This paper proposes a set of new error criteria and learning approaches, Adaptive Normalized Risk-Averting Training (ANRAT), to attack the non-convex optimization problem in training deep neural networks (DNNs). Theoretically, we demonstrate its effectiveness on global and local convexity lower-bounded by the standard $L_p$-norm error. By analyzing the gradient on the convexity index $\lambda$, we explain the reason why to learn $\lambda$ adaptively using gradient descent works. In practice, we show how this method improves training of deep neural networks to solve visual recognition tasks on the MNIST and CIFAR-10 datasets. Without using pretraining or other tricks, we obtain results comparable or superior to those reported in recent literature on the same tasks using standard ConvNets + MSE/cross entropy. Performance on deep/shallow multilayer perceptrons and Denoised Auto-encoders is also explored. ANRAT can be combined with other quasi-Newton training methods, innovative network variants, regularization techniques and other specific tricks in DNNs. Other than unsupervised pretraining, it provides a new perspective to address the non-convex optimization problem in DNNs.

This work presents a fast and non-convex algorithm for robust subspace recovery. The data sets considered include inliers drawn around a low-dimensional subspace of a higher dimensional ambient space, and a possibly large portion of outliers that do not lie nearby this subspace. The proposed algorithm, which we refer to as Fast Median Subspace (FMS), is designed to robustly determine the underlying subspace of such data sets, while having lower computational complexity than existing methods. We prove convergence of the FMS iterates to a stationary point. Further, under a special model of data, FMS converges to a point which is near to the global minimum with overwhelming probability. Under this model, we show that the iteration complexity is globally bounded and locally $r$-linear. The latter theorem holds for any fixed fraction of outliers (less than 1) and any fixed positive distance between the limit point and the global minimum. Numerical experiments on synthetic and real data demonstrate its competitive speed and accuracy.

Hiding posts on Facebook isn't a new feature, but stopping them from hitting your timeline is. The social media giant is rolling out an option that lets desktop users publish updates only to News Feed - concealing them from the dedicated timeline where content is permanently save. Called'Hide From Your Timeline', this allows users to publish quick questions for friends or post statuses they'd rather not see on their own Timeline. Facebook is rolling out a new option that lets desktop users publish updates only to News Feed -- concealing them from the dedicated timeline where content is permanently save. Called'Hide From Your Timeline', this allows users to publish quick questions for friends or post statuses they'd rather not see on their own timeline Hide From Your Timeline lets desktop users to publish quick questions for friends or post statuses they'd rather not see on their own timeline.

When IBM's deep blue beat Gary Kasparov in chess in 1997, researchers thought it would take computers decades to beat humans at Go: a 2000 year old board game and even as early as last year, researchers thought we were at least 10 years away. But in early 2016, Google's DeepMind, using advances in Deep Learning, beat Lee Sedol the world champion of Go. In addition to games, Deep Learning is opening new business opportunities in various areas which were simply not possible a few years back. Come listen to leaders from Nervana Systems, Natural Selection, Mtell, and Netradyne who will be sharing their personal experiences of starting an AI company, opportunities that exist for starting AI based businesses and lessons learned throughout their career. Alex has a decade of experience in large-scale machine learning and the industrial IoT.

More details have emerged about the sweeping scope of Google/DeepMind's ambitions for pushing its algorithmic fingers deep into the healthcare sector -- including wanting to apply machine learning processing to UK NHS data within five years. New Scientist has obtained a Memorandum of Understanding between DeepMind and the Royal Free NHS Trust in London, which describes what the pair envisage as a "broad ranging, mutually beneficial partnership, engaging in high levels of collaborative activity and maximizing the potential to work on genuinely innovative and transformational projects". Envisaged benefits of the collaboration include improvements in clinical outcomes, patient safety and cost reductions -- the latter being a huge ongoing pressure-point for the free-at-the-point-of-use NHS as demand for its services continues to rise yet government austerity cuts bite into public sector budgets. The MoU sets out a long list of "areas of mutual interest" where the pair see what they dub as "future potential" to work together over the five-year period of collaboration envisaged in the memorandum. The document, only parts of which are legally binding, was signed on January 28 this year.

I met up with founder Ali Parsa at the recent unveiling of a new AI based triage service that aims to make it easier and more effective for patients to take those first steps towards good health. The service was tested both live against experienced doctors and nurses on the day and over a more prolonged period and featured strongly on both occasions. Indeed, the AI system was found to be both more accurate and considerably faster (and therefore cheaper) than human based triage services.

Keras is a deep learning library that wraps the efficient numerical libraries Theano and TensorFlow. In this post you will discover how to develop and evaluate neural network models using Keras for a regression problem. Regression Tutorial with Keras Deep Learning Library in Python Photo by Salim Fadhley, some rights reserved. The problem that we will look at in this tutorial is the Boston house price dataset. You can download this dataset and save it to your current working directly with the file name housing.csv.

Students and faculty in several of Colorado State University's online programs will begin using "intelligent tutoring" technology in courses this fall. This comes as CSU Online announced last week a new partnership with Cognii, Inc., a leading provider of Artificial Intelligence-based educational technology. CSU faculty and instructional designers will work with Cognii to develop learning and assessment tools powered by Cognii's Virtual Learning Assistant, which is designed to improve students' learning outcomes, increase instructors' productivity, and enable high-quality personalized education at a large scale. "The use of Cognii in the classroom is expected to improve learning outcomes, turning assessment into learning while enhancing the effectiveness of the time our faculty devote to teaching," said Mike Palmquist, CSU's Associate Provost for Instructional Innovation. "Through this partnership, CSU is on the cutting edge of recent research and innovation in the fields of natural language processing, cognitive sciences, and machine learning, and an example of how the University is taking bold steps toward transforming access to quality education."

Norwalk, Conn.--June 8, 2016-- TMC announced today the conference program for the Smart Machine Conference, held at Caesars Palace in Las Vegas, NV from July 11 – 14, 2016. The Smart Machines conference is dedicated to exploring the business advantages of technologies like AI, Autonomous Robots, Chat bots, Neural Networks, Deep Learning, Cloud, Big Data, Virtual Reality Assistants and so much more. The conference program includes keynotes, case studies, networking opportunities and a robust exhibit floor. "The smart machine revolution is here and disrupting society as a whole whether it's fraud detection powered by machine learning, customer service chat bots or a champion chess playing computer," said TMC's CEO, Rich Tehrani, the conference's Executive Producer. "At the Smart Machine Conference attendees will learn the competitive advantages of these technologies and what the industry's future holds."

As readers of my blog will know, I want to learn more about machine learning. I've managed to run some samples and I've built my own first little samples. It feels like the next step is to understand more about the different algorithms, for example when to pick which one and how to tune the parameters to achieve the best results. To learn more, I've started to watch the first hours of the awesome courses below. The courses are a great introduction to machine learning and very different from most other videos I found which often seem to assume you are already a data scientist.