We present a deep recurrent neural network architecture to solve a class of stochastic optimal control problems described by fully nonlinear Hamilton Jacobi Bellman partial differential equations. Such PDEs arise when one considers stochastic dynamics characterized by uncertainties that are additive and control multiplicative. Stochastic models with the aforementioned characteristics have been used in computational neuroscience, biology and aerospace systems and provide a more accurate representation of actuation than models with additive uncertainty. Previous literature has established the inadequacy of the linear HJB theory and and instead rely on a non-linear Feynman-Kac lemma resulting in a second order forward-backward stochastic differential equations representation. However, the proposed solutions that use this representation suffer from compounding errors and computational complexity leading to lack of scalability. In this paper, we propose a deep learning based algorithm that leverages the second order Forward-Bacward SDE representation along with importance sampling and LSTM based recurrent neural networks to not only solve such Stochastic Optimal Control problems but also overcome the problems faced by previous approaches and scales well to high dimensional systems. The resulting control algorithm is tested on three non-linear systems to demonstrate feasibility and out-performance against previous methods.

The field of deep learning is experiencing a trend towards producing reproducible research. Nevertheless, it is still often a frustrating experience to reproduce scientific results. This is especially true in the machine learning community, where it is considered acceptable to have black boxes in your experiments. We present DeepDIVA, a framework designed to facilitate easy experimentation and their reproduction. This framework allows researchers to share their experiments with others, while providing functionality that allows for easy experimentation, such as: boilerplate code, experiment management, hyper-parameter optimization, verification of data integrity and visualization of data and results. Additionally, the code of DeepDIVA is well-documented and supported by several tutorials that allow a new user to quickly familiarize themselves with the framework.

The main purpose of incremental learning is to learn new knowledge while not forgetting the knowledge which have been learned before. At present, the main challenge in this area is the catastrophe forgetting, namely the network will lose their performance in the old tasks after training for new tasks. In this paper, we introduce an ensemble method of incremental classifier to alleviate this problem, which is based on the cosine distance between the output feature and the pre-defined center, and can let each task to be preserved in different networks. During training, we make use of PEDCC-Loss to train the CNN network. In the stage of testing, the prediction is determined by the cosine distance between the network latent features and pre-defined center. The experimental results on EMINST and CIFAR100 show that our method outperforms the recent LwF method, which use the knowledge distillation, and iCaRL method, which keep some old samples while training for new task. The method can achieve the goal of not forgetting old knowledge while training new classes, and solve the problem of catastrophic forgetting better.

We introduce a data-free quantization method for deep neural networks that does not require fine-tuning or hyperparameter selection. It achieves near-original model performance on common computer vision architectures and tasks. 8-bit fixed-point quantization is essential for efficient inference in modern deep learning hardware architectures. However, quantizing models to run in 8-bit is a non-trivial task, frequently leading to either significant performance reduction or engineering time spent on training a network to be amenable to quantization. Our approach relies on equalizing the weight ranges in the network by making use of a scale-equivariance property of activation functions. In addition the method corrects biases in the error that are introduced during quantization. This improves quantization accuracy performance, and can be applied ubiquitously to almost any model with a straight-forward API call. For common architectures, such as the MobileNet family, we achieve state-of-the-art quantized model performance. We further show that the method also extends to other computer vision architectures and tasks such as semantic segmentation and object detection.

Effective modeling of electronic health records (EHR) is rapidly becoming an important topic in both academia and industry. A recent study showed that utilizing the graphical structure underlying EHR data (e.g. relationship between diagnoses and treatments) improves the performance of prediction tasks such as heart failure diagnosis prediction. However, EHR data do not always contain complete structure information. Moreover, when it comes to claims data, structure information is completely unavailable to begin with. Under such circumstances, can we still do better than just treating EHR data as a flat-structured bag-of-features? In this paper, we study the possibility of utilizing the implicit structure of EHR by using the Transformer for prediction tasks on EHR data. Specifically, we argue that the Transformer is a suitable model to learn the hidden EHR structure, and propose the Graph Convolutional Transformer, which uses data statistics to guide the structure learning process. Our model empirically demonstrated superior prediction performance to previous approaches on both synthetic data and publicly available EHR data on encounter-based prediction tasks such as graph reconstruction and readmission prediction, indicating that it can serve as an effective general-purpose representation learning algorithm for EHR data.

