In recent years, considerable progress has been made towards a vehicle's ability to operate autonomously. An end-to-end approach attempts to achieve autonomous driving using a single, comprehensive software component. Recent breakthroughs in deep learning have significantly increased end-to-end systems' capabilities, and such systems are now considered a possible alternative to the current state-of-the-art solutions. This paper examines end-to-end learning for autonomous vehicles in simulated urban environments containing other vehicles, traffic lights, and speed limits. Furthermore, the paper explores end-to-end systems' ability to execute navigational commands and examines whether improved performance can be achieved by utilizing temporal dependencies between subsequent visual cues. Two end-to-end architectures are proposed: a traditional Convolutional Neural Network and an extended design combining a Convolutional Neural Network with a recurrent layer. The models are trained using expert driving data from a simulated urban setting, and are evaluated by their driving performance in an unseen simulated environment. The results of this paper indicate that end-to-end systems can operate autonomously in simple urban environments. Moreover, it is found that the exploitation of temporal information in subsequent images enhances a system's ability to judge movement and distance.

Estimating and optimizing Mutual Information (MI) is core to many problems in machine learning; however, bounding MI in high dimensions is challenging. To establish tractable and scalable objectives, recent work has turned to variational bounds parameterized by neural networks, but the relationships and tradeoffs between these bounds remains unclear. In this work, we unify these recent developments in a single framework. We find that the existing variational lower bounds degrade when the MI is large, exhibiting either high bias or high variance. To address this problem, we introduce a continuum of lower bounds that encompasses previous bounds and flexibly trades off bias and variance. On high-dimensional, controlled problems, we empirically characterize the bias and variance of the bounds and their gradients and demonstrate the effectiveness of our new bounds for estimation and representation learning.

Many real-world time series, such as in health, have changepoints where the system's structure or parameters change. Since changepoints can indicate critical events such as onset of illness, it is highly important to detect them. However, existing methods for changepoint detection (CPD) often require user-specified models and cannot recognize changes that occur gradually or at multiple time-scales. To address both, we show how CPD can be treated as a supervised learning problem, and propose a new deep neural network architecture to efficiently identify both abrupt and gradual changes at multiple timescales from multivariate data. Our proposed pyramid recurrent neural network (PRN) provides scale-invariance using wavelets and pyramid analysis techniques from multi-scale signal processing. Through experiments on synthetic and real-world datasets, we show that PRN can detect abrupt and gradual changes with higher accuracy than the state of the art and can extrapolate to detect changepoints at novel scales not seen in training.

Access to sufficient annotated data is a common challenge in training deep neural networks on medical images. As annotating data is expensive and time-consuming, it is difficult for an individual medical center to reach large enough sample sizes to build their own, personalized models. As an alternative, data from all centers could be pooled to train a centralized model that everyone can use. However, such a strategy is often infeasible due to the privacy-sensitive nature of medical data. Recently, federated learning (FL) has been introduced to collaboratively learn a shared prediction model across centers without the need for sharing data. In FL, clients are locally training models on site-specific datasets for a few epochs and then sharing their model weights with a central server, which orchestrates the overall training process. Importantly, the sharing of models does not compromise patient privacy. A disadvantage of FL is the dependence on a central server, which requires all clients to agree on one trusted central body, and whose failure would disrupt the training process of all clients. In this paper, we introduce BrainTorrent, a new FL framework without a central server, particularly targeted towards medical applications. BrainTorrent presents a highly dynamic peer-to-peer environment, where all centers directly interact with each other without depending on a central body. We demonstrate the overall effectiveness of FL for the challenging task of whole brain segmentation and observe that the proposed server-less BrainTorrent approach does not only outperform the traditional server-based one but reaches a similar performance to a model trained on pooled data.

The recent use of `Big Code' with state-of-the-art deep learning methods offers promising avenues to ease program source code writing and correction. As a first step towards automatic code repair, we implemented a graph neural network model that predicts token types for Javascript programs. The predictions achieve an accuracy above $90\%$, which improves on previous similar work.

We consider the problem of recovering a common latent source with independent components from multiple views. This applies to settings in which a variable is measured with multiple experimental modalities, and where the goal is to synthesize the disparate measurements into a single unified representation. We consider the case that the observed views are a nonlinear mixing of component-wise corruptions of the sources. When the views are considered separately, this reduces to nonlinear Independent Component Analysis (ICA) for which it is provably impossible to undo the mixing. We present novel identifiability proofs that this is possible when the multiple views are considered jointly, showing that the mixing can theoretically be undone using function approximators such as deep neural networks. In contrast to known identifiability results for nonlinear ICA, we prove that independent latent sources with arbitrary mixing can be recovered as long as multiple, sufficiently different noisy views are available.

This paper describes a language representation model which combines the Bidirectional Encoder Representations from Transformers (BERT) learning mechanism described in Devlin et al. (2018) with a generalization of the Universal Transformer model described in Dehghani et al. (2018). We further improve this model by adding a latent variable that represents the persona and topics of interests of the writer for each training example. We also describe a simple method to improve the usefulness of our language representation for solving problems in a specific domain at the expense of its ability to generalize to other fields. Finally, we release a pre-trained language representation model for social texts that was trained on 100 million tweets.

Data-driven modelling and synthesis of motion data is an active research area with applications that include animation and games. This paper introduces a new class of probabilistic, generative motion-data models based on normalising flows, specifically Glow. Models of this kind can describe highly complex distributions (unlike many classical approaches like GMMs) yet can be trained stably and efficiently using standard maximum likelihood (unlike GANs). Several model variants are described: unconditional fixed-length sequence models, conditional (i.e., controllable) fixed-length sequence models, and finally conditional, variable-length sequence models. The last type uses LSTMs to enable arbitrarily long time-dependencies and is, importantly, causal, meaning it only depends on control and pose information from current and previous timesteps. This makes it suitable for generating controllable motion in real-time applications. Every model type can in principle be applied to any motion since they do not make restrictive assumptions such as the motion being cyclic in nature. Experiments on a motion-capture dataset of human locomotion confirm that motion (sequences of 3D joint coordinates) sampled randomly from the new methods is judged as convincingly natural by human observers.

The recent success of Generative Adversarial Networks (GAN) is a result of their ability to generate high quality images from a latent vector space. An important application is the generation of images from a text description, where the text description is encoded and further used in the conditioning of the generated image. Thus the generative network has to additionally learn a mapping from the text latent vector space to a highly complex and multi-modal image data distribution, which makes the training of such models challenging. To handle the complexities of fashion image and meta data, we propose Ontology Generative Adversarial Networks (O-GANs) for fashion image synthesis that is conditioned on an hierarchical fashion ontology in order to improve the image generation fidelity. We show that the incorporation of the ontology leads to better image quality as measured by Fr\'{e}chet Inception Distance and Inception Score. Additionally, we show that the O-GAN achieves better conditioning results evaluated by implicit similarity between the text and the generated image.

We've discussed the idea of differentiable programming, where we incorporate existing programs into deep learning models. But if you're a researcher building, say, a self-driving car, what does differentiable programming mean in practice? How does it affect the way we express our problem, train our model, curate our dataset, and ultimately the results we achieve? This article shows what DP can bring to some simple but classic control problems, where we would normally use Reinforcement Learning (RL). DP-based models not only learn far more effective control strategies than RL, but also train orders of magnitude faster.