Learning representations that accurately capture long-range dependencies in sequential inputs --- including text, audio, and genomic data --- is a key problem in deep learning. Feed-forward convolutional models capture only feature interactions within finite receptive fields while recurrent architectures can be slow and difficult to train due to vanishing gradients. Here, we propose Temporal Feature-Wise Linear Modulation (TFiLM) --- a novel architectural component inspired by adaptive batch normalization and its extensions --- that uses a recurrent neural network to alter the activations of a convolutional model. This approach expands the receptive field of convolutional sequence models with minimal computational overhead. Empirically, we find that TFiLM significantly improves the learning speed and accuracy of feed-forward neural networks on a range of generative and discriminative learning tasks, including text classification and audio super-resolution

The success of Deep Learning and its potential use in many safety-critical applications has motivated research on formal verification of Neural Network (NN) models. In this context, verification means verifying whether a NN model satisfies certain input-output properties. Despite the reputation of learned NN models as black boxes, and the theoretical hardness of proving useful properties about them, researchers have been successful in verifying some classes of models by exploiting their piecewise linear structure and taking insights from formal methods such as Satisifiability Modulo Theory. However, these methods are still far from scaling to realistic neural networks. To facilitate progress on this crucial area, we make two key contributions. First, we present a unified framework based on branch and bound that encompasses previous methods. This analysis results in the identification of new methods that combine the strengths of multiple existing approaches, accomplishing a speedup of two orders of magnitude compared to the previous state of the art. Second, we propose a new data set of benchmarks which includes a collection of previously released test cases. We use the benchmark to provide a thorough experimental comparison of existing algorithms and identify the factors impacting the hardness of verification problems.

The constant introduction of standardized benchmarks in the literature has helped accelerating the recent advances in meta-learning research. They offer a way to get a fair comparison between different algorithms, and the wide range of datasets available allows full control over the complexity of this evaluation. However, for a large majority of code available online, the data pipeline is often specific to one dataset, and testing on another dataset requires significant rework. We introduce Torchmeta, a library built on top of PyTorch that enables seamless and consistent evaluation of meta-learning algorithms on multiple datasets, by providing data-loaders for most of the standard benchmarks in few-shot classification and regression, with a new meta-dataset abstraction. It also features some extensions for PyTorch to simplify the development of models compatible with meta-learning algorithms. The code is available here: https://github.com/tristandeleu/pytorch-meta

Real-world graph applications, such as advertisements and product recommendations make profits based on accurately classify the label of the nodes. However, in such scenarios, there are high incentives for the adversaries to attack such graph to reduce the node classification performance. Previous work on graph adversarial attacks focus on modifying existing graph structures, which is infeasible in most real-world applications. In contrast, it is more practical to inject adversarial nodes into existing graphs, which can also potentially reduce the performance of the classifier. In this paper, we study the novel node injection poisoning attacks problem which aims to poison the graph. We describe a reinforcement learning based method, namely NIPA, to sequentially modify the adversarial information of the injected nodes. We report the results of experiments using several benchmark data sets that show the superior performance of the proposed method NIPA, relative to the existing state-of-the-art methods.

Gaussian process classification (GPC) provides a flexible and powerful statistical framework describing joint distributions over function space. Conventional GPCs however suffer from (i) poor scalability for big data due to the full kernel matrix, and (ii) intractable inference due to the non-Gaussian likelihoods. Hence, various scalable GPCs have been proposed through (i) the sparse approximation built upon a small inducing set to reduce the time complexity; and (ii) the approximate inference to derive analytical evidence lower bound (ELBO). However, these scalable GPCs equipped with analytical ELBO are limited to specific likelihoods or additional assumptions. In this work, we present a unifying framework which accommodates scalable GPCs using various likelihoods. Analogous to GP regression (GPR), we introduce additive noises to augment the probability space for (i) the GPCs with step, (multinomial) probit and logit likelihoods via the internal variables; and particularly, (ii) the GPC using softmax likelihood via the noise variables themselves. This leads to unified scalable GPCs with analytical ELBO by using variational inference. Empirically, our GPCs showcase better results than state-of-the-art scalable GPCs for extensive binary/multi-class classification tasks with up to two million data points.

