We propose a general approach for supervised learning with structured output spaces, such as combinatorial and polyhedral sets, that is based on minimizing estimated conditional risk functions. Given a loss function defined over pairs of output labels, we first estimate the conditional risk function by solving a (possibly infinite) collection of regularized least squares problems. A prediction is made by solving an inference problem that minimizes the estimated conditional risk function over the output space. We show that this approach enables, in some cases, efficient training and inference without explicitly introducing a convex surrogate for the original loss function, even when it is discontinuous. Empirical evaluations on real-world and synthetic data sets demonstrate the effectiveness of our method in adapting to a variety of loss functions.

The notion of developing statistical methods in machine learning which are robust to adversarial perturbations in the underlying data has been the subject of increasing interest in recent years. A common feature of this work is that the adversarial robustification often corresponds exactly to regularization methods which appear as a loss function plus a penalty. In this paper we deepen and extend the understanding of the connection between robustification and regularization (as achieved by penalization) in regression problems. Specifically, (a) in the context of linear regression, we characterize precisely under which conditions on the model of uncertainty used and on the loss function penalties robustification and regularization are equivalent, and (b) we extend the characterization of robustification and regularization to matrix regression problems (matrix completion and Principal Component Analysis).

Noname manuscript No. (will be inserted by the editor) Abstract Recently, two-dimensional canonical correlation analysis (2DCCA) has been successfully applied for image feature extraction. The method instead of concatenating the columns of the images to the one-dimensional vectors, directly works with two-dimensional image matrices. Although 2DCCA works well in different recognition tasks, it lacks a probabilistic interpretation. In this paper, we present a probabilistic framework for 2DCCA called probabilistic 2DCCA (P2DCCA) and an iterative EM based algorithm for optimizing the parameters. Experimental results on synthetic and real data demonstrate superior performance in loading factor estimation for P2DCCA compared to 2DCCA. For real data, three subsets of AR face database and also the UMIST face database confirm the robustness of the proposed algorithm in face recognition tasks with different illumination conditions, facial expressions, poses and occlusions. Keywords Canonical Correlation Analysis (CCA) · Two-dimensional CCA · Probabilistic Feature extraction · Dimension Reduction · Face recognition 1 Introduction Although many real-world applications encounter high dimensional data, the most informative part of the data can be modeled in a low dimensional space. Moreover, processing high-dimensional data is a time consuming process and requires lots of resources. To tackle these problems, feature extraction has been used as a tool for finding a compact and meaningful data representation. For single-mode source data, some subspace learning methods are conducted to learn more semantic description subspaces.

Stochastic approximation algorithms are sequential nonparametric methods for finding a zero or minimum of a function in the situation where only the noisy observations of the function values are available. Two timescale stochastic approximation algorithms represent one of the most general subclasses of stochastic approximation methods. These algorithms consist of two coupled recursions which are updated with different (one is considerably smaller than the other) step sizes which in turn facilitate convergence for such algorithms. Two timescale stochastic approximation algorithms [19] have successfully been applied to several complex problems arising in the areas of reinforcement learning, signal processing and admission control in communication networks. There are many reinforcement learning applications (precisely those where parameterization of value function is implemented) where non-additive Markov noise is present in one or both iterates thus requiring the current two timescale framework to be extended to include Markov noise (for example, in [13, p. 5] it is mentioned that in order to generalize the analysis to Markov noise, the theory of two timescale stochastic approximation needs to include the latter).

Baidu (NASDAQ:BIDU) has recently reported Q4 results, beating earnings estimates but with a slight miss on revenue. Shares fell 5% after the news mainly due to the relatively weak guidance. The good news is that the effects of new regulation on online advertising is mostly gone and the core search business will soon be growing again, helped by the integration of AI. Baidu's core search results were actually better than I thought. I expected the biggest impact of the new regulation to have its greatest impact on Q4 results, and I was expecting a bigger contraction.

AI techniques are widely used as part of modern applications to improve application capabilities. Now, software developers are starting to look at how artificial intelligence techniques, such as deep learning, can be used to improve the ability to understand complex software, said Steven Lowe, principal consultant at ThoughtWorks Inc., a Chicago-based application development consultancy, at DeveloperWeek in San Francisco. Lowe has been working on a novel approach for using deep learning to analyze how the structure of software can quickly gain insights from old codebases. Software engineers are approaching development and enterprise design in an entirely new way, thanks to the cloud. In this expert handbook, explore how your peers are leveraging the cloud to streamline app lifecycle management, save money, and make production and security more efficient.

Normally it would be considered rude to call a class "bullshit," but here's one time you can get away with it. Two University of Washington professors are teaching a course to help students "think critically about the data and models that constitute evidence in the social and natural sciences," according to the introduction to the course. The 160-seat seminar, titled "Calling Bullshit in the Age of Big Data," begins in late March and continues for roughly 10 weeks. Members of the general public can follow the course syllabus, including readings and recordings of lectures, at the course's website. At the end of the course, students should be able to "provide your crystals-and-homeopathy aunt or casually racist uncle with an accessible and persuasive explanation of why a claim is bullshit," according to the syllabus.

As tech companies continue to dump mountains of cash into artificial intelligence (AI) development, the technology promises to greatly improve our digital lives. However, the AI ecosystem still has major problems to solve before it can advance, a new report said. The report, released Friday from Edison Investment Research, said that AI has the potential to be a major differentiator but it is still in the early stages of its development. Most of what is referred to as AI, currently, is "simply advanced statistics," the report claims. According to the Edison Investment Research report, there are three goals that must be solved for AI to move out of its infancy.

We have just taken another momentous step in the journey to unveil the hidden wonders of our own planet! Since the launch of the Shell Ocean Discovery XPRIZE at the American Geophysical Union Fall Meeting in San Francisco in December 2015, individuals from around the world have been racing to form Teams and develop a range of groundbreaking technologies to access the deep-sea. Registration closed at the end of September 2016 with 32 bold Teams stepping forward to take on the challenge of mapping and imaging our ocean as never before. Today, we announce the 21 semi-finalists Teams advancing in the Ocean Discovery XPRIZE. These innovative semi-finalist Teams, consisting of almost 350 individuals from 25 countries, represent a broad, impressive diversity of backgrounds and expertise, including middle and high school students, university students, maker-movement enthusiasts, and water and ocean industry professionals.

Ocado, the world's largest online-only supermarket, has been evaluating the feasibility of robotic picking and packing of shopping orders in its highly-automated warehouses through the SoMa project, a Horizon 2020 framework programme for research and innovation funded by the European Union. One of the main challenges of robotic manipulation has been the handling of easily damageable and unpredictably shaped objects such as fruit and vegetable groceries. These products have unique shapes and should be handled in a way that does not cause damage or bruising. To avoid damaging sensitive items, the project uses a compliant gripper (i.e. one that possesses spring-like properties) in conjunction with an industrial robot arm. The variation in shape of the target objects imposes another set of constraints on the design of a suitable gripper.