Neural networks play an increasingly important role in the field of machine learning and are included in many applications in society. Unfortunately, neural networks suffer from adversarial samples generated to attack them. However, most of the generation approaches either assume that the attacker has full knowledge of the neural network model or are limited by the type of attacked model. In this paper, we propose a new approach that generates a black-box attack to neural networks based on the swarm evolutionary algorithm. Benefiting from the improvements in the technology and theoretical characteristics of evolutionary algorithms, our approach has the advantages of effectiveness, black-box attack, generality, and randomness. Our experimental results show that both the MNIST images and the CIFAR-10 images can be perturbed to successful generate a black-box attack with 100\% probability on average. In addition, the proposed attack, which is successful on distilled neural networks with almost 100\% probability, is resistant to defensive distillation. The experimental results also indicate that the robustness of the artificial intelligence algorithm is related to the complexity of the model and the data set. In addition, we find that the adversarial samples to some extent reproduce the characteristics of the sample data learned by the neural network model.

This paper aims to provide a better understanding of a symmetric loss. First, we show that using a symmetric loss is advantageous in the balanced error rate (BER) minimization and area under the receiver operating characteristic curve (AUC) maximization from corrupted labels. Second, we prove general theoretical properties of symmetric losses, including a classification-calibration condition, excess risk bound, conditional risk minimizer, and AUC-consistency condition. Third, since all nonnegative symmetric losses are non-convex, we propose a convex barrier hinge loss that benefits significantly from the symmetric condition, although it is not symmetric everywhere. Finally, we conduct experiments on BER and AUC optimization from corrupted labels to validate the relevance of the symmetric condition.

Model pruning is a popular mechanism to make a network more efficient for inference. In this paper, we explore the use of pruning to also make the training of such neural networks more efficient. Unlike all prior model pruning methods that sparsify a pre-trained model and then prune it, we train the network from scratch, while gradually and structurally pruning parameters during the training. We build on our key observations: 1) once parameters are sparsified via regularization, they rarely re-appear in later steps, and 2) setting the appropriate regularization penalty at the beginning of training effectively converges the loss. We train ResNet and VGG networks on CIFAR10/100 and ImageNet datasets from scratch, and achieve 30-50% improvement in training FLOPs and 20-30% improvement in measured training time on modern GPUs.

Autonomous driving is getting a lot of attention in the last decade and will be the hot topic at least until the first successful certification of a car with Level 5 autonomy. There are many public datasets in the academic community. However, they are far away from what a robust industrial production system needs. There is a large gap between academic and industrial setting and a substantial way from a research prototype, built on public datasets, to a deployable solution which is a challenging task. In this paper, we focus on bad practices that often happen in the autonomous driving from an industrial deployment perspective. Data design deserves at least the same amount of attention as the model design. There is very little attention paid to these issues in the scientific community, and we hope this paper encourages better formalization of dataset design. More specifically, we focus on the datasets design and validation scheme for autonomous driving, where we would like to highlight the common problems, wrong assumptions, and steps towards avoiding them, as well as some open problems.

Toxicity prediction of chemical compounds is a grand challenge. Lately, it achieved significant progress in accuracy but using a huge set of features, implementing a complex blackbox technique such as a deep neural network, and exploiting enormous computational resources. In this paper, we strongly argue for the models and methods that are simple in machine learning characteristics, efficient in computing resource usage, and powerful to achieve very high accuracy levels. To demonstrate this, we develop a single task-based chemical toxicity prediction framework using only 2D features that are less compute intensive. We effectively use a decision tree to obtain an optimum number of features from a collection of thousands of them. We use a shallow neural network and jointly optimize it with decision tree taking both network parameters and input features into account. Our model needs only a minute on a single CPU for its training while existing methods using deep neural networks need about 10 min on NVidia Tesla K40 GPU. However, we obtain similar or better performance on several toxicity benchmark tasks. We also develop a cumulative feature ranking method which enables us to identify features that can help chemists perform prescreening of toxic compounds effectively.

