As everyone knows Machine learning studies computer algorithms for learning to do stuff. We might, for instance, be interested in learning to complete a task, or to make accurate predictions, or to behave intelligently. The learning that is being done is always based on some sort of observations or data, such as examples…direct experience, or instruction. So in general, machine learning is about learning to do better in the future based on what was experienced in the past. Machine learning is being used in a lot of real-world applications for various purpose.

In this paper, we introduce a new learning strategy based on a seminal idea of Mojirsheibani (1999, 2000, 2002a, 2002b), who proposed a smart method for combining several classifiers, relying on a consensus notion. In many aggregation methods, the prediction for a new observation x is computed by building a linear or convex combination over a collection of basic estimators r1(x),. .. , rm(x) previously calibrated using a training data set. Mojirsheibani proposes to compute the prediction associated to a new observation by combining selected outputs of the training examples. The output of a training example is selected if some kind of consensus is observed: the predictions computed for the training example with the different machines have to be "similar" to the prediction for the new observation. This approach has been recently extended to the context of regression in Biau et al. (2016). In the original scheme, the agreement condition is actually required to hold for all individual estimators, which appears inadequate if there is one bad initial estimator. In practice, a few disagreements are allowed ; for establishing the theoretical results, the proportion of estimators satisfying the condition is required to tend to 1. In this paper, we propose an alternative procedure, mixing the previous consensus ideas on the predictions with the Euclidean distance computed between entries. This may be seen as an alternative approach allowing to reduce the effect of a possibly bad estimator in the initial list, using a constraint on the inputs. We prove the consistency of our strategy in classification and in regression. We also provide some numerical experiments on simulated and real data to illustrate the benefits of this new aggregation method. On the whole, our practical study shows that our method may perform much better than the original combination technique, and, in particular, exhibit far less variance. We also show on simulated examples that this procedure mixing inputs and outputs is still robust to high dimensional inputs.

Low-precision computation is often used to lower the time and energy cost of machine learning, and recently hardware accelerators have been developed to support it. Still, it has been used primarily for inference - not training. Previous low-precision training algorithms suffered from a fundamental tradeoff: as the number of bits of precision is lowered, quantization noise is added to the model, which limits statistical accuracy. To address this issue, we describe a simple low-precision stochastic gradient descent variant called HALP. HALP converges at the same theoretical rate as full-precision algorithms despite the noise introduced by using low precision throughout execution. The key idea is to use SVRG to reduce gradient variance, and to combine this with a novel technique called bit centering to reduce quantization error. We show that on the CPU, HALP can run up to $4 \times$ faster than full-precision SVRG and can match its convergence trajectory. We implemented HALP in TensorQuant, and show that it exceeds the validation performance of plain low-precision SGD on two deep learning tasks.

Structured prediction energy networks (SPENs; Belanger & McCallum 2016) use neural network architectures to define energy functions that can capture arbitrary dependencies among parts of structured outputs. Prior work used gradient descent for inference, relaxing the structured output to a set of continuous variables and then optimizing the energy with respect to them. We replace this use of gradient descent with a neural network trained to approximate structured argmax inference. This "inference network" outputs continuous values that we treat as the output structure. We develop large-margin training criteria for joint training of the structured energy function and inference network. On multi-label classification we report speedups of 10-60x compared to (Belanger et al., 2017) while also improving accuracy. For sequence labeling with simple structured energies, our approach performs comparably to exact inference while being much faster at test time. We then demonstrate improved accuracy by augmenting the energy with a "label language model" that scores entire output label sequences, showing it can improve handling of long-distance dependencies in part-of-speech tagging. Finally, we show how inference networks can replace dynamic programming for test-time inference in conditional random fields, suggestive for their general use for fast inference in structured settings.

Aberrations in complex biological systems develop in the background of diverse genetic and environmental factors and are associated with multiple complex molecular events. These include changes in the genome, transcriptome, proteome and metabolome, as well as epigenetic effects. Advances in high-throughput profiling techniques have enabled a systematic and comprehensive exploration of the genetic and epigenetic basis of various diseases, including cancer (Kaushik et al., 2016; Lee et al., 2016), diabetes (Sas et al., 2018; Yuan et al., 2014), chronic kidney disease (Atzler et al., 2014), etc. Further, such multi-Omics collections have become available for patients belonging to different, but related disease subtypes, with The Cancer Genome Atlas (TCGA: Tomczak et al. (2015)) being a prototypical one. Hence, there is an increasing need for models that can integrate such complex data both vertically across multiple modalities and horizontally across different disease subtypes. Figure 1 provides a schematic representation of the horizontal and vertical structure of such heterogeneous multi-modal Omics data as outlined above.

