Existing on-policy imitation learning algorithms, such as DAgger, assume access to a fixed supervisor. However, there are many settings where the supervisor may converge during policy learning, such as a human performing a novel task or an improving algorithmic controller. We formalize imitation learning from a "converging supervisor" and provide sublinear static and dynamic regret guarantees against the best policy in hindsight with labels from the converged supervisor, even when labels during learning are only from intermediate supervisors. We then show that this framework is closely connected to a recent class of reinforcement learning (RL) algorithms known as dual policy iteration (DPI), which alternate between training a reactive learner with imitation learning and a model-based supervisor with data from the learner. Experiments suggest that when this framework is applied with the state-of-the-art deep model-based RL algorithm PETS as an improving supervisor, it outperforms deep RL baselines on continuous control tasks and provides up to an 80-fold speedup in policy evaluation.

In recent studies on model-based reinforcement learning (MBRL), incorporating uncertainty in forward dynamics is a state-of-the-art strategy to enhance learning performance, making MBRLs competitive to cutting-edge model free methods, especially in simulated robotics tasks. Probabilistic ensembles with trajectory sampling (PETS) is a leading type of MBRL, which employs Bayesian inference to dynamics modeling and model predictive control (MPC) with stochastic optimization via the cross entropy method (CEM). In this paper, we propose a novel extension to the uncertainty-aware MBRL. Our main contributions are twofold: Firstly, we introduce a variational inference MPC, which reformulates various stochastic methods, including CEM, in a Bayesian fashion. Secondly, we propose a novel instance of the framework, called probabilistic action ensembles with trajectory sampling (PaETS). As a result, our Bayesian MBRL can involve multimodal uncertainties both in dynamics and optimal trajectories. In comparison to PETS, our method consistently improves asymptotic performance on several challenging locomotion tasks.

Probabilistic programming languages (PPLs) are receiving widespread attention for performing Bayesian inference in complex generative models. However, applications to science remain limited because of the impracticability of rewriting complex scientific simulators in a PPL, the computational cost of inference, and the lack of scalable implementations. To address these, we present a novel PPL framework that couples directly to existing scientific simulators through a cross-platform probabilistic execution protocol and provides Markov chain Monte Carlo (MCMC) and deep-learning-based inference compilation (IC) engines for tractable inference. To guide IC inference, we perform distributed training of a dynamic 3DCNN--LSTM architecture with a PyTorch-MPI-based framework on 1,024 32-core CPU nodes of the Cori supercomputer with a global minibatch size of 128k: achieving a performance of 450 Tflop/s through enhancements to PyTorch. We demonstrate a Large Hadron Collider (LHC) use-case with the C++ Sherpa simulator and achieve the largest-scale posterior inference in a Turing-complete PPL.

In this work, we propose a (linearized) Alternating Direction Method-of-Multipliers (ADMM) algorithm for minimizing a convex function subject to a nonconvex constraint. We focus on the special case where such constraint arises from the specification that a variable should lie in the range of a neural network. This is motivated by recent successful applications of Generative Adversarial Networks (GANs) in tasks like compressive sensing, denoising and robustness against adversarial examples. The derived rates for our algorithm are characterized in terms of certain geometric properties of the generator network, which we show hold for feedforward architectures, under mild assumptions. Unlike gradient descent (GD), it can efficiently handle non-smooth objectives as well as exploit efficient partial minimization procedures, thus being faster in many practical scenarios.

The exclusive or (xor) function is one of the simplest examples that illustrate why nonlinear feedforward networks are superior to linear regression for machine learning applications. We review the xor representation and approximation problems and discuss their solutions in terms of probabilistic logic and associative copula functions. After briefly reviewing the specification of feedforward networks, we compare the dynamics of learned error surfaces with different activation functions such as RELU and tanh through a set of colorful three-dimensional charts. The copula representations extend xor from Boolean to real values, thereby providing a convenient way to demonstrate the concept of cross-validation on in-sample and out-sample data sets. Our approach is pedagogical and is meant to be a machine learning prolegomenon. Keywords: machine learning; neural networks; probabilistic logic; copulas; error surfaces; xor.

