The expressiveness of search space is a key concern in neural architecture search (NAS). Previous block-level approaches mainly focus on searching networks that chain one operation after another. Incorporating multi-path search space with the one-shot doctrine remains untackled. In this paper, we investigate the supernet behavior under the multi-path setting, which we call MixPath. For a sampled training, simply switching multiple paths on and off incurs severe feature inconsistency which deteriorates the convergence. To remedy this effect, we employ what we term as \emph{shadow batch normalizations} (SBN) to follow various path patterns. Experiments performed on CIFAR-10 show that our approach is effective regardless of the number of allowable paths. Further experiments are conducted on ImageNet to have a fair comparison with the latest NAS methods. Our code will be available https://github.com/xiaomi-automl/MixPath.git .
Convolutional neural networks are widely adopted in Acoustic Scene Classification (ASC) tasks, but they generally carry a heavy computational burden. In this work, we propose a lightweight yet high-performing baseline network inspired by MobileNetV2, which replaces square convolutional kernels with unidirectional ones to extract features alternately in temporal and frequency dimensions. Furthermore, we explore a dynamic architecture space built on the basis of the proposed baseline with the recent Neural Architecture Search (NAS) paradigm, which first trains a supernet that incorporates all candidate networks and then applies a well-known evolutionary algorithm NSGA-II to discover more efficient networks with higher accuracy and lower computational cost. Experimental results demonstrate that our searched network is competent in ASC tasks, which achieves 90.3% F1-score on the DCASE2018 task 5 evaluation set, marking a new state-of-the-art performance while saving 25% of FLOPs compared to our baseline network.
Variational inference is a fundamental problem in Variational Auto-Encoder (VAE). By virtue of high-dimensional geometry, we propose a very simple algorithm completely different from existing ones to solve the inference problem in VAE. We analyze the unique characteristics of random variables on spheres in high dimensions and prove that the Wasserstein distance between two arbitrary datasets randomly drawn from a sphere are nearly identical when the dimension is sufficiently large. Based on our theory, a novel algorithm for distribution-robust sampling is devised. Moreover, we reform the latent space of VAE by constraining latent random variables on the sphere, thus freeing VAE from the approximate optimization pertaining to the variational posterior probability. The new algorithm is named as Spherical Auto-Encoder (SAE), which is in essence the vanilla autoencoder with the spherical constraint on the latent space. The associated inference is called the spherical inference, which is geometrically deterministic but is much more robust to various probabilistic priors than the variational inference in VAE for sampling. The experiments on sampling and inference validate our theoretical analysis and the superiority of SAE.
Generative adversarial networks (GANs) have shown promise in image generation and classification given limited supervision. Existing methods extend the unsupervised GAN framework to incorporate supervision heuristically. Specifically, a single discriminator plays two incompatible roles of identifying fake samples and predicting labels and it only estimates the data without considering the labels. The formulation intrinsically causes two problems: (1) the generator and the discriminator (i.e., the classifier) may not converge to the data distribution at the same time; and (2) the generator cannot control the semantics of the generated samples. In this paper, we present the triple generative adversarial network (Triple-GAN), which consists of three players---a generator, a classifier, and a discriminator. The generator and the classifier characterize the conditional distributions between images and labels, and the discriminator solely focuses on identifying fake image-label pairs. We design compatible objective functions to ensure that the distributions characterized by the generator and the classifier converge to the data distribution. We evaluate Triple-GAN in two challenging settings, namely, semi-supervised learning and the extreme low data regime. In both settings, Triple-GAN can achieve state-of-the-art classification results among deep generative models and generate meaningful samples in a specific class simultaneously.
The great success of deep learning poses urgent challenges for understanding its working mechanism and rationality. The depth, structure, and massive size of the data are recognized to be three key ingredients for deep learning. Most of the recent theoretical studies for deep learning focus on the necessity and advantages of depth and structures of neural networks. In this paper, we aim at rigorous verification of the importance of massive data in embodying the out-performance of deep learning. To approximate and learn spatially sparse and smooth functions, we establish a novel sampling theorem in learning theory to show the necessity of massive data. We then prove that implementing the classical empirical risk minimization on some deep nets facilitates in realization of the optimal learning rates derived in the sampling theorem. This perhaps explains why deep learning performs so well in the era of big data.
