Collaborating Authors Machine Learning

Domain Adaptive Transfer Attack (DATA)-based Segmentation Networks for Building Extraction from Aerial Images Machine Learning

Semantic segmentation models based on convolutional neural networks (CNNs) have gained much attention in relation to remote sensing and have achieved remarkable performance for the extraction of buildings from high-resolution aerial images. However, the issue of limited generalization for unseen images remains. When there is a domain gap between the training and test datasets, CNN-based segmentation models trained by a training dataset fail to segment buildings for the test dataset. In this paper, we propose segmentation networks based on a domain adaptive transfer attack (DATA) scheme for building extraction from aerial images. The proposed system combines the domain transfer and adversarial attack concepts. Based on the DATA scheme, the distribution of the input images can be shifted to that of the target images while turning images into adversarial examples against a target network. Defending adversarial examples adapted to the target domain can overcome the performance degradation due to the domain gap and increase the robustness of the segmentation model. Cross-dataset experiments and the ablation study are conducted for the three different datasets: the Inria aerial image labeling dataset, the Massachusetts building dataset, and the WHU East Asia dataset. Compared to the performance of the segmentation network without the DATA scheme, the proposed method shows improvements in the overall IoU. Moreover, it is verified that the proposed method outperforms even when compared to feature adaptation (FA) and output space adaptation (OSA).

On The Problem of Relevance in Statistical Inference Machine Learning

How many statistical inference tools we have for inference from massive data? A huge number, but only when we are ready to assume the given database is homogenous, consisting of a large cohort of "similar" cases. Why we need the homogeneity assumption? To make `learning from the experience of others' or `borrowing strength' possible. But, what if, we are dealing with a massive database of heterogeneous cases (which is a norm in almost all modern data-science applications including neuroscience, genomics, healthcare, and astronomy)? How many methods we have in this situation? Not much, if not ZERO. Why? It's not obvious how to go about gathering strength when each piece of information is fuzzy. The danger is that, if we include irrelevant cases, borrowing information might heavily damage the quality of the inference! This raises some fundamental questions for big data inference: When (not) to borrow? Whom (not) to borrow? How (not) to borrow? These questions are at the heart of the "Problem of Relevance" in statistical inference -- a puzzle that has remained too little addressed since its inception nearly half a century ago. Here we offer the first practical theory of relevance with precisely describable statistical formulation and algorithm. Through examples, we demonstrate how our new statistical perspective answers previously unanswerable questions in a realistic and feasible way.

CodNN -- Robust Neural Networks From Coded Classification Machine Learning

Deep Neural Networks (DNNs) are a revolutionary force in the ongoing information revolution, and yet their intrinsic properties remain a mystery. In particular, it is widely known that DNNs are highly sensitive to noise, whether adversarial or random. This poses a fundamental challenge for hardware implementations of DNNs, and for their deployment in critical applications such as autonomous driving. In this paper we construct robust DNNs via error correcting codes. By our approach, either the data or internal layers of the DNN are coded with error correcting codes, and successful computation under noise is guaranteed. Since DNNs can be seen as a layered concatenation of classification tasks, our research begins with the core task of classifying noisy coded inputs, and progresses towards robust DNNs. We focus on binary data and linear codes. Our main result is that the prevalent parity code can guarantee robustness for a large family of DNNs, which includes the recently popularized binarized neural networks. Further, we show that the coded classification problem has a deep connection to Fourier analysis of Boolean functions. In contrast to existing solutions in the literature, our results do not rely on altering the training process of the DNN, and provide mathematically rigorous guarantees rather than experimental evidence.

Theoretical Aspects of Group Equivariant Neural Networks Machine Learning

Group equivariant neural networks have been explored in the past few years and are interesting from theoretical and practical standpoints. They leverage concepts from group representation theory, non-commutative harmonic analysis and differential geometry that do not often appear in machine learning. In practice, they have been shown to reduce sample and model complexity, notably in challenging tasks where input transformations such as arbitrary rotations are present. We begin this work with an exposition of group representation theory and the machinery necessary to define and evaluate integrals and convolutions on groups. Then, we show applications to recent SO(3) and SE(3) equivariant networks, namely the Spherical CNNs, Clebsch-Gordan Networks, and 3D Steerable CNNs. We proceed to discuss two recent theoretical results. The first, by Kondor and Trivedi (ICML'18), shows that a neural network is group equivariant if and only if it has a convolutional structure. The second, by Cohen et al. (NeurIPS'19), generalizes the first to a larger class of networks, with feature maps as fields on homogeneous spaces.

On the Neural Tangent Kernel of Deep Networks with Orthogonal Initialization Machine Learning

In recent years, a critical initialization scheme of orthogonal initialization on deep nonlinear networks has been proposed. The orthogonal weights are crucial to achieve {\it dynamical isometry} for random networks, where the entire spectrum of singular values of an input-output Jacobian are around one. The strong empirical evidence that orthogonal initialization in linear networks and the linear regime of nonlinear networks can speed up training than Gaussian initialization raise great interests. One recent work has proven the benefit of orthogonal initialization in linear networks. However, the dynamics behind it have not been revealed on nonlinear networks. In this work, we study the Neural Tangent Kernel (NTK), which can describe dynamics of gradient descent training of wide network, and focus on fully-connected and nonlinear networks with orthogonal initialization. We prove that NTK of Gaussian and orthogonal weights are equal when the network width is infinite, resulting in a conclusion that orthogonal initialization can speed up training is a finite-width effect in the small learning rate regime. Then we find that during training, the NTK of infinite-width network with orthogonal initialization stays constant theoretically and varies at a rate of the same order as Gaussian ones empirically, as the width tends to infinity. Finally, we conduct a thorough empirical investigation of training speed on CIFAR10 datasets and show the benefit of orthogonal initialization lies in the large learning rate and depth phase in a linear regime of nonlinear network.

