Deep learning has arguably achieved tremendous success in recent years. In simple words, deep learning uses the composition of many nonlinear functions to model the complex dependency between input features and labels. While neural networks have a long history, recent advances have greatly improved their performance in computer vision, natural language processing, etc. From the statistical and scientific perspective, it is natural to ask: What is deep learning? What are the new characteristics of deep learning, compared with classical methods? What are the theoretical foundations of deep learning? To answer these questions, we introduce common neural network models (e.g., convolutional neural nets, recurrent neural nets, generative adversarial nets) and training techniques (e.g., stochastic gradient descent, dropout, batch normalization) from a statistical point of view. Along the way, we highlight new characteristics of deep learning (including depth and over-parametrization) and explain their practical and theoretical benefits. We also sample recent results on theories of deep learning, many of which are only suggestive. While a complete understanding of deep learning remains elusive, we hope that our perspectives and discussions serve as a stimulus for new statistical research.

We propose a dynamic neighborhood aggregation (DNA) procedure guided by (multi-head) attention for representation learning on graphs. In contrast to current graph neural networks which follow a simple neighborhood aggregation scheme, our DNA procedure allows for a selective and node-adaptive aggregation of neighboring embeddings of potentially differing locality. In order to avoid overfitting, we propose to control the channel-wise connections between input and output by making use of grouped linear projections. In a number of transductive nodeclassification experiments, we demonstrate the effectiveness of our approach. Graph neural networks (GNNs) have become the de facto standard for representation learning on relational data (Bronstein et al., 2017; Gilmer et al., 2017; Battaglia et al., 2018).

Irony and sarcasm are two complex linguistic phenomena that are widely used in everyday language and especially over the social media, but they represent two serious issues for automated text understanding. Many labelled corpora have been extracted from several sources to accomplish this task, and it seems that sarcasm is conveyed in different ways for different domains. Nonetheless, very little work has been done for comparing different methods among the available corpora. Furthermore, usually, each author collects and uses its own dataset to evaluate his own method. In this paper, we show that sarcasm detection can be tackled by applying classical machine learning algorithms to input texts sub-symbolically represented in a Latent Semantic space. The main consequence is that our studies establish both reference datasets and baselines for the sarcasm detection problem that could serve to the scientific community to test newly proposed methods.

We propose a new image denoising algorithm, dubbed as Fully Convolutional Adaptive Image DEnoiser (FC-AIDE), that can learn from offline supervised training set with a fully convolutional neural network as well as adaptively fine-tune the supervised model for each given noisy image. We significantly extend the framework of the recently proposed Neural AIDE, which formulates the denoiser to be context-based pixelwise mappings and utilizes the unbiased estimator of MSE for such denoisers. The two main contributions we make are; 1) implementing a novel fully convolutional architecture that boosts the base supervised model, and 2) introducing regularization methods for the adaptive fine-tuning such that a stronger and more robust adaptivity can be attained. As a result, FC-AIDE is shown to possess many desirable features; it outperforms the recent CNN-based state-of-the-art denoisers on all of the benchmark datasets we tested, and gets particularly strong for various challenging scenarios, e.g., with mismatched image/noise characteristics or with scarce supervised training data.

Cyber-physical systems, especially in critical infrastructures, have become primary hacking targets in international conflicts and diplomacy. However, cyber-physical systems present unique challenges to defenders, starting with an inability to communicate. This paper outlines the results of our interviews with information technology (IT) defenders and operational technology (OT) operators and seeks to address lessons learned from them in the structure of our notional solutions. We present two problems in this paper: (1) the difficulty of coordinating detection and response between defenders who work on the cyber/IT and physical/OT sides of cyber-physical infrastructures, and (2) the difficulty of estimating the safety state of a cyber-physical system while an intrusion is underway but before damage can be effected by the attacker. To meet these challenges, we propose two solutions: (1) a visualization that will enable communication between IT defenders and OT operators, and (2) a machine-learning approach that will estimate the distance from normal the physical system is operating and send information to the visualization.

