Convolutional neural networks are among the most successful architectures in deep learning. This success is at least partially attributable to the efficacy of spatial invariance as an inductive bias. Locally connected layers, which differ from convolutional layers in their lack of spatial invariance, usually perform poorly in practice. However, these observations still leave open the possibility that some degree of relaxation of spatial invariance may yield a better inductive bias than either convolution or local connectivity. To test this hypothesis, we design a method to relax the spatial invariance of a network layer in a controlled manner. In particular, we create a extit{low-rank} locally connected layer, where the filter bank applied at each position is constructed as a linear combination of basis set of filter banks. By varying the number of filter banks in the basis set, we can control the degree of departure from spatial invariance. In our experiments, we find that relaxing spatial invariance improves classification accuracy over both convolution and locally connected layers across MNIST, CIFAR-10, and CelebA datasets. These results suggest that spatial invariance in convolution layers may be overly restrictive.

How Does Gender Balance In Training Data Affect Face Recognition Accuracy?.

Even though deep learning methods have greatly increased the overall accuracy of face recognition, an old problem still persists: accuracy is higher for men than for women. Previous researchers have speculated that the difference could be due to cosmetics, head pose, or hair covering the face. It is also often speculated that the lower accuracy for women is caused by women being under-represented in the training data. This work aims to investigate if gender imbalance in the training data is actually the cause of lower accuracy for females. Using a state-of-the-art deep CNN, three different loss functions, and two training datasets, we train each on seven subsets with different male/female ratios, totaling forty two train-ings. The trained face matchers are then tested on three different testing datasets. Results show that gender-balancing the dataset has an overall positive effect, with higher accuracy for most of the combinations of loss functions and datasets when a balanced subset is used. However, for the best combination of loss function and dataset, the original training dataset shows better accuracy on 3 out of 4 times. We observe that test accuracy for males is higher when the training data is all male. However, test accuracy for females is not maximized when the training data is all female. Fora number of combinations of loss function and test dataset, accuracy for females is higher when only 75% of the train-ing data is female than when 100% of the training data is female. This suggests that lower accuracy for females is nota simple result of the fraction of female training data. By clustering face features, we show that in general, male faces are closer to other male faces than female faces, and female faces are closer to other female faces than male faces

Image-to-Image Translation with Text Guidance.

The goal of this paper is to embed controllable factors, i.e., natural language descriptions, into image-to-image translation with generative adversarial networks, which allows text descriptions to determine the visual attributes of synthetic images. We propose four key components: (1) the implementation of part-of-speech tagging to filter out non-semantic words in the given description, (2) the adoption of an affine combination module to effectively fuse different modality text and image features, (3) a novel refined multi-stage architecture to strengthen the differential ability of discriminators and the rectification ability of generators, and (4) a new structure loss to further improve discriminators to better distinguish real and synthetic images. Extensive experiments on the COCO dataset demonstrate that our method has a superior performance on both visual realism and semantic consistency with given descriptions.

Ensemble of Deep Convolutional Neural Networks for Automatic Pavement Crack Detection and Measurement.

Automated pavement crack detection and measurement are important road issues. Agencies have to guarantee the improvement of road safety. Conventional crack detection and measurement algorithms can be extremely time-consuming and low efficiency. Therefore, recently, innovative algorithms have received increased attention from researchers. In this paper, we propose an ensemble of convolutional neural networks (without a pooling layer) based on probability fusion for automated pavement crack detection and measurement. Specifically, an ensemble of convolutional neural networks was employed to identify the structure of small cracks with raw images. Secondly, outputs of the individual convolutional neural network model for the ensemble were averaged to produce the final crack probability value of each pixel, which can obtain a predicted probability map. Finally, the predicted morphological features of the cracks were measured by using the skeleton extraction algorithm. To validate the proposed method, some experiments were performed on two public crack databases (CFD and AigleRN) and the results of the different state-of-the-art methods were compared. The experimental results show that the proposed method outperforms the other methods. For crack measurement, the crack length and width can be measure based on different crack types (complex, common, thin, and intersecting cracks.). The results show that the proposed algorithm can be effectively applied for crack measurement.

PointHop++: A Lightweight Learning Model on Point Sets for 3D Classification.

The PointHop method was recently proposed by Zhang et al. for 3D point cloud classification with unsupervised feature extraction. It has an extremely low training complexity while achieving state-of-the-art classification performance. In this work, we improve the PointHop method furthermore in two aspects: 1) reducing its model complexity in terms of the model parameter number and 2) ordering discriminant features automatically based on the cross-entropy criterion. The resulting method is called PointHop++. The first improvement is essential for wearable and mobile computing while the second improvement bridges statistics-based and optimization-based machine learning methodologies. With experiments conducted on the ModelNet40 benchmark dataset, we show that the PointHop++ method performs on par with deep neural network (DNN) solutions and surpasses other unsupervised feature extraction methods.

Finding novelty with uncertainty.

