In machine learning, there is an ongoing revival of the use of variational Bayes (VB) to deal with complex probabilistic models with hidden variables. The revival is driven by the use of stochastic methods to approximate the VB lower bound and associated gradients. See for example [1, 2, 5]. The advantages include automated inference [3] and also that they are applicable to a much wider class of probabilistic models. While the basic recipes are limited in the flexibility of the approximate hidden-variable posteriors, there are ongoing efforts to make them more flexible.

We introduce a deep multitask architecture to integrate multityped representations of multimodal objects. This multitype exposition is less abstract than the multimodal characterization, but more machine-friendly, and thus is more precise to model. For example, an image can be described by multiple visual views, which can be in the forms of bag-of-words (counts) or color/texture histograms (real-valued). At the same time, the image may have several social tags, which are best described using a sparse binary vector. Our deep model takes as input multiple type-specific features, narrows the cross-modality semantic gaps, learns cross-type correlation, and produces a high-level homogeneous representation. At the same time, the model supports heterogeneously typed tasks. We demonstrate the capacity of the model on two applications: social image retrieval and multiple concept prediction. The deep architecture produces more compact representation, naturally integrates multiviews and multimodalities, exploits better side information, and most importantly, performs competitively against baselines.

Saddle points have long been regarded as a tremendous obstacle for continuous optimization. There are many well known examples when worst case initialization of gradient descent provably converge to saddle points [20, Section 1.2.3], and hardness results which show that finding even a local minimizer of nonconvex functions is NP-Hard in the worst case [19]. However, such worst-case analyses have not daunted practitioners, and high quality solutions of continuous optimization problems are readily found by a variety of simple algorithms. Building on tools from the theory of dynamical systems, this paper demonstrates that, under very mild regularity conditions, saddle points are indeed of little concern for the gradient method.

In practice, there are often explicit constraints on what representations or decisions are acceptable in an application of machine learning. For example it may be a legal requirement that a decision must not favour a particular group. Alternatively it can be that that representation of data must not have identifying information. We address these two related issues by learning flexible representations that minimize the capability of an adversarial critic. This adversary is trying to predict the relevant sensitive variable from the representation, and so minimizing the performance of the adversary ensures there is little or no information in the representation about the sensitive variable. We demonstrate this adversarial approach on two problems: making decisions free from discrimination and removing private information from images. We formulate the adversarial model as a minimax problem, and optimize that minimax objective using a stochastic gradient alternate min-max optimizer. We demonstrate the ability to provide discriminant free representations for standard test problems, and compare with previous state of the art methods for fairness, showing statistically significant improvement across most cases. The flexibility of this method is shown via a novel problem: removing annotations from images, from unaligned training examples of annotated and unannotated images, and with no a priori knowledge of the form of annotation provided to the model.

Tackling pattern recognition problems in areas such as computer vision, bioinformatics, speech or text recognition is often done best by taking into account task-specific statistical relations between output variables. In structured prediction, this internal structure is used to predict multiple outputs simultaneously, leading to more accurate and coherent predictions. Structural support vector machines (SSVMs) are nonprobabilistic models that optimize a joint input-output function through margin-based learning. Because SSVMs generally disregard the interplay between unary and interaction factors during the training phase, final parameters are suboptimal. Moreover, its factors are often restricted to linear combinations of input features, limiting its generalization power. To improve prediction accuracy, this paper proposes: (i) Joint inference and learning by integration of back-propagation and loss-augmented inference in SSVM subgradient descent; (ii) Extending SSVM factors to neural networks that form highly nonlinear functions of input features. Image segmentation benchmark results demonstrate improvements over conventional SSVM training methods in terms of accuracy, highlighting the feasibility of end-to-end SSVM training with neural factors.

