on data augmentation semi-supervised supervised tasks. In this paper, we instead focus unsupervised anomaly detection propose a novel framework optimized task. By using GAN variant known as autoencoder (AAE), impose distribution latent space dataset systematically sample to generate...

Image blur and image noise are common distortions during acquisition. In this paper, we systematically study the effect of on deep neural network (DNN) classifiers. First, examine DNN classifier performance under four types distortions. Second, propose two approaches to alleviate distortion: re-training fine-tuning with noisy images. Our results suggest that, certain conditions, images can much due distorted inputs, is more practical than re-training.

High-frame-rate (HFR) video is emerging in popular gaming applications to enhance the smooth experience perceived by end users. However, it challenging guarantee delivery quality of HFR mobile cloud scenarios because high transmission rate and limited wireless resources. To address this critical problem, we develop a novel scheduling framework dubbed AdaPtive vIdeo Streaming (APHIS). The term adaptive indicates scheme's capability dynamically adjusting traffic load forward error correction...

We address the vehicle detection and classification problems using Deep Neural Networks (DNNs) approaches. Here we answer to questions that are specific our application including how utilize DNN for detection, what features useful classification, extend a model trained on limited size dataset, cases of extreme lighting condition. Answering these propose approach outperforms state-of-the-art methods, achieves promising results image with conditions.

Few shot image classification aims at learning a classifier from limited labeled data. Generating the weights has been applied in many meta-learning methods for few due to its simplicity and effectiveness. In this work, we present Attentive Weights Generation via Information Maximization (AWGIM), which introduces two novel contributions: i) Mutual information maximization between generated data within task; enables retain of task specific query sample. ii) Self-attention cross-attention...

This paper presents a novel randomized algorithm for robust point cloud registration without correspondences. Most existing approaches require set of putative correspondences obtained by extracting invariant descriptors. However, such descriptors could become unreliable in noisy and contaminated settings. In these settings, methods that directly handle input sets are preferable. Without correspondences, however, conventional techniques very large number samples order to reach satisfactory...

Person Re-Identification (ReID) requires comparing two images of person captured under different conditions. Existing work based on neural networks often computes the similarity feature maps from one single convolutional layer. In this work, we propose an efficient, end-to-end fully Siamese network that similarities at multiple levels. We demonstrate multi-level can improve accuracy considerably using low-complexity structures in ReID problem. Specifically, first, use several layers to...

Modern GANs excel at generating high quality and diverse images. However, when transferring the pretrained on small target data (e.g., 10-shot), generator tends to replicate training samples. Several methods have been proposed address this few-shot image generation task, but there is a lack of effort analyze them under unified framework. As our first contribution, we propose framework existing during adaptation. Our analysis discovers that while some disproportionate focus diversity...

While much of multiview video coding focuses on the rate-distortion performance compressing all frames views for storage or non-interactive delivery over networks, we address problem designing a frame structure to enable interactive streaming, where clients can interactively switch during playback. Thus, as client is playing back successive (in time) given view, it send request server different view while continuing uninterrupted temporal Noting that standard tools random access (i.e.,...

In this paper, we propose a multimodal multi-stream deep learning framework to tackle the egocentric activity recognition problem, using both video and sensor data. First, experiment extend Convolutional Neural Network learn spatial temporal features from videos. Second, multistream Long Short-Term Memory architecture multiple streams (accelerometer, gyroscope, etc.). Third, use two-level fusion technique different pooling techniques compute prediction results. Experimental results dataset...

Delivering high-quality mobile video with the limited radio resources is challenging due to time-varying channel status and stringent Quality of Service (QoS) requirements. Multi-homing support enables terminals establish multiple simultaneous associations for enhancing transmission performance. In this paper, we study multi-homed communication delay-constrained High Definition (HD) in heterogeneous wireless networks. The low-delay encoded HD streaming consists exclusively Intra (I)...

We investigate the design of an entire mobile imaging system for early detection melanoma. Different from previous work, we focus on smartphone-captured visible light images. Our addresses two major challenges. First, images acquired using a smartphone under loosely-controlled environmental conditions may be subject to various distortions, and this makes melanoma more difficult. Second, processing performed is stringent computation memory constraints. In our propose that optimized run...

While Geerative Adversarial Networks (GANs) are fundamental to many generative modelling applications, they suffer from numerous issues. In this work, we propose a principled framework simultaneously mitigate two issues in GANs: catastrophic forgetting of the discriminator and mode collapse generator. We achieve by employing for GANs contrastive learning mutual information maximization approach, perform extensive analyses understand sources improvements. Our approach significantly stabilizes...

CNN-based generative modelling has evolved to produce synthetic images indistinguishable from real in the RGB pixel space. Recent works have observed that CNN-generated share a systematic shortcoming replicating high frequency Fourier spectrum decay attributes. Furthermore, these successfully exploited this detect reporting up 99% accuracy across multiple state-of-the-art GAN models.In work, we investigate validity of assertions claiming are unable achieve spectral consistency. We...

Cross-domain few-shot classification task (CD-FSC) combines with the requirement to generalize across domains represented by datasets. This setup faces challenges originating from limited labeled data in each class and, additionally, domain shift between training and test sets. In this paper, we introduce a novel approach for existing FSC models. It leverages on explanation scores, obtained methods when applied predictions of models, computed intermediate feature maps Firstly, tailor...

Model inversion (MI) attacks aim to infer and reconstruct private training data by abusing access a model. MI have raised concerns about the leaking of sen-sitive information (e.g. face images used in recognition system). Recently, several algorithms for been proposed improve attack performance. In this work, we revisit MI, study two fundamental issues pertaining all state-of-the-art (SOTA) algorithms, propose solutions these which lead significant boost performance SOTA MI. particular, our...