Reinforcement learning algorithms are known to be sample inefficient, and often performance on one task can be substantially improved by leveraging information (e.g., via pre-training) on other related tasks. In this work, we propose a technique to achieve such knowledge transfer in cases where agent trajectories contain sensitive or private information, such as in the healthcare domain. Our approach leverages a differentially private policy evaluation algorithm to initialize an actor-critic model and improve the effectiveness of learning in downstream tasks. We empirically show this technique increases sample efficiency in resource-constrained control problems while preserving the privacy of trajectories collected in an upstream task.

Deep neural networks have been used in various machine learning applications and achieved tremendous empirical successes. However, training deep neural networks is a challenging task. Many alternatives have been proposed in place of end-to-end back-propagation. Layer-wise training is one of them, which trains a single layer at a time, rather than trains the whole layers simultaneously. In this paper, we study a layer-wise training using a block coordinate gradient descent (BCGD) for deep linear networks. We establish a general convergence analysis of BCGD and found the optimal learning rate, which results in the fastest decrease in the loss. More importantly, the optimal learning rate can directly be applied in practice, as it does not require any prior knowledge. Thus, tuning the learning rate is not needed at all. Also, we identify the effects of depth, width, and initialization in the training process. We show that when the orthogonal-like initialization is employed, the width of intermediate layers plays no role in gradient-based training, as long as the width is greater than or equal to both the input and output dimensions. We show that under some conditions, the deeper the network is, the faster the convergence is guaranteed. This implies that in an extreme case, the global optimum is achieved after updating each weight matrix only once. Besides, we found that the use of deep networks could drastically accelerate convergence when it is compared to those of a depth 1 network, even when the computational cost is considered. Numerical examples are provided to justify our theoretical findings and demonstrate the performance of layer-wise training by BCGD.

Data augmentation techniques, e.g., flipping or cropping, which systematically enlarge the training dataset by explicitly generating more training samples, are effective in improving the generalization performance of deep neural networks. In the supervised setting, a common practice for data augmentation is to assign the same label to all augmented samples of the same source. However, if the augmentation results in large distributional discrepancy among them (e.g., rotations), forcing their label invariance may be too difficult to solve and often hurts the performance. To tackle this challenge, we suggest a simple yet effective idea of learning the joint distribution of the original and self-supervised labels of augmented samples. The joint learning framework is easier to train, and enables an aggregated inference combining the predictions from different augmented samples for improving the performance. Further, to speed up the aggregation process, we also propose a knowledge transfer technique, self-distillation, which transfers the knowledge of augmentation into the model itself. We demonstrate the effectiveness of our data augmentation framework on various fully-supervised settings including the few-shot and imbalanced classification scenarios.

Gaussian Processes (GPs) with an appropriate kernel are known to provide accurate predictions and uncertainty estimates even with very small amounts of labeled data. However, GPs are generally unable to learn a good representation that can encode intricate structures in high dimensional data. The representation power of GPs depends heavily on kernel functions used to quantify the similarity between data points. Traditional GP kernels are not very effective at capturing similarity between high dimensional data points, while methods that use deep neural networks to learn a kernel are not sample-efficient. To overcome these drawbacks, we propose deep probabilistic kernels which use a probabilistic neural network to map high-dimensional data to a probability distribution in a low dimensional subspace, and leverage the rich work on kernels between distributions to capture the similarity between these distributions. Experiments on a variety of datasets show that building a GP using this covariance kernel solves the conflicting problems of representation learning and sample efficiency. Our model can be extended beyond GPs to other small-data paradigms such as few-shot classification where we show competitive performance with state-of-the-art models on the mini-Imagenet dataset.

Many modern large-scale machine learning problems benefit from decentralized and stochastic optimization. Recent works have shown that utilizing both decentralized computing and local stochastic gradient estimates can outperform state-of-the-art centralized algorithms, in applications involving highly non-convex problems, such as training deep neural networks. In this work, we propose a decentralized stochastic algorithm to deal with certain smooth non-convex problems where there are $m$ nodes in the system, and each node has a large number of samples (denoted as $n$). Differently from the majority of the existing decentralized learning algorithms for either stochastic or finite-sum problems, our focus is given to both reducing the total communication rounds among the nodes, while accessing the minimum number of local data samples. In particular, we propose an algorithm named D-GET (decentralized gradient estimation and tracking), which jointly performs decentralized gradient estimation (which estimates the local gradient using a subset of local samples) and gradient tracking (which tracks the global full gradient using local estimates). We show that, to achieve certain $\epsilon$ stationary solution of the deterministic finite sum problem, the proposed algorithm achieves an $\mathcal{O}(mn^{1/2}\epsilon^{-1})$ sample complexity and an $\mathcal{O}(\epsilon^{-1})$ communication complexity. These bounds significantly improve upon the best existing bounds of $\mathcal{O}(mn\epsilon^{-1})$ and $\mathcal{O}(\epsilon^{-1})$, respectively. Similarly, for online problems, the proposed method achieves an $\mathcal{O}(m \epsilon^{-3/2})$ sample complexity and an $\mathcal{O}(\epsilon^{-1})$ communication complexity, while the best existing bounds are $\mathcal{O}(m\epsilon^{-2})$ and $\mathcal{O}(\epsilon^{-2})$, respectively.