The quality control of fetal sonographic (FS) images is essential for the correct biometric measurements and fetal anomaly diagnosis. However, quality control requires professional sonographers to perform and is often labor-intensive. To solve this problem, we propose an automatic image quality assessment scheme based on multi-task learning to assist in FS image quality control. An essential criterion for FS image quality control is that all the essential anatomical structures in the section should appear full and remarkable with a clear boundary. Therefore, our scheme aims to identify those essential anatomical structures to judge whether an FS image is the standard image, which is achieved by three convolutional neural networks. The Feature Extraction Network aims to extract deep level features of FS images. Based on the extracted features, the Class Prediction Network determines whether the structure meets the standard and Region Proposal Network identifies its position. The scheme has been applied to three types of fetal sections, which are the head, abdominal, and heart. The experimental results show that our method can make a quality assessment of an FS image within less a second. Also, our method achieves competitive performance in both the detection and classification compared with state-of-the-art methods.
Differential Architecture Search (DARTS) is now a widely disseminated weight-sharing neural architecture search method. However, there are two fundamental weaknesses remain untackled. First, we observe that the well-known aggregation of skip connections during optimization is caused by an unfair advantage in an exclusive competition. Second, there is a non-negligible incongruence when discretizing continuous architectural weights to a one-hot representation. Because of these two reasons, DARTS delivers a biased solution that might not even be suboptimal. In this paper, we present a novel approach to curing both frailties. Specifically, as unfair advantages in a pure exclusive competition easily induce a monopoly, we relax the choice of operations to be collaborative, where we let each operation have an equal opportunity to develop its strength. We thus call our method Fair DARTS. Moreover, we propose a zero-one loss to directly reduce the discretization gap. Experiments are performed on two mainstream search spaces, in which we achieve new state-of-the-art networks on ImageNet. Our code is available on https://github.com/xiaomi-automl/fairdarts.
Bayesian neural networks (BNNs) introduce uncertainty estimation to deep networks by performing Bayesian inference on network weights. However, such models bring the challenges of inference, and further BNNs with weight uncertainty rarely achieve superior performance to standard models. In this paper, we investigate a new line of Bayesian deep learning by performing Bayesian reasoning on the structure of deep neural networks. Drawing inspiration from the neural architecture search, we define the network structure as gating weights on the redundant operations between computational nodes, and apply stochastic variational inference techniques to learn the structure distributions of networks. Empirically, the proposed method substantially surpasses the advanced deep neural networks across a range of classification and segmentation tasks. More importantly, our approach also preserves benefits of Bayesian principles, producing improved uncertainty estimation than the strong baselines including MC dropout and variational BNNs algorithms (e.g. noisy EK-FAC).
Tens of millions of women suffer from infertility worldwide each year. In vitro fertilization (IVF) is the best choice for many such patients. However, IVF is expensive, time-consuming, and both physically and emotionally demanding. The first question that a patient usually asks before the IVF is how likely she will conceive, given her basic medical examination information. This paper proposes three approaches to predict the cumulative pregnancy rate after multiple oocyte pickup cycles. Experiments on 11,190 patients showed that first clustering the patients into different groups and then building a support vector machine model for each group can achieve the best overall performance. Our model could be a quick and economic approach for reliably estimating the cumulative pregnancy rate for a patient, given only her basic medical examination information, well before starting the actual IVF procedure. The predictions can help the patient make optimal decisions on whether to use her own oocyte or donor oocyte, how many oocyte pickup cycles she may need, whether to use embryo frozen, etc. They will also reduce the patient's cost and time to pregnancy, and improve her quality of life.
Improving the performance of click-through rate (CTR) prediction remains one of the core tasks in online advertising systems. With the rise of deep learning, CTR prediction models with deep networks remarkably enhance model capacities. In deep CTR models, exploiting users' historical data is essential for learning users' behaviors and interests. As existing CTR prediction works neglect the importance of the temporal signals when embed users' historical clicking records, we propose a time-aware attention model which explicitly uses absolute temporal signals for expressing the users' periodic behaviors and relative temporal signals for expressing the temporal relation between items. Besides, we propose a regularized adversarial sampling strategy for negative sampling which eases the classification imbalance of CTR data and can make use of the strong guidance provided by the observed negative CTR samples. The adversarial sampling strategy significantly improves the training efficiency, and can be co-trained with the time-aware attention model seamlessly. Experiments are conducted on real-world CTR datasets from both in-station and out-station advertising places.