Bayesian optimization (BO) is an effective method of finding the global optima of black-box functions. Recently BO has been applied to neural architecture search and shows better performance than pure evolutionary strategies. All these methods adopt Gaussian processes (GPs) as surrogate function, with the handcraft similarity metrics as input. In this work, we propose a Bayesian graph neural network as a new surrogate, which can automatically extract features from deep neural architectures, and use such learned features to fit and characterize black-box objectives and their uncertainty. Based on the new surrogate, we then develop a graph Bayesian optimization framework to address the challenging task of deep neural architecture search. Experiment results show our method significantly outperforms the comparative methods on benchmark tasks.

Convolutional Neural Network (CNN) has demonstrated impressive ability to represent hyperspectral images and to achieve promising results in hyperspectral image classification. However, traditional CNN models can only operate convolution on regular square image regions with fixed size and weights, so they cannot universally adapt to the distinct local regions with various object distributions and geometric appearances. Therefore, their classification performances are still to be improved, especially in class boundaries. To alleviate this shortcoming, we consider employing the recently proposed Graph Convolutional Network (GCN) for hyperspectral image classification, as it can conduct the convolution on arbitrarily structured non-Euclidean data and is applicable to the irregular image regions represented by graph topological information. Different from the commonly used GCN models which work on a fixed graph, we enable the graph to be dynamically updated along with the graph convolution process, so that these two steps can be benefited from each other to gradually produce the discriminative embedded features as well as a refined graph. Moreover, to comprehensively deploy the multi-scale information inherited by hyperspectral images, we establish multiple input graphs with different neighborhood scales to extensively exploit the diversified spectral-spatial correlations at multiple scales. Therefore, our method is termed 'Multi-scale Dynamic Graph Convolutional Network' (MDGCN). The experimental results on three typical benchmark datasets firmly demonstrate the superiority of the proposed MDGCN to other state-of-the-art methods in both qualitative and quantitative aspects.

Reducing the test time resource requirements of a neural network while preserving test accuracy is crucial for running inference on resource-constrained devices. To achieve this goal, we introduce a novel network reparameterization based on the Kronecker-factored eigenbasis (KFE), and then apply Hessian-based structured pruning methods in this basis. As opposed to existing Hessian-based pruning algorithms which do pruning in parameter coordinates, our method works in the KFE where different weights are approximately independent, enabling accurate pruning and fast computation. We demonstrate empirically the effectiveness of the proposed method through extensive experiments. In particular, we highlight that the improvements are especially significant for more challenging datasets and networks. With negligible loss of accuracy, an iterative-pruning version gives a 10$\times$ reduction in model size and a 8$\times$ reduction in FLOPs on wide ResNet32.

In this work, we propose a novel technique to boost training efficiency of a neural network. Our work is based on an excellent idea that whitening the inputs of neural networks can achieve a fast convergence speed. Given the well-known fact that independent components must be whitened, we introduce a novel Independent-Component (IC) layer before each weight layer, whose inputs would be made more independent. However, determining independent components is a computationally intensive task. To overcome this challenge, we propose to implement an IC layer by combining two popular techniques, Batch Normalization and Dropout, in a new manner that we can rigorously prove that Dropout can quadratically reduce the mutual information and linearly reduce the correlation between any pair of neurons with respect to the dropout layer parameter $p$. As demonstrated experimentally, the IC layer consistently outperforms the baseline approaches with more stable training process, faster convergence speed and better convergence limit on CIFAR10/100 and ILSVRC2012 datasets. The implementation of our IC layer makes us rethink the common practices in the design of neural networks. For example, we should not place Batch Normalization before ReLU since the non-negative responses of ReLU will make the weight layer updated in a suboptimal way, and we can achieve better performance by combining Batch Normalization and Dropout together as an IC layer.

