Nowadays, Internet of Things (IoT) gives rise to a huge amount of data. IoT nodes equipped with smart sensors can immediately extract meaningful knowledge from the data through machine learning technologies. Deep learning (DL) is constantly contributing significant progress in smart sensing due to its dramatic superiorities over traditional machine learning. The promising prospect of wide-range applications puts forwards demands on the ubiquitous deployment of DL under various contexts. As a result, performing DL on mobile or embedded platforms is becoming a common requirement. Nevertheless, a typical DL application can easily exhaust an embedded or mobile device owing to a large amount of multiply and accumulate (MAC) operations and memory access operations. Consequently, it is a challenging task to bridge the gap between deep learning and resource-limited platforms. We summarize typical applications of resource-limited deep learning and point out that deep learning is an indispensable impetus of pervasive computing. Subsequently, we explore the underlying reasons for the high computational overhead of DL through reviewing the fundamental concepts including capacity, generalization, and backpropagation of a neural network. Guided by these concepts, we investigate on principles of representative research works, as well as three types of solutions: algorithmic design, computational optimization, and hardware revolution. In pursuant to these solutions, we identify challenges to be addressed.

Since microaneurysms (MAs) can be seen as the earliest lesions in diabetic retinopathy, its detection plays a critical role in the diabetic retinopathy diagnosis. In recent years, many machine-learning methods have been developed for MA detection. Generally, MA candidates are first identified and then a set of features for these candidates are extracted. Finally, machine-learning methods are applied for candidate classification. In this paper, we present a novel unsupervised classification method based on sparse posterior cerebral artery (PCA) for MA detection. Since it does not have to consider a non-MA training set, the class imbalance problem can be avoided. Furthermore, effective features can be selected due to the characteristic of sparse PCA, which combines the elastic net penalty with the PCA. Meanwhile, a single T ² statistic is introduced, and the control limit can be determined for distinguishing true MAs from spurious candidates automatically. Experiment results on the retinopathy online challenge competition database show the effectiveness of our proposed method.

Exudates can be regarded as one of the most prevalent clinical signs of diabetic retinopathy, and the detection of exudates has important clinical significance in diabetic retinopathy diagnosis. In this paper, a novel approach named superpixel multi-feature classification for the automatic detection of exudates is developed. First, an entire image is segmented into a series of superpixels considered as candidates. Then, a total of 20 features, including 19 multi-channel intensity features and a novel contextual feature, are proposed for characterizing each candidate. A supervised multi-variable classification algorithm is also introduced to distinguish the true exudates from the spurious candidates. Finally, a novel optic disc detection technique is designed to further improve the performance of classification accuracy. Extensive experiments are carried out on two publicly available online databases, DiaretDB1, and e-ophtha EX. Compared with other state-of-the-art approaches, the experimental results show the advantages and effectiveness of the proposed approach.