Urban dispersal events are processes where an unusually large number of people leave the same area in a short period. Early prediction of dispersal events is important in mitigating congestion and safety risks and making better dispatching decisions for taxi and ride-sharing fleets. Existing work mostly focuses on predicting taxi demand in the near future by learning patterns from historical data. However, they fail in case of abnormality because dispersal events with abnormally high demand are non-repetitive and violate common assumptions such as smoothness in demand change over time. Instead, in this paper we argue that dispersal events follow a complex pattern of trips and other related features in the past, which can be used to predict such events. Therefore, we formulate the dispersal event prediction problem as a survival analysis problem. We propose a two-stage framework (DILSA), where a deep learning model combined with survival analysis is developed to predict the probability of a dispersal event and its demand volume. We conduct extensive case studies and experiments on the NYC Yellow taxi dataset from 2014-2016. Results show that DILSA can predict events in the next 5 hours with F1-score of 0.7 and with average time error of 18 minutes. It is orders of magnitude better than the state-ofthe-art deep learning approaches for taxi demand prediction.

Meta-learning is a tool that allows us to build sample-efficient learning systems. Here we show that, once meta-trained, LSTM Meta-Learners aren't just faster learners than their sample-inefficient deep learning (DL) and reinforcement learning (RL) brethren, but that they actually pursue fundamentally different learning trajectories. We study their learning dynamics on three sets of structured tasks for which the corresponding learning dynamics of DL and RL systems have been previously described: linear regression (Saxe et al., 2013), nonlinear regression (Rahaman et al., 2018; Xu et al., 2018), and contextual bandits (Schaul et al., 2019). In each case, while sample-inefficient DL and RL Learners uncover the task structure in a staggered manner, meta-trained LSTM Meta-Learners uncover almost all task structure concurrently, congruent with the patterns expected from Bayes-optimal inference algorithms. This has implications for research areas wherever the learning behaviour itself is of interest, such as safety, curriculum design, and human-in-the-loop machine learning.

Most of the semantic segmentation approaches have been developed for single image segmentation, and hence, video sequences are currently segmented by processing each frame of the video sequence separately. The disadvantage of this is that temporal image information is not considered, which improves the performance of the segmentation approach. One possibility to include temporal information is to use recurrent neural networks. However, there are only a few approaches using recurrent networks for video segmentation so far. These approaches extend the encoder-decoder network architecture of well-known segmentation approaches and place convolutional LSTM layers between encoder and decoder. However, in this paper it is shown that this position is not optimal, and that other positions in the network exhibit better performance. Nowadays, state-of-the-art segmentation approaches rarely use the classical encoder-decoder structure, but use multi-branch architectures. These architectures are more complex, and hence, it is more difficult to place the recurrent units at a proper position. In this work, the multi-branch architectures are extended by convolutional LSTM layers at different positions and evaluated on two different datasets in order to find the best one. It turned out that the proposed approach outperforms the pure CNN-based approach for up to 1.6 percent.

A visual-relational knowledge graph (KG) is a multi-relational graph whose entities are associated with images. We explore novel machine learning approaches for answering visual-relational queries in web-extracted knowledge graphs. To this end, we have created ImageGraph, a KG with 1,330 relation types, 14,870 entities, and 829,931 images crawled from the web. With visual-relational KGs such as ImageGraph one can introduce novel probabilistic query types in which images are treated as first-class citizens. Both the prediction of relations between unseen images as well as multi-relational image retrieval can be expressed with specific families of visual-relational queries. We introduce novel combinations of convolutional networks and knowledge graph embedding methods to answer such queries. We also explore a zero-shot learning scenario where an image of an entirely new entity is linked with multiple relations to entities of an existing KG. The resulting multi-relational grounding of unseen entity images into a knowledge graph serves as a semantic entity representation. We conduct experiments to demonstrate that the proposed methods can answer these visual-relational queries efficiently and accurately.