Convolutional neural networks (CNNs) demand huge DRAM bandwidth for computational imaging tasks, and block-based processing has recently been applied to greatly reduce the bandwidth. However, the induced additional computation for feature recomputing or the large SRAM for feature reusing will degrade the performance or even forbid the usage of state-of-the-art models. In this paper, we address these issues by considering the overheads and hardware constraints in advance when constructing CNNs. We investigate a novel model family---ERNet---which includes temporary layer expansion as another means for increasing model capacity. We analyze three ERNet variants in terms of hardware requirement and introduce a hardware-aware model optimization procedure. Evaluations on Full HD and 4K UHD applications will be given to show the effectiveness in terms of image quality, pixel throughput, and SRAM usage. The results also show that, for block-based inference, ERNet can outperform the state-of-the-art FFDNet and EDSR-baseline models for image denoising and super-resolution respectively.

Emotions play an important role in people's life. Understanding and recognising is not only important for interpersonal communication, but also has promising applications in Human-Computer Interaction, automobile safety and medical research. This project focuses on extending the emotion recognition database, and training the CNN + RNN emotion recognition neural networks with emotion category representation and valence \& arousal representation. The combined models are constructed by training the two representations simultaneously. The comparison and analysis between the three types of model are discussed. The inner-relationship between two emotion representations and the interpretability of the neural networks are investigated. The findings suggest that categorical emotion recognition performance can benefit from training with a combined model. And the mapping of emotion category and valence \& arousal values can explain this phenomenon.

Generative Adversarial Networks (GANs) were proposed in 2014 by Goodfellow et al., and have since been extended into multiple computer vision applications. This report provides a thorough survey of recent GAN research, outlining the various architectures and applications, as well as methods for training GANs and dealing with latent space. This is followed by a discussion of potential areas for future GAN research, including: evaluating GANs, better understanding GANs, and techniques for training GANs. The second part of this report outlines the compilation of a dataset of images `in the wild' representing each of the 7 basic human emotions, and analyses experiments done when training a StarGAN on this dataset combined with the FER2013 dataset.

Mean field theory has been successfully used to analyze deep neura l networks (DNN) in the infinite size limit. Given the finite size of realistic D NN, we utilize the large deviation theory and path integral analysis to study the deviation of functions represented by DNN from their typical mean field solution s. The parameter perturbations investigated include weight sparsification (dilution) a nd binarization, which are commonly used in model simplification, for both ReLU and sign activation functions. We find that random networks with ReLU activation are m ore robust to parameter perturbations with respect to their counterparts wit h sign activation, which arguably is reflected in the simplicity of the functions they generate . Keywords: large deviation theory, path integral, deep neural networks, fu nction sensitivity 1. Introduction Learning machines realized by deep neural networks (DNN) have ac hieved impressive success in performing various machine learning tasks, such as spee ch recognition, image classification and natural language processing [1].

Recent work in signal propagation theory has shown that dropout limits the depth to which information can propagate through a neural network. In this paper, we investigate the effect of initialisation on training speed and generalisation for ReLU networks within this depth limit. We ask the following research question: given that critical initialisation is crucial for training at large depth, if dropout limits the depth at which networks are trainable, does initialising critically still matter? We conduct a large-scale controlled experiment, and perform a statistical analysis of over $12000$ trained networks. We find that (1) trainable networks show no statistically significant difference in performance over a wide range of non-critical initialisations; (2) for initialisations that show a statistically significant difference, the net effect on performance is small; (3) only extreme initialisations (very small or very large) perform worse than criticality. These findings also apply to standard ReLU networks of moderate depth as a special case of zero dropout. Our results therefore suggest that, in the shallow-to-moderate depth setting, critical initialisation provides zero performance gains when compared to off-critical initialisations and that searching for off-critical initialisations that might improve training speed or generalisation, is likely to be a fruitless endeavour.