Pronounced variability due to the growth of renewable energy sources, flexible loads, and distributed generation is challenging residential distribution systems. This context, motivates well fast, efficient, and robust reactive power control. Real-time optimal reactive power control is possible in theory by solving a non-convex optimization problem based on the exact model of distribution flow. However, lack of high-precision instrumentation and reliable communications, as well as the heavy computational burden of non-convex optimization solvers render computing and implementing the optimal control challenging in practice. Taking a statistical learning viewpoint, the input-output relationship between each grid state and the corresponding optimal reactive power control is parameterized in the present work by a deep neural network, whose unknown weights are learned offline by minimizing the power loss over a number of historical and simulated training pairs. In the inference phase, one just feeds the real-time state vector into the learned neural network to obtain the `optimal' reactive power control with only several matrix-vector multiplications. The merits of this novel statistical learning approach are computational efficiency as well as robustness to random input perturbations. Numerical tests on a 47-bus distribution network using real data corroborate these practical merits.

We show that there is shared structure between different NLP datasets that HUBERT, but not BERT, is able to learn and leverage. Our experiment results show that untangling data-specific semantics from general language structure is key for better transfer among NLP tasks. Built on the Transformer architecture (V aswani et al., 2017), the BERT model (Devlin et al., 2018) has demonstrated great power for providing general-purpose vector embeddings of natural language: its representations have served as the basis of many successful deep Natural Language Processing (NLP) models on a variety of tasks (e.g., Liu et al., 2019a;b; Zhang et al., 2019). Recent studies (Coenen et al., 2019; Hewitt & Manning, 2019; Lin et al., 2019; Tenney et al., 2019) have shown that BERT representations carry considerable information about grammatical structure, which, by design, is a deep and general encapsulation of linguistic information. Symbolic computation over structured symbolic representations such as parse trees has long been used to formalize linguistic knowledge. To strengthen the generality of BERT's representations, we propose to import into its architecture this type of computation. Symbolic linguistic representations support the important distinction between content and form information. The form consists of a structure devoid of content, such as an unlabeled tree, a collection of nodes defined by their structural positions or roles (Newell, 1980), such as root, left-child-of-root, right-child-of-left-child-of root, etc. In a particular linguistic expression such as "Kim referred to herself during the speech", these purely-structural roles are filled with particular content-bearing symbols, including terminal words like Kim and non-terminal categories like NounPhrase . These role fillers have their own identities, which are preserved as they move from role to role across expressions: Kim retains its referent and its semantic properties whether it fills the subject or the object role in a sentence.

In this paper, we present Gamma-LSTM, an enhanced long short term memory (LSTM) unit, to enable learning of hierarchical representations through multiple stages of temporal abstractions. Gamma memory, a hierarchical memory unit, forms the central memory of Gamma-LSTM with gates to regulate the information flow into various levels of hierarchy, thus providing the unit with a control to pick the appropriate level of hierarchy to process the input at a given instant of time. We demonstrate better performance of Gamma-LSTM model regular and stacked LSTMs in two settings (pixel-by-pixel MNIST digit classification and natural language inference) placing emphasis on the ability to generalize over long sequences.

Work in progress generative models written as programs. Conditions on these random variables are imposed through observe statements, while the sample statements define latent variables we seek to draw inference about. Common to the different languages is the existence of an inference backend, which contains one or more general inference methods. Recent research has addressed the task of making repeated inference less computationally expensive, by using upfront computation to reduce the cost of later executions, an approach known as amortized inference (Gershman and Goodman, 2014). One new method called inference compilation (IC) (Le et al., 2017) enables fast inference on arbitrarily complex and non-differentiable generative models. The approximate posterior distribution it learns can be combined with importance sampling at inference time, so that inference is asymptotically correct.

We present relay policy learning, a method for imitation and reinforcement learning that can solve multi-stage, long-horizon robotic tasks. This general and universally-applicable, two-phase approach consists of an imitation learning stage that produces goal-conditioned hierarchical policies, and a reinforcement learning phase that finetunes these policies for task performance. Our method, while not necessarily perfect at imitation learning, is very amenable to further improvement via environment interaction, allowing it to scale to challenging long-horizon tasks. We simplify the long-horizon policy learning problem by using a novel data-relabeling algorithm for learning goal-conditioned hierarchical policies, where the low-level only acts for a fixed number of steps, regardless of the goal achieved. While we rely on demonstration data to bootstrap policy learning, we do not assume access to demonstrations of every specific tasks that is being solved, and instead leverage unstructured and unsegmented demonstrations of semantically meaningful behaviors that are not only less burdensome to provide, but also can greatly facilitate further improvement using reinforcement learning. We demonstrate the effectiveness of our method on a number of multi-stage, long-horizon manipulation tasks in a challenging kitchen simulation environment. Videos are available at https://relay-policy-learning.github.io/