Image synthesis is a core problem in modern deep learning, and many recent architectures such as autoencoders and Generative Adversarial networks produce spectacular results on highly complex data, such as images of faces or landscapes. While these results open up a wide range of new, advanced synthesis applications, there is also a severe lack of theoretical understanding of how these networks work. This results in a wide range of practical problems, such as difficulties in training, the tendency to sample images with little or no variability, and generalisation problems. In this paper, we propose to analyse the ability of the simplest generative network, the autoencoder, to encode and decode two simple geometric attributes : size and position. We believe that, in order to understand more complicated tasks, it is necessary to first understand how these networks process simple attributes. For the first property, we analyse the case of images of centred disks with variable radii. We explain how the autoencoder projects these images to and from a latent space of smallest possible dimension, a scalar. In particular, we describe a closed-form solution to the decoding training problem in a network without biases, and show that during training, the network indeed finds this solution. We then investigate the best regularisation approaches which yield networks that generalise well. For the second property, position, we look at the encoding and decoding of Dirac delta functions, also known as `one-hot' vectors. We describe a hand-crafted filter that achieves encoding perfectly, and show that the network naturally finds this filter during training. We also show experimentally that the decoding can be achieved if the dataset is sampled in an appropriate manner.

Endoscopy is a routine imaging technique used for both diagnosis and minimally invasive surgical treatment. Artifacts such as motion blur, bubbles, specular reflections, floating objects and pixel saturation impede the visual interpretation and the automated analysis of endoscopy videos. Given the widespread use of endoscopy in different clinical applications, we contend that the robust and reliable identification of such artifacts and the automated restoration of corrupted video frames is a fundamental medical imaging problem. Existing state-of-the-art methods only deal with the detection and restoration of selected artifacts. However, typically endoscopy videos contain numerous artifacts which motivates to establish a comprehensive solution. We propose a fully automatic framework that can: 1) detect and classify six different primary artifacts, 2) provide a quality score for each frame and 3) restore mildly corrupted frames. To detect different artifacts our framework exploits fast multi-scale, single stage convolutional neural network detector. We introduce a quality metric to assess frame quality and predict image restoration success. Generative adversarial networks with carefully chosen regularization are finally used to restore corrupted frames. Our detector yields the highest mean average precision (mAP at 5% threshold) of 49.0 and the lowest computational time of 88 ms allowing for accurate real-time processing. Our restoration models for blind deblurring, saturation correction and inpainting demonstrate significant improvements over previous methods. On a set of 10 test videos we show that our approach preserves an average of 68.7% which is 25% more frames than that retained from the raw videos.

Recently, deep models have been successfully applied in several applications, especially with low-level representations. However, sparse, noisy samples and structured domains (with multiple objects and interactions) are some of the open challenges in most deep models. Column Networks, a deep architecture, can succinctly capture such domain structure and interactions, but may still be prone to sub-optimal learning from sparse and noisy samples. Inspired by the success of human-advice guided learning in AI, especially in data-scarce domains, we propose Knowledge-augmented Column Networks that leverage human advice/knowledge for better learning with noisy/sparse samples. Our experiments demonstrate that our approach leads to either superior overall performance or faster convergence (i.e., both effective and efficient).

Most of the classical denoising methods restore clear results by selecting and averaging pixels in the noisy input. Instead of relying on hand-crafted selecting and averaging strategies, we propose to explicitly learn this process with deep neural networks. Specifically, we propose deformable 2D kernels for image denoising where the sampling locations and kernel weights are both learned. The proposed kernel naturally adapts to image structures and could effectively reduce the oversmoothing artifacts. Furthermore, we develop 3D deformable kernels for video denoising to more efficiently sample pixels across the spatial-temporal space. Our method is able to solve the misalignment issues of large motion from dynamic scenes. For better training our video denoising model, we introduce the trilinear sampler and a new regularization term. We demonstrate that the proposed method performs favorably against the state-of-the-art image and video denoising approaches on both synthetic and real-world data.

Compared with global average pooling in existing deep convolutional neural networks (CNNs), global covariance pooling can capture richer statistics of deep features, having potential for improving representation and generalization abilities of deep CNNs. However, integration of global covariance pooling into deep CNNs brings two challenges: (1) robust covariance estimation given deep features of high dimension and small sample; (2) appropriate use of geometry of covariances. To address these challenges, we propose a global Matrix Power Normalized COVariance (MPN-COV) Pooling. Our MPN-COV conforms to a robust covariance estimator, very suitable for scenario of high dimension and small sample. It can also be regarded as power-Euclidean metric between covariances, effectively exploiting their geometry. Furthermore, a global Gaussian embedding method is proposed to incorporate first-order statistics into MPN-COV. For fast training of MPN-COV networks, we propose an iterative matrix square root normalization, avoiding GPU unfriendly eigen-decomposition inherent in MPN-COV. Additionally, progressive 1x1 and group convolutions are introduced to compact covariance representations. The MPN-COV and its variants are highly modular, readily plugged into existing deep CNNs. Extensive experiments are conducted on large-scale object classification, scene categorization, fine-grained visual recognition and texture classification, showing our methods are superior to the counterparts and achieve state-of-the-art performance.