Sequence to sequence learning has recently emerged as a new paradigm in supervised learning. To date, most of its applications focused on only one task and not much work explored this framework for multiple tasks. This paper examines three multi-task learning (MTL) settings for sequence to sequence models: (a) the oneto-many setting - where the encoder is shared between several tasks such as machine translation and syntactic parsing, (b) the many-to-one setting - useful when only the decoder can be shared, as in the case of translation and image caption generation, and (c) the many-to-many setting - where multiple encoders and decoders are shared, which is the case with unsupervised objectives and translation. Our results show that training on a small amount of parsing and image caption data can improve the translation quality between English and German by up to 1.5 BLEU points over strong single-task baselines on the WMT benchmarks. Furthermore, we have established a new state-of-the-art result in constituent parsing with 93.0 F1. Lastly, we reveal interesting properties of the two unsupervised learning objectives, autoencoder and skip-thought, in the MTL context: autoencoder helps less in terms of perplexities but more on BLEU scores compared to skip-thought.

Representation learning systems typically rely on massive amounts of labeled data in order to be trained to high accuracy. Recently, high-dimensional parametric models like neural networks have succeeded in building rich representations using either compressive, reconstructive or supervised criteria. However, the semantic structure inherent in observations is oftentimes lost in the process. Human perception excels at understanding semantics but cannot always be expressed in terms of labels. Thus, \emph{oracles} or \emph{human-in-the-loop systems}, for example crowdsourcing, are often employed to generate similarity constraints using an implicit similarity function encoded in human perception. In this work we propose to combine \emph{generative unsupervised feature learning} with a \emph{probabilistic treatment of oracle information like triplets} in order to transfer implicit privileged oracle knowledge into explicit nonlinear Bayesian latent factor models of the observations. We use a fast variational algorithm to learn the joint model and demonstrate applicability to a well-known image dataset. We show how implicit triplet information can provide rich information to learn representations that outperform previous metric learning approaches as well as generative models without this side-information in a variety of predictive tasks. In addition, we illustrate that the proposed approach compartmentalizes the latent spaces semantically which allows interpretation of the latent variables.

While most approaches to automatically recognizing entailment relations have used classifiers employing hand engineered features derived from complex natural language processing pipelines, in practice their performance has been only slightly better than bag-of-word pair classifiers using only lexical similarity. The only attempt so far to build an end-to-end differentiable neural network for entailment failed to outperform such a simple similarity classifier. In this paper, we propose a neural model that reads two sentences to determine entailment using long short-term memory units. We extend this model with a word-by-word neural attention mechanism that encourages reasoning over entailments of pairs of words and phrases. Furthermore, we present a qualitative analysis of attention weights produced by this model, demonstrating such reasoning capabilities. On a large entailment dataset this model outperforms the previous best neural model and a classifier with engineered features by a substantial margin. It is the first generic end-to-end differentiable system that achieves state-of-the-art accuracy on a textual entailment dataset.

We adapt the ideas underlying the success of Deep Q-Learning to the continuous action domain. We present an actor-critic, model-free algorithm based on the deterministic policy gradient that can operate over continuous action spaces. Using the same learning algorithm, network architecture and hyper-parameters, our algorithm robustly solves more than 20 simulated physics tasks, including classic problems such as cartpole swing-up, dexterous manipulation, legged locomotion and car driving. Our algorithm is able to find policies whose performance is competitive with those found by a planning algorithm with full access to the dynamics of the domain and its derivatives. We further demonstrate that for many of the tasks the algorithm can learn policies end-to-end: directly from raw pixel inputs.

We develop a scalable deep non-parametric generative model by augmenting deep Gaussian processes with a recognition model. Inference is performed in a novel scalable variational framework where the variational posterior distributions are reparametrized through a multilayer perceptron. The key aspect of this reformulation is that it prevents the proliferation of variational parameters which otherwise grow linearly in proportion to the sample size. We derive a new formulation of the variational lower bound that allows us to distribute most of the computation in a way that enables to handle datasets of the size of mainstream deep learning tasks. We show the efficacy of the method on a variety of challenges including deep unsupervised learning and deep Bayesian optimization.