We present a framework for automatically structuring and training fast, approximate, deep neural surrogates of existing stochastic simulators. Unlike traditional approaches to surrogate modeling, our surrogates retain the interpretable structure of the reference simulators. The particular way we achieve this allows us to replace the reference simulator with the surrogate when undertaking amortized inference in the probabilistic programming sense. The fidelity and speed of our surrogates allow for not only faster "forward" stochastic simulation but also for accurate and substantially faster inference. We support these claims via experiments that involve a commercial composite-materials curing simulator. Employing our surrogate modeling technique makes inference an order of magnitude faster, opening up the possibility of doing simulator-based, non-invasive, just-in-time parts quality testing; in this case inferring safety-critical latent internal temperature profiles of composite materials undergoing curing from surface temperature profile measurements.

Graph Attention Networks (GA Ts) are the state-of-the-art neural architecture for representation learning with graphs. GA Ts learn attention functions that assign weights to nodes so that different nodes have different influences in the feature aggregation steps. In practice, however, induced attention functions are prone to over-fitting due to increasing number of parameters and the lack of direct supervision on attention weights. GA Ts also suffer from over-smoothing at the decision boundary of nodes. Here we propose a framework to address their weaknesses via margin-based constraints on attention during training. We first theoretically demonstrate the over-smoothing behavior of GA Ts and then develop an approach using constraint on the attention weights according to the class boundary and feature aggregation pattern. Furthermore, to alleviate the over-fitting problem, we propose additional constraints on graph structure. Extensive experiments and ablation studies on common benchmark datasets demonstrate the effectiveness of our method, which leads to significant improvements over the previous state-of-the-art graph attention methods on all datasets. Introduction Many real world applications involve graph data, like social networks (Zhang and Chen 2018), chemical molecules (Gilmer et al. 2017), and recommender systems (Berg, Kipf, and Welling 2017). The complicated structures of these graphs have inspired new machine learning methods (Cai, Zheng, and Chang 2018; Wu et al. 2019b). Recently much attention and progress has been made on graph neural networks, which have been successfully applied to social network analysis (Battaglia et al. 2016), recommendation systems (Ying et al. 2018), and machine reading comprehension (Tu et al. 2019; De Cao, Aziz, and Titov 2018). Recently, a novel architecture leveraging attention mechanism in Graph Neural Networks (GNNs) called Graph Attention Networks (GA Ts) was introduced (V eli ˇ ckovi c et al. 2017). GA T was motivated by attention mechanism in natural language processing (V aswani et al. 2017; Devlin et al. 2018).

Training neural-networks is computationally hard. However, in practice they are trained efficiently using gradient-based algorithms, achieving remarkable performance on natural data. To bridge this gap, we observe the property of local correlation: correlation between small patterns of the input and the target label. We focus on learning deep neural-networks with a variant of gradient-descent, when the target function is a tree-structured Boolean circuit. We show that in this case, the existence of correlation between the gates of the circuit and the target label determines whether the optimization succeeds or fails. Using this result, we show that neural-networks can learn the (log n)-parity problem for most product distributions. These results hint that local correlation may play an important role in differentiating between distributions that are hard or easy to learn.

In this paper, we consider same-day delivery with a heterogeneous fleet of vehicles and drones. Customers make delivery requests over the course of the day and the dispatcher dynamically dispatches vehicles and drones to deliver the goods to customers before their delivery deadline. Vehicles can deliver multiple packages in one route but travel relatively slowly due to the urban traffic. Drones travel faster, but they have limited capacity and require charging or battery swaps. To exploit the different strengths of the fleets, we propose a deep Q-learning approach. Our method learns the value of assigning a new customer to either drones or vehicles as well as the option to not offer service at all. To aid feature selection, we present an analytical analysis that demonstrates the role that different types of information have on the value function and decision making. In a systematic computational analysis, we show the superiority of our policy compared to benchmark policies and the effectiveness of our deep Q-learning approach.

This paper provides a framework to analyze stochastic gradient algorithms in a mean squared error (MSE) sense using the asymptotic normality result of the stochastic gradient descent (SGD) iterates. We perform this analysis by taking the asymptotic normality result and applying it to the finite iteration case. Specifically, we look at problems where the gradient estimators are biased and have reduced variance and compare the iterates generated by these gradient estimators to the iterates generated by the SGD algorithm. We use the work of Fabian to characterize the mean and the variance of the distribution of the iterates in terms of the bias and the covariance matrix of the gradient estimators. We introduce the sliding window SGD (SW-SGD) algorithm, with its proof of convergence, which incurs a lower MSE than the SGD algorithm on quadratic and convex problems. Lastly, we present some numerical results to show the effectiveness of this framework and the superiority of SW-SGD algorithm over the SGD algorithm.