We study distributed stochastic gradient (D-SG) method and its accelerated variant (D-ASG) for solving decentralized strongly convex stochastic optimization problems where the objective function is distributed over several computational units, lying on a fixed but arbitrary connected communication graph, subject to local communication constraints where noisy estimates of the gradients are available. We develop a framework which allows to choose the stepsize and the momentum parameters of these algorithms in a way to optimize performance by systematically trading off the bias, variance, robustness to gradient noise and dependence to network effects. When gradients do not contain noise, we also prove that distributed accelerated methods can \emph{achieve acceleration}, requiring $\mathcal{O}(\kappa \log(1/\varepsilon))$ gradient evaluations and $\mathcal{O}(\kappa \log(1/\varepsilon))$ communications to converge to the same fixed point with the non-accelerated variant where $\kappa$ is the condition number and $\varepsilon$ is the target accuracy. To our knowledge, this is the first acceleration result where the iteration complexity scales with the square root of the condition number in the context of \emph{primal} distributed inexact first-order methods. For quadratic functions, we also provide finer performance bounds that are tight with respect to bias and variance terms. Finally, we study a multistage version of D-ASG with parameters carefully varied over stages to ensure exact $\mathcal{O}(-k/\sqrt{\kappa})$ linear decay in the bias term as well as optimal $\mathcal{O}(\sigma^2/k)$ in the variance term. We illustrate through numerical experiments that our approach results in practical algorithms that are robust to gradient noise and that can outperform existing methods.

The fields of artificial intelligence and neuroscience have a long history of fertile bi-directional interactions. On the one hand, important inspiration for the development of artificial intelligence systems has come from the study of natural systems of intelligence, the mammalian neocortex in particular. On the other, important inspiration for models and theories of the brain have emerged from artificial intelligence research. A central question at the intersection of these two areas is concerned with the processes by which neocortex learns, and the extent to which they are analogous to the back-propagation training algorithm of deep networks. Matching the data efficiency, transfer and generalization properties of neocortical learning remains an area of active research in the field of deep learning. Recent advances in our understanding of neuronal, synaptic and dendritic physiology of the neocortex suggest new approaches for unsupervised representation learning, perhaps through a new class of objective functions, which could act alongside or in lieu of back-propagation. Such local learning rules have implicit rather than explicit objectives with respect to the training data, facilitating domain adaptation and generalization. Incorporating them into deep networks for representation learning could better leverage unlabelled datasets to offer significant improvements in data efficiency of downstream supervised readout learning, and reduce susceptibility to adversarial perturbations, at the cost of a more restricted domain of applicability.

Generative seq2seq dialogue systems are trained to predict the next word in dialogues that have already occurred. They can learn from large unlabeled conversation datasets, build a deeper understanding of conversational context, and generate a wide variety of responses. This flexibility comes at the cost of control, a concerning tradeoff in doctor/patient interactions. Inaccuracies, typos, or undesirable content in the training data will be reproduced by the model at inference time. We trade a small amount of labeling effort and some loss of response variety in exchange for quality control. More specifically, a pretrained language model encodes the conversational context, and we finetune a classification head to map an encoded conversational context to a response class, where each class is a noisily labeled group of interchangeable responses. Experts can update these exemplar responses over time as best practices change without retraining the classifier or invalidating old training data. Expert evaluation of 775 unseen doctor/patient conversations shows that only 12% of the discriminative model's responses are worse than the what the doctor ended up writing, compared to 18% for the generative model.

We optimize a six degrees of freedom hovering policy using reinforcement meta-learning. The policy maps flash LIDAR measurements directly to on/off spacecraft body-frame thrust commands, allowing hovering at a fixed position and attitude in the asteroid body-fixed reference frame. Importantly, the policy does not require position and velocity estimates, and can operate in environments with unknown dynamics, and without an asteroid shape model or navigation aids. Indeed, during optimization the agent is confronted with a new randomly generated asteroid for each episode, insuring that it does not learn an asteroid's shape, texture, or environmental dynamics. This allows the deployed policy to generalize well to novel asteroid characteristics, which we demonstrate in our experiments. The hovering controller has the potential to simplify mission planning by allowing asteroid body-fixed hovering immediately upon the spacecraft's arrival to an asteroid. This in turn simplifies shape model generation and allows resource mapping via remote sensing immediately upon arrival at the target asteroid.