Transfer learning through fine-tuning a pre-trained neural network with an extremely large dataset, such as ImageNet, can significantly accelerate training while the accuracy is frequently bottlenecked by the limited dataset size of the new target task. To solve the problem, some regularization methods, constraining the outer layer weights of the target network using the starting point as references (SPAR), have been studied. In this paper, we propose a novel regularized transfer learning framework DELTA, namely DEep Learning Transfer using Feature Map with Attention. Instead of constraining the weights of neural network, DELTA aims to preserve the outer layer outputs of the target network. Specifically, in addition to minimizing the empirical loss, DELTA intends to align the outer layer outputs of two networks, through constraining a subset of feature maps that are precisely selected by attention that has been learned in an supervised learning manner. We evaluate DELTA with the state-of-the-art algorithms, including L2 and L2-SP. The experiment results show that our proposed method outperforms these baselines with higher accuracy for new tasks.

We consider the problem of selective prediction (also known as reject option) in deep neural networks, and introduce SelectiveNet, a deep neural architecture with an integrated reject option. Existing rejection mechanisms are based mostly on a threshold over the prediction confidence of a pre-trained network. In contrast, SelectiveNet is trained to optimize both classification (or regression) and rejection simultaneously, end-to-end. The result is a deep neural network that is optimized over the covered domain. In our experiments, we show a consistently improved risk-coverage trade-off over several well-known classification and regression datasets, thus reaching new state-of-the-art results for deep selective classification.

Microarray gene expression has widely attracted the eyes of the public as an efficient tool for cancer diagnosis and classification. However, the very-high dimensionality and the small number of samples make it difficult for traditional machine learning algorithms to address this problem due to the high amount of computations required and overfitting. So far, the existing approaches of processing microarray datasets are still far from satisfactory and they employ two phases, feature selection (or extraction) followed by a machine learning algorithm. In this paper, we show that MultiLayer Perceptrons (MLPs) with adaptive sparse connectivity can directly handle this problem without features selection. Tested on four datasets, our novel results demonstrate that deep learning methods can be applied directly also to high dimensional non-grid like data, while learning from a small amount of labeled examples with imbalanced classes and achieving better accuracy than the traditional two phases approach. Moreover, we have been able to create sparse MLP models with over one million neurons and to train them on a typical laptop without GPU. This is with two orders of magnitude more than the largest MLPs which can run currently on commodity hardware.

We prove that for an $L$-layer fully-connected linear neural network, if the width of every hidden layer is $\tilde\Omega (L \cdot r \cdot d_{\mathrm{out}} \cdot \kappa^3 )$, where $r$ and $\kappa$ are the rank and the condition number of the input data, and $d_{\mathrm{out}}$ is the output dimension, then gradient descent with Gaussian random initialization converges to a global minimum at a linear rate. The number of iterations to find an $\epsilon$-suboptimal solution is $O(\kappa \log(\frac{1}{\epsilon}))$. Our polynomial upper bound on the total running time for wide deep linear networks and the $\exp\left(\Omega\left(L\right)\right)$ lower bound for narrow deep linear neural networks [Shamir, 2018] together demonstrate that wide layers are necessary for optimizing deep models.

We study the training process of Deep Neural Networks (DNNs) from the Fourier analysis perspective. Our starting point is a Frequency Principle (F-Principle) --- DNNs initialized with small parameters often fit target functions from low to high frequencies --- which was first proposed by Xu et al. (2018) and Rahaman et al. (2018) on synthetic datasets. In this work, we first show the universality of the F-Principle by demonstrating this phenomenon on high-dimensional benchmark datasets, such as MNIST and CIFAR10. Then, based on experiments, we show that the F-Principle provides insight into both the success and failure of DNNs in different types of problems. Based on the F-Principle, we further propose that DNN can be adopted to accelerate the convergence of low frequencies for scientific computing problems, in which most of the conventional methods (e.g., Jacobi method) exhibit the opposite convergence behavior --- faster convergence for higher frequencies. Finally, we prove a theorem for DNNs of one hidden layer as a first step towards a mathematical explanation of the F-Principle. Our work indicates that the F-Principle with Fourier analysis is a promising approach to the study of DNNs because it seems ubiquitous, applicable, and explainable.