A recent line of research has shown that gradient-based algorithms with random initialization can converge to the global minima of the training loss for over-parameterized (i.e., sufficiently wide) deep neural networks. However, the condition on the width of the neural network to ensure the global convergence is very stringent, which is often a high-degree polynomial in the training sample size $n$ (e.g., $O(n^{24})$). In this paper, we provide an improved analysis of the global convergence of (stochastic) gradient descent for training deep neural networks, which only requires a milder over-parameterization condition than previous work in terms of the training sample size and other problem-dependent parameters. The main technical contributions of our analysis include (a) a tighter gradient lower bound that leads to a faster convergence of the algorithm, and (b) a sharper characterization of the trajectory length of the algorithm. By specializing our result to two-layer (i.e., one-hidden-layer) neural networks, it also provides a milder over-parameterization condition than the best-known result in prior work.

Convolutional neural networks (CNNs) are widely used for classification problems. However, they often require large amounts of computation and memory which are not readily available in resource constrained systems. Pruning unimportant parameters from CNNs to reduce these requirements has been a subject of intensive research in recent years. However, novel approaches in pruning signals are sometimes difficult to compare against each other. We propose a taxonomy that classifies pruning signals based on four mostly-orthogonal components of the signal. We also empirically evaluate 396 pruning signals including existing ones, and new signals constructed from the components of existing signals. We find that some of our newly constructed signals outperform the best existing pruning signals.

Data have often to be moved between servers and clients during the inference phase. For instance, modern virtual assistants collect data on mobile devices and the data are sent to remote servers for the analysis. A related scenario is that clients have to access and download large amounts of data stored on servers in order to apply machine learning models. Depending on the available bandwidth, this data transfer can be a serious bottleneck, which can significantly limit the application machine learning models. In this work, we propose a simple yet effective framework that allows to select certain parts of the input data needed for the subsequent application of a given neural network. Both the masks as well as the neural network are trained simultaneously such that a good model performance is achieved while, at the same time, only a minimal amount of data is selected by the masks. During the inference phase, only the parts selected by the masks have to be transferred between the server and the client. Our experimental evaluation indicates that it is, for certain learning tasks, possible to significantly reduce the amount of data needed to be transferred without affecting the model performance much.

Symbol detection for Massive Multiple-Input Multiple-Output (MIMO) is a challenging problem for which traditional algorithms are either impractical or suffer from performance limitations. Several recently proposed learning-based approaches achieve promising results on simple channel models (e.g., i.i.d. Gaussian). However, their performance degrades significantly on real-world channels with spatial correlation. We propose MMNet, a deep learning MIMO detection scheme that significantly outperforms existing approaches on realistic channels with the same or lower computational complexity. MMNet's design builds on the theory of iterative soft-thresholding algorithms and uses a novel training algorithm that leverages temporal and spectral correlation to accelerate training. Together, these innovations allow MMNet to train online for every realization of the channel. On i.i.d. Gaussian channels, MMNet requires two orders of magnitude fewer operations than existing deep learning schemes but achieves near-optimal performance. On spatially-correlated channels, it achieves the same error rate as the next-best learning scheme (OAMPNet) at 2.5dB lower SNR and with at least 10x less computational complexity. MMNet is also 4--8dB better overall than a classic linear scheme like the minimum mean square error (MMSE) detector.

With the widespread success of deep neural networks in science and technology, it is becoming increasingly important to quantify the uncertainty of the predictions produced by deep learning. In this paper, we introduce a new method that attaches an explicit uncertainty statement to the probabilities of classification using deep neural networks. Precisely, we view that the classification probabilities are sampled from an unknown distribution, and we propose to learn this distribution through the Dirichlet mixture that is flexible enough for approximating any continuous distribution on the simplex. We then construct credible intervals from the learned distribution to assess the uncertainty of the classification probabilities. Our approach is easy to implement, computationally efficient, and can be coupled with any deep neural network architecture. Our method leverages the crucial observation that, in many classification applications such as medical diagnosis, more than one class labels are available for each observational unit. We demonstrate the usefulness of our approach through simulations and a real data example.