Medical images are often used to detect and characterize pathology and disease; however, automatically identifying and segmenting pathology in medical images is challenging because the appearance of pathology across diseases varies widely. To address this challenge, we propose a Bayesian deep learning method that learns to translate healthy computed tomography images to magnetic resonance images and simultaneously calculates voxel-wise uncertainty. Since high uncertainty occurs in pathological regions of the image, this uncertainty can be used for unsupervised anomaly segmentation. We show encouraging experimental results on an unsupervised anomaly segmentation task by combining two types of uncertainty into a novel quantity we call scibilic uncertainty.

Validating uncertainty in medical image translation.

Medical images are increasingly used as input to deep neural networks to produce quantitative values that aid researchers and clinicians. However, standard deep neural networks do not provide a reliable measure of uncertainty in those quantitative values. Recent work has shown that using dropout during training and testing can provide estimates of uncertainty. In this work, we investigate using dropout to estimate epistemic and aleatoric uncertainty in a CT-to-MR image translation task. We show that both types of uncertainty are captured, as defined, providing confidence in the output uncertainty estimates.

Adaptivity of Stochastic Gradient Methods for Nonconvex Optimization.

Adaptivity is an important yet under-studied property in modern optimization theory. The gap between the state-of-the-art theory and the current practice is striking in that algorithms with desirable theoretical guarantees typically involve drastically different settings of hyperparameters, such as step-size schemes and batch sizes, in different regimes. Despite the appealing theoretical results, such divisive strategies provide little, if any, insight to practitioners to select algorithms that work broadly without tweaking the hyperparameters. In this work, blending the "geometrization" technique introduced by Lei & Jordan 2016 and the exttt{SARAH} algorithm of Nguyen et al., 2017, we propose the Geometrized exttt{SARAH} algorithm for non-convex finite-sum and stochastic optimization. Our algorithm is proved to achieve adaptivity to both the magnitude of the target accuracy and the Polyak-L{}ojasiewicz (PL) constant if present. In addition, it achieves the best-available convergence rate for non-PL objectives simultaneously while outperforming existing algorithms for PL objectives.

Weakly-Supervised Disentanglement Without Compromises.

Intelligent agents should be able to learn useful representations by observing changes in their environment. We model such observations as pairs of non-i.i.d. images sharing at least one of the underlying factors of variation. First, we theoretically show that only knowing how many factors have changed, but not which ones, is sufficient to learn disentangled representations. Second, we provide practical algorithms that learn disentangled representations from pairs of images without requiring annotation of groups, individual factors, or the number of factors that have changed. Third, we perform a large-scale empirical study and show that such pairs of observations are sufficient to reliably learn disentangled representations on several benchmark data sets. Finally, we evaluate our learned representations and find that they are simultaneously useful on a diverse suite of tasks, including generalization under covariate shifts, fairness, and abstract reasoning. Overall, our results demonstrate that weak supervision enables learning of useful disentangled representations in realistic scenarios.

Capsules with Inverted Dot-Product Attention Routing.

We introduce a new routing algorithm for capsule networks, in which a child capsule is routed to a parent based only on agreement between the parents state and the childs vote. The new mechanism 1) designs routing via inverted dot-product attention; 2) imposes Layer Normalization as normalization; and 3) replaces sequential iterative routing with concurrent iterative routing. When compared to previously proposed routing algorithms, our method improves performance on benchmark datasets such as CIFAR-10 and CIFAR-100, and it performs at-par with a powerful CNN (ResNet-18) with 4x fewer parameters. On a different task of recognizing digits from overlayed digit images, the proposed capsule model performs favorably against CNNs given the same number of layers and neurons per layer. We believe that our work raises the possibility of applying capsule networks to complex real-world tasks. Our code is publicly available at: https://github.com/apple/ml-capsules-inverted-attention-routing

Cyclic Boosting -- an explainable supervised machine learning algorithm.

Supervised machine learning algorithms have seen spectacular advances and surpassed human level performance in a wide range of specific applications. However, using complex ensemble or deep learning algorithms typically results in black box models, where the path leading to individual predictions cannot be followed in detail. In order to address this issue, we propose the novel "Cyclic Boosting" machine learning algorithm, which allows to efficiently perform accurate regression and classification tasks while at the same time allowing a detailed understanding of how each individual prediction was made.

DynamicPPL: Stan-like Speed for Dynamic Probabilistic Models.

We present the preliminary high-level design and features of DynamicPPL.jl, a modular library providing a lightning-fast infrastructure for probabilistic programming. Besides a computational performance that is often close to or better than Stan, DynamicPPL provides an intuitive DSL that allows the rapid development of complex dynamic probabilistic programs. Being entirely written in Julia, a high-level dynamic programming language for numerical computing, DynamicPPL inherits a rich set of features available through the Julia ecosystem. Since DynamicPPL is a modular, stand-alone library, any probabilistic programming system written in Julia, such as Turing.jl, can use DynamicPPL to specify models and trace their model parameters. The main features of DynamicPPL are: 1) a meta-programming based DSL for specifying dynamic models using an intuitive tilde-based notation; 2) a tracing data-structure for tracking RVs in dynamic probabilistic models; 3) a rich contextual dispatch system allowing tailored behaviour during model execution; and 4) a user-friendly syntax for probabilistic queries. Finally, we show in a variety of experiments that DynamicPPL, in combination with Turing.jl, achieves computational performance that is often close to or better than Stan.