This paper presents a framework for estimating the remaining useful life (RUL) of mechanical systems. The framework consists of a multi-layer perceptron and an evolutionary algorithm for optimizing the data-related parameters. The framework makes use of a strided time window to estimate the RUL for mechanical components. Tuning the data-related parameters can become a very time consuming task. The framework presented here automatically reshapes the data such that the efficiency of the model is increased. Furthermore, the complexity of the model is kept low, e.g. neural networks with few hidden layers and few neurons at each layer. Having simple models has several advantages like short training times and the capacity of being in environments with limited computational resources such as embedded systems. The proposed method is evaluated on the publicly available C-MAPSS dataset, its accuracy is compared against other state-of-the art methods for the same dataset.

As the application of deep learning has expanded to real-world problems with insufficient volume of training data, transfer learning recently has gained much attention as means of improving the performance in such small-data regime. However, when existing methods are applied between heterogeneous architectures and tasks, it becomes more important to manage their detailed configurations and often requires exhaustive tuning on them for the desired performance. To address the issue, we propose a novel transfer learning approach based on meta-learning that can automatically learn what knowledge to transfer from the source network to where in the target network. Given source and target networks, we propose an efficient training scheme to learn meta-networks that decide (a) which pairs of layers between the source and target networks should be matched for knowledge transfer and (b) which features and how much knowledge from each feature should be transferred. We validate our meta-transfer approach against recent transfer learning methods on various datasets and network architectures, on which our automated scheme significantly outperforms the prior baselines that find "what and where to transfer" in a hand-crafted manner.

Semi-parametric survival analysis methods like the Cox Proportional Hazards (CPH) regression (Cox, 1972) are a popular approach for survival analysis. These methods involve fitting of the log-proportional hazard as a function of the covariates and are convenient as they do not require estimation of the baseline hazard rate. Recent approaches have involved learning non-linear representations of the input covariates and demonstrate improved performance. In this paper we argue against such deep parameterizations for survival analysis and experimentally demonstrate that more interpretable semi-parametric models inspired from mixtures of experts perform equally well or in some cases better than such overly parameterized deep models.

Deep neural networks have achieved remarkable success in challenging tasks. However, the black-box approach of training and testing of such networks is not acceptable to critical applications. In particular, the existence of adversarial examples and their overgeneralization to irrelevant inputs makes it difficult, if not impossible, to explain decisions by commonly used neural networks. In this paper, we analyze the underlying mechanism of generalization of deep neural networks and propose an ($n$, $k$) consensus algorithm to be insensitive to adversarial examples and at the same time be able to reject irrelevant samples. Furthermore, the consensus algorithm is able to improve classification accuracy by using multiple trained deep neural networks. To handle the complexity of deep neural networks, we cluster linear approximations and use cluster means to capture feature importance. Due to weight symmetry, a small number of clusters are sufficient to produce a robust interpretation. Experimental results on a health dataset show the effectiveness of our algorithm in enhancing the prediction accuracy and interpretability of deep neural network models on one-year patient mortality prediction.

Deep neural networks provide best-in-class performance for a number of computer vision problems. However, training these networks is computationally intensive and requires fine-tuning various hyperparameters. In addition, performance swings widely as the network converges making it hard to decide when to stop training. In this paper, we introduce a trio of techniques (PSWA, PWALKS, and PSWM) centered around periodic sampling of model weights that provide consistent and more robust convergence on a variety of vision problems (classification, detection, segmentation) and gradient update methods (vanilla SGD, Momentum, Adam) with marginal additional computation time. Our techniques use existing optimal training policies but converge in a less volatile fashion with performance improvements that are approximately monotonic. Our analysis of the loss surface shows that these techniques also produce minima that are deeper and wider than those found by SGD.

We propose a novel Bayesian nonparametric method to learn translation-invariant relationships on non-Euclidean domains. The resulting graph convolutional Gaussian processes can be applied to problems in machine learning for which the input observations are functions with domains on general graphs. The structure of these models allows for high dimensional inputs while retaining expressibility, as is the case with convolutional neural networks. We present applications of graph convolutional Gaussian processes to images and triangular meshes, demonstrating their versatility and effectiveness, comparing favorably to existing methods, despite being relatively simple models.