ICML 2018 Papers

Many advanced Learning from Demonstration (LfD) methods consider the decomposition of complex, real-world tasks into simpler sub-tasks. By reusing the corresponding sub-policies within and between tasks, we can provide training data for each policy from different high-level tasks and compose them to perform novel ones. Existing approaches to modular LfD focus either on learning a single high-level task or depend on domain knowledge and temporal segmentation. In contrast, we propose a weakly supervised, domain-agnostic approach based on task sketches, which include only the sequence of sub-tasks performed in each demonstration. Our approach simultaneously aligns the sketches with the observed demonstrations and learns the required sub-policies. This improves generalisation in comparison to separate optimisation procedures. We evaluate the approach on multiple domains, including a simulated 3D robot arm control task using purely image-based observations. The results show that our approach performs commensurately with fully supervised approaches, while requiring significantly less annotation effort.

Standard methods for generating adversarial examples for neural networks do not consistently fool neural network classifiers in the physical world due to a combination of viewpoint shifts, camera noise, and other natural transformations, limiting their relevance to real-world systems. We demonstrate the existence of robust 3D adversarial objects, and we present the first algorithm for synthesizing examples that are adversarial over a chosen distribution of transformations. We synthesize two-dimensional adversarial images that are robust to noise, distortion, and affine transformation. We apply our algorithm to complex three-dimensional objects, using 3D-printing to manufacture the first physical adversarial objects. Our results demonstrate the existence of 3D adversarial objects in the physical world.

Advances in deep generative networks have led to impressive results in recent years. Nevertheless, such models can often waste their capacity on the minutiae of datasets, presumably due to weak inductive biases in their decoders. This is where graphics engines may come in handy since they abstract away low-level details and represent images as high-level programs. Current methods that combine deep learning and renderers are limited by hand-crafted likelihood or distance functions, a need for large amounts of supervision, or difficulties in scaling their inference algorithms to richer datasets. To mitigate these issues, we present SPIRAL, an adversarially trained agent that generates a program which is executed by a graphics engine to interpret and sample images. The goal of this agent is to fool a discriminator network that distinguishes between real and rendered data, trained with a distributed reinforcement learning setup without any supervision. A surprising finding is that using the discriminator’s output as a reward signal is the key to allow the agent to make meaningful progress at matching the desired output rendering. To the best of our knowledge, this is the first demonstration of an end-to-end, unsupervised and adversarial inverse graphics agent on challenging real world (MNIST, Omniglot, CelebA) and synthetic 3D datasets. A video of the agent can be found at https://youtu.be/iSyvwAwa7vk.

We consider a generalization of the classic linear regression problem to the case when the loss is an Orlicz norm. An Orlicz norm is parameterized by a non-negative convex function G: R_+ - > R_+ with G(0) = 0: the Orlicz norm of a n-dimensional vector x is defined as |x|_G = inf{ alpha > 0 | sum_{i = 1}^n G( |x_i| / alpha ) < = 1 }. We consider the cases where the function G grows subquadratically. Our main result is based on a new oblivious embedding which embeds the column space of a given nxd matrix A with Orlicz norm into a lower dimensional space with L2 norm. Specifically, we show how to efficiently find an mxn embedding matrix S (m < n), such that for every d-dimensional vector x, we have Omega(1/(d log n)) |Ax|_G < = |SAx|_2 < = O(d^2 log n) |Ax|_G. By applying this subspace embedding technique, we show an approximation algorithm for the regression problem min_x |Ax-b|_G, up to a O( d log^2 n ) factor. As a further application of our techniques, we show how to also use them to improve on the algorithm for the Lp low rank matrix approximation problem for 1 < = p < 2.

We introduce submodular hypergraphs, a family of hypergraphs that have different submodular weights associated with different cuts of hyperedges. Submodular hypergraphs arise in cluster- ing applications in which higher-order structures carry relevant information. For such hypergraphs, we define the notion of p-Laplacians and derive corresponding nodal domain theorems and k-way Cheeger inequalities. We conclude with the description of algorithms for computing the spectra of 1- and 2-Laplacians that constitute the basis of new spectral hypergraph clustering methods.

In this work, we propose “global style tokens” (GSTs), a bank of embeddings that are jointly trained within Tacotron, a state-of-the-art end-to-end speech synthesis system. The embeddings are trained with no explicit labels, yet learn to model a large range of acoustic expressiveness. GSTs lead to a rich set of significant results. The soft interpretable “labels” they generate can be used to control synthesis in novel ways, such as varying speed and speaking style – independently of the text content. They can also be used for style transfer, replicating the speaking style of a single audio clip across an entire long-form text corpus. When trained on noisy, unlabeled found data, GSTs learn to factorize noise and speaker identity, providing a path towards highly scalable but robust speech synthesis.