Scan-based Semantic Segmentation of LiDAR Point Clouds: An Experimental Study Machine Learning

Autonomous vehicles need to have a semantic understanding of the three-dimensional world around them in order to reason about their environment. State of the art methods use deep neural networks to predict semantic classes for each point in a LiDAR scan. A powerful and efficient way to process LiDAR measurements is to use two-dimensional, image-like projections. In this work, we perform a comprehensive experimental study of image-based semantic segmentation architectures for LiDAR point clouds. We demonstrate various techniques to boost the performance and to improve runtime as well as memory constraints. First, we examine the effect of network size and suggest that much faster inference times can be achieved at a very low cost to accuracy. Next, we introduce an improved point cloud projection technique that does not suffer from systematic occlusions. We use a cyclic padding mechanism that provides context at the horizontal field-of-view boundaries. In a third part, we perform experiments with a soft Dice loss function that directly optimizes for the intersection-over-union metric. Finally, we propose a new kind of convolution layer with a reduced amount of weight-sharing along one of the two spatial dimensions, addressing the large difference in appearance along the vertical axis of a LiDAR scan. We propose a final set of the above methods with which the model achieves an increase of 3.2% in mIoU segmentation performance over the baseline while requiring only 42% of the original inference time.

Perturb More, Trap More: Understanding Behaviors of Graph Neural Networks Machine Learning

While graph neural networks (GNNs) have shown a great potential in various tasks on graph, the lack of transparency has hindered understanding how GNNs arrived at its predictions. Although few explainers for GNNs are explored, the consideration of local fidelity, indicating how the model behaves around an instance should be predicted, is neglected. In this paper, we first propose a novel post-hoc framework based on local fidelity for any trained GNNs - TraP2, which can generate a high-fidelity explanation. Considering that both relevant graph structure and important features inside each node need to be highlighted, a three-layer architecture in TraP2 is designed: i) interpretation domain are defined by Translation layer in advance; ii) local predictive behavior of GNNs being explained are probed and monitored by Perturbation layer, in which multiple perturbations for graph structure and feature-level are conducted in interpretation domain; iii) high faithful explanations are generated by fitting the local decision boundary through Paraphrase layer. Finally, TraP2 is evaluated on six benchmark datasets based on five desired attributions: accuracy, fidelity, decisiveness, insight and inspiration, which achieves $10.2\%$ higher explanation accuracy than the state-of-the-art methods.

Machine learning for causal inference: on the use of cross-fit estimators Machine Learning

Modern causal inference methods allow machine learning to be used to weaken parametric modeling assumptions. However, the use of machine learning may result in bias and incorrect inferences due to overfitting. Cross-fit estimators have been proposed to eliminate this bias and yield better statistical properties. We conducted a simulation study to assess the performance of several different estimators for the average causal effect (ACE). The data generating mechanisms for the simulated treatment and outcome included log-transforms, polynomial terms, and discontinuities. We compared singly-robust estimators (g-computation, inverse probability weighting) and doubly-robust estimators (augmented inverse probability weighting, targeted maximum likelihood estimation). Nuisance functions were estimated with parametric models and ensemble machine learning, separately. We further assessed cross-fit doubly-robust estimators. With correctly specified parametric models, all of the estimators were unbiased and confidence intervals achieved nominal coverage. When used with machine learning, the cross-fit estimators substantially outperformed all of the other estimators in terms of bias, variance, and confidence interval coverage. Due to the difficulty of properly specifying parametric models in high dimensional data, doubly-robust estimators with ensemble learning and cross-fitting may be the preferred approach for estimation of the ACE in most epidemiologic studies. However, these approaches may require larger sample sizes to avoid finite-sample issues.

Towards Understanding Normalization in Neural ODEs Machine Learning

Normalization is an important and vastly investigated technique in deep learning. However, its role for Ordinary Differential Equation based networks (neural ODEs) is still poorly understood. This paper investigates how different normalization techniques affect the performance of neural ODEs. Particularly, we show that it is possible to achieve 93% accuracy in the CIFAR-10 classification task, and to the best of our knowledge, this is the highest reported accuracy among neural ODEs tested on this problem.

A Benchmark Study on Time Series Clustering Machine Learning

This paper presents the first time series clustering benchmark utilizing all time series datasets currently available in the University of California Riverside (UCR) archive -- the state of the art repository of time series data. Specifically, the benchmark examines eight popular clustering methods representing three categories of clustering algorithms (partitional, hierarchical and density-based) and three types of distance measures (Euclidean, dynamic time warping, and shape-based). We lay out six restrictions with special attention to making the benchmark as unbiased as possible. A phased evaluation approach was then designed for summarizing dataset-level assessment metrics and discussing the results. The benchmark study presented can be a useful reference for the research community on its own; and the dataset-level assessment metrics reported may be used for designing evaluation frameworks to answer different research questions.