The study of the LSE has received considerable attention in statistics as well as machine learning; see [9, 36, 41, 54, 55] for important contributions. LSEs are particularly useful when F is known to satisfy some shape constraints such as monotonicity, convexity, or unimodality. In such cases, the LSEs are tuning parameter-free, can be computed as the solution to convex optimization problems, and are adaptive, i.e., the rate of convergence of the LSE changes depending on the "structure" of f 0 [4, 23, 28, 43, 47]. For example, if F is the class of monotone functions and f 0 is a strictly increasing function then ˆ f converges at an n 1/ 3 rate (i.e., null ˆ f f 0null O p(n 1/ 3)); however, if f 0 0, then ˆ f converges at an n 1/ 2 rate [10, 25, 56]. Tsirelson [50] and van de Geer and Wegkamp [53] have established necessary and sufficient conditions on F and null for consistency of ˆ f . Our goal in this work is to provide some general sufficient conditions on F and null under which the LSE is "rate-optimal"; see discussions after Corollary 2.1 for more details, also see Example 3.1. Our results significantly expand the scenarios under which LSE can be proven to be rate-optimal or "be a safe choice" for the model at hand. Theorem 3.2.5 of [55] implies that the LSE ˆ f defined on F satisfies null ˆ f f 0null O p(r n) for any r n such that E null sup f F: null f f 0null r nnull null G n[2null ( f f 0)( X) (f f 0) 2 (X)]null null null C nr 2 n, (3) where C denotes a constant 1 . In the rest of this paper, we make the convention that the constant C is not necessarily the same on each occurrence.

We propose a novel learning framework to answer questions such as "if a user is purchasing a shirt, what other items will (s)he need with the shirt?" Our framework learns distributed representations for items from available textual data, with the learned representations representing items in a latent space expressing functional complementarity as well similarity. In particular, our framework places functionally similar items close together in the latent space, while also placing complementary items closer than non-complementary items, but farther away than similar items. In this study, we introduce a new dataset of similar, complementary, and negative items derived from the Amazon co-purchase dataset. For evaluation purposes, we focus our approach on clothing and fashion verticals. As per our knowledge, this is the first attempt to learn similar and complementary relationships simultaneously through just textual title metadata. Our framework is applicable across a broad set of items in the product catalog and can generate quality complementary item recommendations at scale.

In this paper, we introduce an extension of our presented cognitive-based emotion model [27][28]and [30], where we enhance our knowledge-based emotion unit of the architecture by embedding a fuzzy rule-based system to it. The model utilizes the cognitive parameters dependency and their corresponding weights to regulate the robot's behavior and fuse their behavior data to achieve the final decision in their interaction with the environment. Using this fuzzy system, our previous model can simulate linguistic parameters for better controlling and generating understandable and flexible behaviors in the robots. We implement our model on an assistive healthcare robot, named Robot Nurse Assistant (RNA) and test it with human subjects. Our model records all the emotion states and essential information based on its predefined rules and learning system. Our results show that our robot interacts with patients in a reasonable, faithful way in special conditions which are defined by rules. This work has the potential to provide better on-demand service for clinical experts to monitor the patients' emotion states and help them make better decisions accordingly.

A consequence of the fragmented and siloed healthcare landscape is that patient care (and data) is split along multitude of different facilities and computer systems and enabling interoperability between these systems is hard. The lack interoperability not only hinders continuity of care and burdens providers, but also hinders effective application of Machine Learning (ML) algorithms. Thus, most current ML algorithms, designed to understand patient care and facilitate clinical decision-support, are trained on limited datasets. This approach is analogous to the Newtonian paradigm of Reductionism in which a system is broken down into elementary components and a description of the whole is formed by understanding those components individually. A key limitation of the reductionist approach is that it ignores the component-component interactions and dynamics within the system which are often of prime significance in understanding the overall behaviour of complex adaptive systems (CAS). Healthcare is a CAS. Though the application of ML on health data have shown incremental improvements for clinical decision support, ML has a much a broader potential to restructure care delivery as a whole and maximize care value. However, this ML potential remains largely untapped: primarily due to functional limitations of Electronic Health Records (EHR) and the inability to see the healthcare system as a whole. This viewpoint (i) articulates the healthcare as a complex system which has a biological and an organizational perspective, (ii) motivates with examples, the need of a system's approach when addressing healthcare challenges via ML and, (iii) emphasizes to unleash EHR functionality - while duly respecting all ethical and legal concerns - to reap full benefits of ML.

Traditional planning and control methods could fail to find a feasible trajectory for an autonomous vehicle to execute amongst dense traffic on roads. This is because the obstacle-free volume in spacetime is very small in these scenarios for the vehicle to drive through. However, that does not mean the task is infeasible since human drivers are known to be able to drive amongst dense traffic by leveraging the cooperativeness of other drivers to open a gap. The traditional methods fail to take into account the fact that the actions taken by an agent affect the behaviour of other vehicles on the road. In this work, we rely on the ability of deep reinforcement learning to implicitly model such interactions and learn a continuous control policy over the action space of an autonomous vehicle. The application we consider requires our agent to negotiate and open a gap in the road in order to successfully merge or change lanes. Our policy learns to repeatedly probe into the target road lane while trying to find a safe spot to move in to. We compare against two model-predictive control-based algorithms and show that our policy outperforms them in simulation.