We tackle the problem of optimizing a black-box objective function defined over a highly-structured input space. This problem is ubiquitous in science and engineering. In machine learning, inferring the structure of a neural network or the Automatic Statistician (AS), where the optimal kernel combination for a Gaussian process is selected, are two important examples. We use the \as as a case study to describe our approach, that can be easily generalized to other domains. We propose an Structure Generating Variational Auto-encoder (SG-VAE) to embed the original space of kernel combinations into some low-dimensional continuous manifold where Bayesian optimization (BO) ideas are used. This is possible when structural knowledge of the problem is available, which can be given via a simulator or any other form of generating potentially good solutions. The right exploration-exploitation balance is imposed by propagating into the search the uncertainty of the latent space of the SG-VAE, that is computed using variational inference. The key aspect of our approach is that the SG-VAE can be used to bias the search towards relevant regions, making it suitable for transfer learning tasks. Several experiments in various application domains are used to illustrate the utility and generality of the approach described in this work.

Bayesian Neural Networks (BNNs) have recently received increasing attention for their ability to provide well-calibrated posterior uncertainties. However, model selection—even choosing the number of nodes—remains an open question. Recent work has proposed the use of a horseshoe prior over node pre-activations of a Bayesian neural network, which effectively turns off nodes that do not help explain the data. In this work, we propose several modeling and inference advances that consistently improve the compactness of the model learned while maintaining predictive performance, especially in smaller-sample settings including reinforcement learning.

Motivated by Supervised Opinion Analysis, we propose a novel framework devoted to Structured Output Learning with Abstention (SOLA). The structure prediction model is able to abstain from predicting some labels in the structured output at a cost chosen by the user in a flexible way. For that purpose, we decompose the problem into the learning of a pair of predictors, one devoted to structured abstention and the other, to structured output prediction. To compare fully labeled training data with predictions potentially containing abstentions, we define a wide class of asymmetric abstention-aware losses. Learning is achieved by surrogate regression in an appropriate feature space while prediction with abstention is performed by solving a new pre-image problem. Thus, SOLA extends recent ideas about Structured Output Prediction via surrogate problems and calibration theory and enjoys statistical guarantees on the resulting excess risk. Instantiated on a hierarchical abstention-aware loss, SOLA is shown to be relevant for fine-grained opinion mining and gives state-of-the-art results on this task. Moreover, the abstention-aware representations can be used to competitively predict user-review ratings based on a sentence-level opinion predictor.

We present a new method of blackbox optimization via gradient approximation with the use of structured random orthogonal matrices, providing more accurate estimators than baselines and with provable theoretical guarantees. We show that this algorithm can be successfully applied to learn better quality compact policies than those using standard gradient estimation techniques. The compact policies we learn have several advantages over unstructured ones, including faster training algorithms and faster inference. These benefits are important when the policy is deployed on real hardware with limited resources. Further, compact policies provide more scalable architectures for derivative-free optimization (DFO) in high-dimensional spaces. We show that most robotics tasks from the OpenAI Gym can be solved using neural networks with less than 300 parameters, with almost linear time complexity of the inference phase, with up to 13x fewer parameters relative to the Evolution Strategies (ES) algorithm introduced by Salimans et al. (2017). We do not need heuristics such as fitness shaping to learn good quality policies, resulting in a simple and theoretically motivated training mechanism.

In recent years, Deep Reinforcement Learning has made impressive advances in solving several important benchmark problems for sequential decision making. Many control applications use a generic multilayer perceptron (MLP) for non-vision parts of the policy network. In this work, we propose a new neural network architecture for the policy network representation that is simple yet effective. The proposed Structured Control Net (SCN) splits the generic MLP into two separate sub-modules: a nonlinear control module and a linear control module. Intuitively, the nonlinear control is for forward-looking and global control, while the linear control stabilizes the local dynamics around the residual of global control. We hypothesize that this will bring together the benefits of both linear and nonlinear policies: improve training sample efficiency, final episodic reward, and generalization of learned policy, while requiring a smaller network and being generally applicable to different training methods. We validated our hypothesis with competitive results on simulations from OpenAI MuJoCo, Roboschool, Atari, and a custom urban driving environment, with various ablation and generalization tests, trained with multiple black-box and policy gradient training methods. The proposed architecture has the potential to improve upon broader control tasks by incorporating problem specific priors into the architecture. As a case study, we demonstrate much improved performance for locomotion tasks by emulating the biological central pattern generators (CPGs) as the nonlinear part of the architecture.