Explore, Discover and Learn: Unsupervised Discovery of State-Covering Skills.

Acquiring abilities in the absence of a task-oriented reward function is at the frontier of reinforcement learning research. This problem has been studied through the lens of empowerment, which draws a connection between option discovery and information theory. Information-theoretic skill discovery methods have garnered much interest from the community, but little research has been conducted in understanding their limitations. Through theoretical analysis and empirical evidence, we show that existing algorithms suffer from a common limitation -- they discover options that provide a poor coverage of the state space. In light of this, we propose Explore, Discover and Learn (EDL), an alternative approach to information-theoretic skill discovery. Crucially, EDL optimizes the same information-theoretic objective derived from the empowerment literature, but addresses the optimization problem using different machinery. We perform an extensive evaluation of skill discovery methods on controlled environments and show that EDL offers significant advantages, such as overcoming the coverage problem, reducing the dependence of learned skills on the initial state, and allowing the user to define a prior over which behaviors should be learned.

Towards Mixture Proportion Estimation without Irreducibility.

extit{Mixture proportion estimation} (MPE) is a fundamental problem of practical significance, where we are given data from only a extit{mixture} and one of its two extit{components} to identify the proportion of each component. All existing MPE methods that are distribution-independent explicitly or implicitly rely on the extit{irreducible} assumption---the unobserved component is not a mixture containing the observable component. If this is not satisfied, those methods will lead to a critical estimation bias. In this paper, we propose extit{Regrouping-MPE} that works without irreducible assumption: it builds a new irreducible MPE problem and solves the new problem. It is worthwhile to change the problem: we prove that if the assumption holds, our method will not affect anything; if the assumption does not hold, the bias from problem changing is less than the bias from violation of the irreducible assumption in the original problem. Experiments show that our method outperforms all state-of-the-art MPE methods on various real-world datasets.

On the Effectiveness of Richardson Extrapolation in Machine Learning.

Richardson extrapolation is a classical technique from numerical analysis that can improve the approximation error of an estimation method by combining linearly several estimates obtained from different values of one of its hyperparameters, without the need to know in details the inner structure of the original estimation method. The main goal of this paper is to study when Richardson extrapolation can be used within machine learning, beyond the existing applications to step-size adaptations in stochastic gradient descent. We identify two situations where Richardson interpolation can be useful: (1) when the hyperparameter is the number of iterations of an existing iterative optimization algorithm, with applications to averaged gradient descent and Frank-Wolfe algorithms (where we obtain asymptotically rates of O(1/k^2) on polytopes, where k is the number of iterations), and (2) when it is a regularization parameter, with applications to Nesterov smoothing techniques for minimizing non-smooth functions (where we obtain asymptotically rates close to O(1/k^2) for non-smooth functions), and ridge regression. In all these cases, we show that extrapolation techniques come with no significant loss in performance, but with sometimes strong gains, and we provide theoretical justifications based on asymptotic developments for such gains, as well as empirical illustrations on classical problems from machine learning.

Learning State Abstractions for Transfer in Continuous Control.

Can simple algorithms with a good representation solve challenging reinforcement learning problems? In this work, we answer this question in the affirmative, where we take "simple learning algorithm" to be tabular Q-Learning, the "good representations" to be a learned state abstraction, and "challenging problems" to be continuous control tasks. Our main contribution is a learning algorithm that abstracts a continuous state-space into a discrete one. We transfer this learned representation to unseen problems to enable effective learning. We provide theory showing that learned abstractions maintain a bounded value loss, and we report experiments showing that the abstractions empower tabular Q-Learning to learn efficiently in unseen tasks.

Consistency of a Recurrent Language Model With Respect to Incomplete Decoding.

Despite strong performance on a variety of tasks, neural sequence models trained with maximum likelihood have been shown to exhibit issues such as length bias and degenerate repetition. We study the related issue of receiving infinite-length sequences from a recurrent language model when using common decoding algorithms. To analyze this issue, we first define inconsistency of a decoding algorithm, meaning that the algorithm can yield an infinite-length sequence that has zero probability under the model. We prove that commonly used incomplete decoding algorithms - greedy search, beam search, top-k sampling, and nucleus sampling - are inconsistent, despite the fact that recurrent language models are trained to produce sequences of finite length. Based on these insights, we propose two remedies which address inconsistency: consistent variants of top-k and nucleus sampling, and a self-terminating recurrent language model. Empirical results show that inconsistency occurs in practice, and that the proposed methods prevent inconsistency.