A key problem in inference for high dimensional unknowns is the design of sampling algorithms whose performance scales favourably with the dimension of the unknown. A typical setting in which these problems arise is the area of Bayesian inverse problems. In such problems, which include graph-based learning, nonparametric regression and PDE-based inversion, the unknown can be viewed as an infinite-dimensional parameter (such as a function) that has been discretised. This results in a high-dimensional space for inference. Here we study robustness of an MCMC algorithm for posterior inference; this refers to MCMC convergence rates that do not deteriorate as the discretisation becomes finer. When a Gaussian prior is employed there is a known methodology for the design of robust MCMC samplers. However, one often requires more flexibility than a Gaussian prior can provide: hierarchical models are used to enable inference of parameters underlying a Gaussian prior; or non-Gaussian priors, such as Besov, are employed to induce sparse MAP estimators; or deep Gaussian priors are used to represent other non-Gaussian phenomena; and piecewise constant functions, which are necessarily non-Gaussian, are required for classification problems. The purpose of this article is to show that the simulation technology available for Gaussian priors can be exported to such non-Gaussian priors. The underlying methodology is based on a white noise representation of the unknown. This is exploited both for robust posterior sampling and for joint inference of the function and parameters involved in the specification of its prior, in which case our framework borrows strength from the well-developed non-centred methodology for Bayesian hierarchical models. The desired robustness of the proposed sampling algorithms is supported by some theory and by extensive numerical evidence from several challenging problems.

In extreme classification problems, learning algorithms are required to map instances to labels from an extremely large label set. We build on a recent extreme classification framework with logarithmic time and space, and on a general approach for error correcting output coding (ECOC), and introduce a flexible and efficient approach accompanied by bounds. Our framework employs output codes induced by graphs, and offers a tradeoff between accuracy and model size. We show how to find the sweet spot of this tradeoff using only the training data. Our experimental study demonstrates the validity of our assumptions and claims, and shows the superiority of our method compared with state-of-the-art algorithms.

We give the first polynomial-time algorithm for performing linear or polynomial regression resilient to adversarial corruptions in both examples and labels. Given a sufficiently large (polynomial-size) training set drawn i.i.d. from distribution D and subsequently corrupted on some fraction of points, our algorithm outputs a linear function whose squared error is close to the squared error of the best-fitting linear function with respect to D, assuming that the marginal distribution of D over the input space is \emph{certifiably hypercontractive}. This natural property is satisfied by many well-studied distributions such as Gaussian, strongly log-concave distributions and, uniform distribution on the hypercube among others. We also give a simple statistical lower bound showing that some distributional assumption is necessary to succeed in this setting. These results are the first of their kind and were not known to be even information-theoretically possible prior to our work. Our approach is based on the sum-of-squares (SoS) method and is inspired by the recent applications of the method for parameter recovery problems in unsupervised learning. Our algorithm can be seen as a natural convex relaxation of the following conceptually simple non-convex optimization problem: find a linear function and a large subset of the input corrupted sample such that the least squares loss of the function over the subset is minimized over all possible large subsets.

Many loss functions in representation learning are invariant under a continuous symmetry transformation. As an example, consider word embeddings (Mikolov et al., 2013b), where the loss remains unchanged if we simultaneously rotate all word and context embedding vectors. We show that representation learning models with a continuous symmetry and a quadratic Markovian time series prior possess so-called Goldstone modes. These are low cost deviations from the optimum which slow down convergence of gradient descent. We use tools from gauge theory in physics to design an optimization algorithm that solves the slow convergence problem. Our algorithm leads to a fast decay of Goldstone modes, to orders of magnitude faster convergence, and to more interpretable representations, as we show for dynamic extensions of matrix factorization and word embedding models. We present an example application, translating modern words into historic language using a shared representation space.

In this paper, we present a technique for generating artificial datasets that retain statistical properties of the real data while providing differential privacy guarantees with respect to this data. We include a Gaussian noise layer in the discriminator of a generative adversarial network to make the output and the gradients differentially private with respect to the training data, and then use the generator component to synthesise privacy-preserving artificial dataset. Our experiments show that under a reasonably small privacy budget we are able to generate data of high quality and successfully train machine learning models on this artificial data.