University of Wisconsin-Madison Abstract Neural networks have great success in many machine learning applications, but the fundamental learning theory behind them remains largely unsolved. Learning neural networks is NPhard, but in practice, simple algorithms like stochastic gradient descent(SGD) often produce good solutions. Moreover, it is observed that overparameterization-- designing networks whose number of parameters is larger than statistically needed to perfectly fit the data -- improves both optimization and generalization, appearing to contradict traditional learning theory. In this work, we extend the theoretical understanding of two and three-layer neural networks in the overparameterized regime. We prove that, using overparameterized neural networks, one can (improperly) learn some notable hypothesis classes, including two and three-layer neural networks with fewer parameters. Moreover, the learning process can be simply done by SGD or its variants in polynomial time using polynomially many samples. We also show that for a fixed sample size, the generalization error of the solution found by some SGD variant can be made almost independent of the number of parameters in the overparameterized network. Authors sorted in alphabetical order. 1 Introduction In contrast to the widely accepted empirical success, much less theory is known. Despite a recent increase of theoretical studies, many questions remain largely open, including fundamental ones about the optimization and generalization in learning neural networks.

In this work we explore the generalization characteristics of unsupervised representation learning by leveraging disentangled VAE's to learn a useful latent space on a set of relational reasoning problems derived from Raven Progressive Matrices. We show that the latent representations, learned by unsupervised training using the right objective function, significantly outperform the same architectures trained with purely supervised learning, especially when it comes to generalization.

Extracting actionable insight from Electronic Health Records (EHRs) poses several challenges for traditional machine learning approaches. Patients are often missing data relative to each other; the data comes in a variety of modalities, such as multivariate time series, free text, and categorical demographic information; important relationships among patients can be difficult to detect; and many others. In this work, we propose a novel approach to address these first three challenges using a representation learning scheme based on message passing. We show that our proposed approach is competitive with or outperforms the state of the art for predicting in-hospital mortality (binary classification), the length of hospital visits (regression) and the discharge destination (multiclass classification).

ABSTRACT Recent work has shown that exploiting relations between labels improves the performance of multi-label classification. We propose a novel framework based on generative adversarial networks (GANs) to model label dependency. The discriminator learns to model label dependency by discriminating real and generated label sets. To fool the discriminator, the classifier, or generator, learns to generate label sets with dependencies close to real data. Extensive experiments and comparisons on two large-scale image classification benchmark datasets (MS-COCO and NUS-WIDE) show that the discriminator improves generalization ability for different kinds of models.

Comprehensive surgical planning require complex patient-specific anatomical models. For instance, functional muskuloskeletal simulations necessitate all relevant structures to be segmented, which could be performed in real-time using deep neural networks given sufficient annotated samples. Such large datasets of multiple structure annotations are costly to procure and are often unavailable in practice. Nevertheless, annotations from different studies and centers can be readily available, or become available in the future in an incremental fashion. We propose a class-incremental segmentation framework for extending a deep network trained for some anatomical structure to yet another structure using a small incremental annotation set. Through distilling knowledge from the current state of the framework, we bypass the need for a full retraining. This is a meta-method to extend any choice of desired deep segmentation network with only a minor addition per structure, which makes it suitable for lifelong class-incremental learning and applicable also for future deep neural network architectures. We evaluated our methods on a public knee dataset of 100 MR volumes. Through varying amount of incremental annotation ratios, we show how our proposed method can retain the previous anatomical structure segmentation performance superior to the conventional finetuning approach. In addition, our framework inherently exploits transferable knowledge from previously trained structures to incremental tasks, demonstrated by superior results compared to non-incremental training. With the presented method, new anatomical structures can be learned without catastrophic forgetting of older structures and without extensive increase of memory and complexity.

Evolution Strategies (ES) emerged as a scalable alternative to popular Reinforcement Learning (RL) techniques, providing an almost perfect speedup when distributed across hundreds of CPU cores thanks to a reduced communication overhead. Despite providing large improvements in wall-clock time, ES is data inefficient when compared to competing RL methods. One of the main causes of such inefficiency is the collection of large batches of experience, which are discarded after each policy update. In this work, we study how to perform more than one update per batch of experience by means of Importance Sampling while preserving the scalability of the original method. The proposed method, Importance Weighted Evolution Strategies (IW-ES), shows promising results and is a first step towards designing efficient ES algorithms.

Graph classification has recently received a lot of attention from various fields of machine learning e.g. kernel methods, sequential modeling or graph embedding. All these approaches offer promising results with different respective strengths and weaknesses. However, most of them rely on complex mathematics and require heavy computational power to achieve their best performance. We propose a simple and fast algorithm based on the spectral decomposition of graph Laplacian to perform graph classification and get a first reference score for a dataset. We show that this method obtains competitive results compared to state-of-the-art algorithms.

Distance metric learning (DML) aims to find an appropriate way to reveal the underlying data relationship. It is critical in many machine learning, pattern recognition and data mining algorithms, and usually require large amount of label information (such as class labels or pair/triplet constraints) to achieve satisfactory performance. However, the label information may be insufficient in real-world applications due to the high-labeling cost, and DML may fail in this case. Transfer metric learning (TML) is able to mitigate this issue for DML in the domain of interest (target domain) by leveraging knowledge/information from other related domains (source domains). Although achieved a certain level of development, TML has limited success in various aspects such as selective transfer, theoretical understanding, handling complex data, big data and extreme cases. In this survey, we present a systematic review of the TML literature. In particular, we group TML into different categories according to different settings and metric transfer strategies, such as direct metric approximation, subspace approximation, distance approximation, and distribution approximation. A summarization and insightful discussion of the various TML approaches and their applications will be presented. Finally, we indicate some challenges and provide possible future directions.

In many reinforcement learning tasks, the goal is to learn a policy to manipulate an agent, whose design is fixed, to maximize some notion of cumulative reward. The design of the agent's physical structure is rarely optimized for the task at hand. In this work, we explore the possibility of learning a version of the agent's design that is better suited for its task, jointly with the policy. We propose a minor alteration to the OpenAI Gym framework, where we parameterize parts of an environment, and allow an agent to jointly learn to modify these environment parameters along with its policy. We demonstrate that an agent can learn a better structure of its body that is not only better suited for the task, but also facilitates policy learning. Joint learning of policy and structure may even uncover design principles that are useful for assisted-design applications. Videos of results at https://designrl.github.io/

We propose a generic framework to calibrate accuracy and confidence (score) of a prediction through stochastic inferences in deep neural networks. We first analyze relation between variation of multiple model parameters for a single example inference and variance of the corresponding prediction scores by Bayesian modeling of stochastic regularization. Our empirical observation shows that accuracy and score of a prediction are highly correlated with variance of multiple stochastic inferences given by stochastic depth or dropout. Motivated by these facts, we design a novel variance-weighted confidence-integrated loss function that is composed of two cross-entropy loss terms with respect to ground-truth and uniform distribution, which are balanced by variance of stochastic prediction scores. The proposed loss function enables us to learn deep neural networks that predict confidence calibrated scores using a single inference. Our algorithm presents outstanding confidence calibration performance and improves classification accuracy with two popular stochastic regularization techniques---stochastic depth and dropout---in multiple models and datasets; it alleviates overconfidence issue in deep neural networks significantly by training networks to achieve prediction accuracy proportional to confidence of prediction.