Existing deep architectures cannot operate on very large signals such as megapixel images due to computational and memory constraints. To tackle this limitation, we propose a fully differentiable end-to-end trainable model that samples and processes only a fraction of the full resolution input image. The locations to process are sampled from an attention distribution computed from a low resolution view of the input. We refer to our method as attention sampling and it can process images of several megapixels with a standard single GPU setup. We show that sampling from the attention distribution results in an unbiased estimator of the full model with minimal variance, and we derive an unbiased estimator of the gradient that we use to train our model end-to-end with a normal SGD procedure. This new method is evaluated on three classification tasks, where we show that it allows to reduce computation and memory footprint by an order of magnitude for the same accuracy as classical architectures. We also show the consistency of the sampling that indeed focuses on informative parts of the input images.

We propose a completely unsupervised method to understand audio scenes observed with random microphone arrangements by decomposing the scene into its constituent sources and their relative presence in each microphone. To this end, we formulate a neural network architecture that can be interpreted as a nonnegative tensor factorization of a multi-channel audio recording. By clustering on the learned network parameters corresponding to channel content, we can learn sources' individual spectral dictionaries and their activation patterns over time. Our method allows us to leverage deep learning advances like end-to-end training, while also allowing stochastic minibatch training so that we can feasibly decompose realistic audio scenes that are intractable to decompose using standard methods. This neural network architecture is easily extensible to other kinds of tensor factorizations.

Seizure detection from EEGs is a challenging and time consuming clinical problem that would benefit from the development of automated algorithms. EEGs can be viewed as structural time series, because they are multivariate time series where the placement of leads on a patient's scalp provides prior information about the structure of interactions. Commonly used deep learning models for time series don't offer a way to leverage structural information, but this would be desirable in a model for structural time series. To address this challenge, we propose the temporal graph convolutional network (TGCN), a model that leverages structural information and has relatively few parameters. TGCNs apply feature extraction operations that are localized and shared over both time and space, thereby providing a useful inductive bias in tasks where one expects similar features to be discriminative across the different sequences. In our experiments we focus on metrics that are most important to seizure detection, and demonstrate that TGCN matches the performance of related models that have been shown to be state-of-the-art in other tasks. Additionally, we investigate interpretability advantages of TGCN by exploring approaches for helping clinicians determine when precisely seizures occur, and the parts of the brain that are most involved.

Recent seminal work at the intersection of deep neural networks practice and random matrix theory has linked the convergence speed and robustness of these networks with the combination of random weight initialization and nonlinear activation function in use. Building on those principles, we introduce a process to transform an existing activation function into another one with better properties. We term such transform \emph{static activation normalization}. More specifically we focus on this normalization applied to the ReLU unit, and show empirically that it significantly promotes convergence robustness, maximum training depth, and anytime performance. We verify these claims by examining empirical eigenvalue distributions of networks trained with those activations. Our static activation normalization provides a first step towards giving benefits similar in spirit to schemes like batch normalization, but without computational cost.

Abstract-- Detecting the intention of drivers is an essential task in self-driving, necessary to anticipate sudden events like lane changes and stops. Turn signals and emergency flashers communicate such intentions, providing seconds of potentially critical reaction time. In this paper, we propose to detect these signals in video sequences by using a deep neural network that reasons about both spatial and temporal information. Our experiments on more than a million frames show high per-frame accuracy in very challenging scenarios. I. INTRODUCTION Autonomous driving has risen as one of the most impactful applications of Artificial Intelligence (AI), where it has the potential to change the way we live.

The inner product operation between tensors is the corner stone of deep neural network architectures, directly inherited from linear algebra. There is a striking contrast between the unicity of this basic construct and the extreme diversity of high level constructs which have been invented to address various application domains. This paper is interested in an intermediate construct, convolution, and its corollary, attention, which have become ubiquitous in many applications, but are still presented in an ad-hoc fashion depending on the application context. We first identify the common problem addressed by most existing forms of convolution, and show how the solution to that problem naturally involves another very generic operation of linear algebra, the outer product between tensors. We then proceed to show that attention is a form of convolution, called "content based" convolution, hence amenable to the generic formulation based on the outer product. The reader looking for yet another architecture yielding better performance results on a specific task is in for some disappointment. The reader aiming at a better, more grounded understanding of familiar concepts may find food for thought.