Today, large amounts of valuable data are distributed among millions of user-held devices, such as personal computers, phones, or Internet-of-things devices. Many companies collect such data with the goal of using it for training machine learning models allowing them to improve their services. However, user-held data is often sensitive, and collecting it is problematic in terms of privacy. We address this issue by proposing a novel way of training a supervised classifier in a distributed setting akin to the recently proposed federated learning paradigm (McMahan et al. 2017), but under the stricter privacy requirement that the server that trains the model is assumed to be untrusted and potentially malicious; we thus preserve user privacy by design, rather than by trust. In particular, our framework, called secret vector machine (SecVM), provides an algorithm for training linear support vector machines (SVM) in a setting in which data-holding clients communicate with an untrusted server by exchanging messages designed to not reveal any personally identifiable information. We evaluate our model in two ways. First, in an offline evaluation, we train SecVM to predict user gender from tweets, showing that we can preserve user privacy without sacrificing classification performance. Second, we implement SecVM's distributed framework for the Cliqz web browser and deploy it for predicting user gender in a large-scale online evaluation with thousands of clients, outperforming baselines by a large margin and thus showcasing that SecVM is practicable in production environments. Overall, this work demonstrates the feasibility of machine learning on data from thousands of users without collecting any personal data. We believe this is an innovative approach that will help reconcile machine learning with data privacy.

We propose two optimization techniques to minimize memory usage and computation while meeting system timing constraints for real-time classification in wearable systems. Our method derives a hierarchical classifier structure for Support Vector Machine (SVM) in order to reduce the amount of computations, based on the probability distribution of output classes occurrences. Also, we propose a memory optimization technique based on SVM parameters, which results in storing fewer support vectors and as a result requiring less memory. To demonstrate the efficiency of our proposed techniques, we performed an activity recognition experiment and were able to save up to 35% and 56% in memory storage when classifying 14 and 6 different activities, respectively. In addition, we demonstrated that there is a trade-off between accuracy of classification and memory savings, which can be controlled based on application requirements.

Canonical correlation analysis (CCA) is a technique to find statistical dependencies between a pair of multivariate data. However, its application to high dimensional data is limited due to the resulting time complexity. While the conventional CCA algorithm requires polynomial time, we have developed an algorithm that approximates CCA with computational time proportional to the logarithm of the input dimensionality using quantum-inspired computation. The computational efficiency and approximation performance of the proposed quantum-inspired CCA (qiCCA) algorithm are experimentally demonstrated. Furthermore, the fast computation of qiCCA allows us to directly apply CCA even after nonlinearly mapping raw input data into very high dimensional spaces. Experiments performed using a benchmark dataset demonstrated that, by mapping the raw input data into the high dimensional spaces with second-order monomials, the proposed qiCCA extracted more correlations than linear CCA and was comparable to deep CCA and kernel CCA. These results suggest that qiCCA is considerably useful and quantum-inspired computation has the potential to unlock a new field in which exponentially large dimensional data can be analyzed.

Materials discovery is crucial for making scientific advances in many domains. Collections of data from experiments and first-principle computations have spurred interest in applying machine learning methods to create predictive models capable of mapping from composition and crystal structures to materials properties. Generally, these are regression problems with the input being a 1D vector composed of numerical attributes representing the material composition and/or crystal structure. While neural networks consisting of fully connected layers have been applied to such problems, their performance often suffers from the vanishing gradient problem when network depth is increased. In this paper, we study and propose design principles for building deep regression networks composed of fully connected layers with numerical vectors as input. We introduce a novel deep regression network with individual residual learning, IRNet, that places shortcut connections after each layer so that each layer learns the residual mapping between its output and input. We use the problem of learning properties of inorganic materials from numerical attributes derived from material composition and/or crystal structure to compare IRNet's performance against that of other machine learning techniques. Using multiple datasets from the Open Quantum Materials Database (OQMD) and Materials Project for training and evaluation, we show that IRNet provides significantly better prediction performance than the state-of-the-art machine learning approaches currently used by domain scientists. We also show that IRNet's use of individual residual learning leads to better convergence during the training phase than when shortcut connections are between multi-layer stacks while maintaining the same number of parameters.

In this paper we present a novel approach for extracting a Bag-of-Words (BoW) representation based on a Neural Network codebook. The conventional BoW model is based on a dictionary (codebook) built from elementary representations which are selected randomly or by using a clustering algorithm on a training dataset. A metric is then used to assign unseen elementary representations to the closest dictionary entries in order to produce a histogram. In the proposed approach, an autoencoder (AE) encompasses the role of both the dictionary creation and the assignment metric. The dimension of the encoded layer of the AE corresponds to the size of the dictionary and the output of its neurons represents the assignment metric. Experimental results for the continuous emotion prediction task on the AVEC 2017 audio dataset have shown an improvement of the Concordance Correlation Coefficient (CCC) from 0.225 to 0.322 for arousal dimension and from 0.244 to 0.368 for valence dimension relative to the conventional BoW version implemented in a baseline system.