Due to the increasing use of machine learning in practice it becomes more and more important to be able to explain the pred iction and behavior of machine learning models. An instance of expl anations are counterfactual explanations which provide an intuitive an d useful explanations of machine learning models. In this survey we review model-specific methods for efficientl y computing counterfactual explanations of many different machine learning models and propose methods for models that have not been considered in l iterature so far.

e-Yantra Robotics Competition (eYRC) is a unique Robotics Competition hosted by IIT Bombay that is actually an Embedded Systems and Robotics MOOC. Registrations have been growing exponentially in each year from 4500 in 2012 to over 34000 in 2019. In this 5-month long competition students learn complex skills under severe time pressure and have access to a discussion forum to post doubts about the learning material. Responding to questions in real-time is a challenge for project staff. Here, we illustrate the advantage of Deep Learning for real-time question answering in the eYRC discussion forum. We illustrate the advantage of Transformer based contextual embedding mechanisms such as Bidirectional Encoder Representation From Transformer (BERT) over word embedding mechanisms such as Word2Vec. We propose a weighted similarity metric as a measure of matching and find it more reliable than Content-Content or Title-Title similarities alone. The automation of replying to questions has brought the turn around response time(TART) down from a minimum of 21 mins to a minimum of 0.3 secs.

We consider the problem of recovering an expert's reward function with inverse reinforcement learning (IRL) when there are missing/incomplete state-action pairs or observations in the demonstrated trajectories. This issue of missing trajectory data or information occurs in many situations, e.g., GPS signals from vehicles moving on a road network are intermittent. In this paper, we propose a tractable approach to directly compute the log-likelihood of demonstrated trajectories with incomplete/missing data. Our algorithm is efficient in handling a large number of missing segments in the demonstrated trajectories, as it performs the training with incomplete data by solving a sequence of systems of linear equations, and the number of such systems to be solved does not depend on the number of missing segments. Empirical evaluation on a real-world dataset shows that our training algorithm outperforms other conventional techniques.

Improving Graph Neural Network Representations of Logical Formulae with Subgraph Pooling Maxwell Crouse * §, Ibrahim Abdelaziz †, Cristina Cornelio †, V eronika Thost ‡, Lingfei Wu †, Kenneth Forbus §, Achille Fokoue † § Northwestern University, † IBM Research, ‡ MIT -IBM Watson AI Lab Abstract --Recent advances in the integration of deep learning with automated theorem proving have centered around the representation of logical formulae as inputs to deep learning systems. In particular, there has been a shift from character and token-level representations to graph-structured representations, in large part driven by the rapidly emerging body of research on geometric deep learning. Typically, structure-aware neural methods for embedding logical formulae have been variants of either Tree LSTMs or GNNs. While more effective than character and token-level approaches, such methods have often made representational tradeoffs that limited their ability to effectively represent the global structure of their inputs. In this work, we introduce a novel approach for embedding logical formulae using DAG LSTMs that is designed to overcome the limitations of both Tree LSTMs and GNNs. The effectiveness of the proposed framework is demonstrated on the tasks of premise selection and proof step classification where it achieves the state-of-the-art performance on two standard datasets. I NTRODUCTION While state-of-the-art classical theorem provers excel at finding complex proofs in restricted domains (e.g. TPTP [1]), they have historically had difficulty when reasoning in broad contexts [2], [3]. With the generation of large logical theories (collections of axioms) for reasoning [4]-[6], there has been interest in extending traditional theorem provers to handle the computational challenges inherent to reasoning at scale.

Off-policy policy evaluation (OPE) is the problem of estimating the online performance of a policy using only pre-collected historical data generated by another policy. Given the increasing interest in deploying learning-based methods for safety-critical applications, many recent OPE methods have recently been proposed. Due to disparate experimental conditions from recent literature, the relative performance of current OPE methods is not well understood. In this work, we present the first comprehensive empirical analysis of a broad suite of OPE methods. Based on thousands of experiments and detailed empirical analyses, we offer a summarized set of guidelines for effectively using OPE in practice, and suggest directions for future research.

Humans and animals are capable of quickly learning new behaviours to solve new tasks. Yet, we often forget that they also rely on a highly specialized morphology that co-adapted with motor control throughout thousands of years. Although compelling, the idea of co-adapting morphology and behaviours in robots is often unfeasible because of the long manufacturing times, and the need to re-design an appropriate controller for each morphology. In this paper, we propose a novel approach to automatically and efficiently co-adapt a robot morphology and its controller. Our approach is based on recent advances in deep reinforcement learning, and specifically the soft actor critic algorithm. Key to our approach is the possibility of leveraging previously tested morphologies and behaviors to estimate the performance of new candidate morphologies. As such, we can make full use of the information available for making more informed decisions, with the ultimate goal of achieving a more data-efficient co-adaptation (i.e., reducing the number of morphologies and behaviors tested). Simulated experiments show that our approach requires drastically less design prototypes to find good morphology-behaviour combinations, making this method particularly suitable for future co-adaptation of robot designs in the real world.