Deep nets generalize well despite having more parameters than the number of training samples. Recent works try to give an explanation using PAC-Bayes and Margin-based analyses, but do not as yet result in sample complexity bounds better than naive parameter counting. The current paper shows generalization bounds that are orders of magnitude better in practice. These rely upon new succinct reparametrizations of the trained net — a compression that is explicit and efficient. These yield generalization bounds via a simple compression-based framework introduced here. Our results also provide some theoretical justification for widespread empirical success in compressing deep nets. Analysis of correctness of our compression relies upon some newly identified noise stability properties of trained deep nets, which are also experimentally verified. The study of these properties and resulting generalization bounds are also extended to convolutional nets, which had eluded earlier attempts on proving generalization.

We study streaming algorithms for principal component analysis (PCA) in noisy settings. We present computationally efficient algorithms with sub-linear regret bounds for PCA in the presence of noise, missing data, and gross outliers.

A large fraction of the arithmetic operations required to evaluate deep neural networks (DNNs) consists of matrix multiplications, in both convolution and fully connected layers. We perform end-to-end learning of low-cost approximations of matrix multiplications in DNN layers by casting matrix multiplications as 2-layer sum-product networks (SPNs) (arithmetic circuits) and learning their (ternary) edge weights from data. The SPNs disentangle multiplication and addition operations and enable us to impose a budget on the number of multiplication operations. Combining our method with knowledge distillation and applying it to image classification DNNs (trained on ImageNet) and language modeling DNNs (using LSTMs), we obtain a first-of-a-kind reduction in number of multiplications (over 99.5%) while maintaining the predictive performance of the full-precision models. Finally, we demonstrate that the proposed framework is able to rediscover Strassen’s matrix multiplication algorithm, learning to multiply $2 \times 2$ matrices using only 7 multiplications instead of 8.

We present a stochastic algorithm to compute the barycenter of a set of probability distributions under the Wasserstein metric from optimal transport. Unlike previous approaches, our method extends to continuous input distributions and allows the support of the barycenter to be adjusted in each iteration. We tackle the problem without regularization, allowing us to recover a sharp output whose support is contained within the support of the true barycenter. We give examples where our algorithm recovers a more meaningful barycenter than previous work. Our method is versatile and can be extended to applications such as generating super samples from a given distribution and recovering blue noise approximations.

Generating video frames that accurately predict future world states is challenging. Existing approaches either fail to capture the full distribution of outcomes, or yield blurry generations, or both. In this paper we introduce a video generation model with a learned prior over stochastic latent variables at each time step. Video frames are generated by drawing samples from this prior and combining them with a deterministic estimate of the future frame. The approach is simple and easily trained end-to-end on a variety of datasets. Sample generations are both varied and sharp, even many frames into the future, and compare favorably to those from existing approaches.

In this paper, we propose a novel reinforcement-learning algorithm consisting in a stochastic variance-reduced version of policy gradient for solving Markov Decision Processes (MDPs). Stochastic variance-reduced gradient (SVRG) methods have proven to be very successful in supervised learning. However, their adaptation to policy gradient is not straightforward and needs to account for I) a non-concave objective function; II) approximations in the full gradient computation; and III) a non-stationary sampling process. The result is SVRPG, a stochastic variance-reduced policy gradient algorithm that leverages on importance weights to preserve the unbiasedness of the gradient estimate. Under standard assumptions on the MDP, we provide convergence guarantees for SVRPG with a convergence rate that is linear under increasing batch sizes. Finally, we suggest practical variants of SVRPG, and we empirically evaluate them on continuous MDPs.

We propose a fast stochastic Hamilton Monte Carlo (HMC) method, for sampling from a smooth and strongly log-concave distribution. At the core of our proposed method is a variance reduction technique inspired by the recent advance in stochastic optimization. We show that, to achieve $\epsilon$ accuracy in 2-Wasserstein distance, our algorithm achieves $\tilde O\big(n+\kappa^{2}d^{1/2}/\epsilon+\kappa^{4/3}d^{1/3}n^{2/3}/\epsilon^{2/3}\big)$ gradient complexity (i.e., number of component gradient evaluations), which outperforms the state-of-the-art HMC and stochastic gradient HMC methods in a wide regime. We also extend our algorithm for sampling from smooth and general log-concave distributions, and prove the corresponding gradient complexity as well. Experiments on both synthetic and real data demonstrate the superior performance of our algorithm.

We propose a stochastic variance-reduced cubic regularized Newton method (SVRC) for non-convex optimization. At the core of our algorithm is a novel semi-stochastic gradient along with a semi-stochastic Hessian, which are specifically designed for cubic regularization method. We show that our algorithm is guaranteed to converge to an $(\epsilon,\sqrt{\epsilon})$-approximate local minimum within $\tilde{O}(n^{4/5}/\epsilon^{3/2})$ second-order oracle calls, which outperforms the state-of-the-art cubic regularization algorithms including subsampled cubic regularization. Our work also sheds light on the application of variance reduction technique to high-order non-convex optimization methods. Thorough experiments on various non-convex optimization problems support our theory.