Former Baidu chief scientist Andrew Ng today announced plans to launch a new business called Deeplearning.ai. The company launched with little information about what to expect beyond an exploration of the "frontiers of AI," but alludes to additional details being shared in August about Deeplearning.ai's Deep learning is a type of AI that involves training large artificial neural networks on a pool of information, then getting them to make inferences about new data. Considered one of the top minds in deep learning today, Ng left Baidu in March and was with the company since 2014 following work as cocreator of the Google Brain AI research project. Ng has also worked as director of the Stanford University's Artificial Intelligence Lab (SAIL) and is cofounder of online education company Coursera.

In this article, we extend the conventional framework of convolutional-Restricted-Boltzmann-Machine to learn highly abstract features among abitrary number of time related input maps by constructing a layer of multiplicative units, which capture the relations among inputs. In many cases, more than two maps are strongly related, so it is wise to make multiplicative unit learn relations among more input maps, in other words, to find the optimal relational-order of each unit. In order to enable our machine to learn relational order, we developed a reinforcement-learning method whose optimality is proven to train the network.

Kernel $k$-means clustering can correctly identify and extract a far more varied collection of cluster structures than the linear $k$-means clustering algorithm. However, kernel $k$-means clustering is computationally expensive when the non-linear feature map is high-dimensional and there are many input points. Kernel approximation, e.g., the Nystr\"om method, has been applied in previous works to approximately solve kernel learning problems when both of the above conditions are present. This work analyzes the application of this paradigm to kernel $k$-means clustering, and shows that applying the linear $k$-means clustering algorithm to $\frac{k}{\epsilon} (1 + o(1))$ features constructed using a so-called rank-restricted Nystr\"om approximation results in cluster assignments that satisfy a $1 + \epsilon$ approximation ratio in terms of the kernel $k$-means cost function, relative to the guarantee provided by the same algorithm without the use of the Nystr\"om method. As part of the analysis, this work establishes a novel $1 + \epsilon$ relative-error trace norm guarantee for low-rank approximation using the rank-restricted Nystr\"om approximation. Empirical evaluations on the $8.1$ million instance MNIST8M dataset demonstrate the scalability and usefulness of kernel $k$-means clustering with Nystr\"om approximation. This work argues that spectral clustering using Nystr\"om approximation---a popular and computationally efficient, but theoretically unsound approach to non-linear clustering---should be replaced with the efficient and theoretically sound combination of kernel $k$-means clustering with Nystr\"om approximation. The superior performance of the latter approach is empirically verified.

Glancing around school classrooms in 2016, it's easy to miss just how far technology has transformed learning over the last decade. The desks, whiteboards and rows of chairs are the same, but so much else has changed that can't be seen. A third of Britain's schools are asking students to bring their own tablets and laptops into the classroom now, coding has been on the national curriculum for three years, and more and more education is happening outside school through apps and digital services. But these changes are just the start. Artificial intelligence (AI) is the next giant leap in learning and, according to those working in the field of education and technology, we haven't seen anything yet.

Granville V., Rasson J.P. Multivariate discriminate analysis and maximum penalized likelihood.... Journal of the Royal Statistical Society, Series B, 57 (1995), 501-517.

Today's educational system is static, generalized and puts less focus on individual self-development than it perhaps should. To make matters worse, students often don't understand why they are learning the things that they're learning, which makes certain classes feel arbitrary and purposeless in the face of their personal ambitions (and has a number of neurological implications we'll soon discuss). With that being said, what could be done to fix these issues and take education to a new level? What could make education more exciting, fun and practical? I believe it comes down to three simple ideas (that aren't new by any means) which can finally be fully explored with smart use of technology.

Full-time faculty members of institutions granting PhD degrees in fields related to Machine Learning are eligible to apply. Awards are structured as one-year unrestricted gifts to academic institutions. Though the funding is not extendable, applicants can submit new proposals for subsequent calls. Project proposals are reviewed by an internal awards panel and the results are communicated to the applicants approximately three months after the submission deadline. Each project will be assigned an Amazon researcher contact.

If you go to startup pitch events, you've seen it happen: An entrepreneur says something--something so naïve, egregious and hackneyed--that it makes the investors, along with educators who are now increasingly in the audience, physically cringe. As funders wince and squirm uncomfortably, some are thinking along the lines of: "How do I respond to this pitch genuinely without coming off like a jerk?" In the interest of fixing this problem at the source, I reached out to some of the advisors and investors in the Dreamit Edtech network to get their "lemon lists" of concepts, statements, and business models that edtech entrepreneurs may want to think twice--or thrice--about. These are not inherently "bad ideas" per se. It's just that the investor community have seen tons of these, and in order to impress you need to jump right to what makes your approach a quantum level better than everything else out there.

Reservoir Computing (RC) refers to a Recurrent Neural Networks (RNNs) framework, frequently used for sequence learning and time series prediction. The RC system consists of a random fixed-weight RNN (the input-hidden reservoir layer) and a classifier (the hidden-output readout layer). Here we focus on the sequence learning problem, and we explore a different approach to RC. More specifically, we remove the non-linear neural activation function, and we consider an orthogonal reservoir acting on normalized states on the unit hypersphere. Surprisingly, our numerical results show that the system's memory capacity exceeds the dimensionality of the reservoir, which is the upper bound for the typical RC approach based on Echo State Networks (ESNs). We also show how the proposed system can be applied to symmetric cryptography problems, and we include a numerical implementation.

It is known that Boosting can be interpreted as a gradient descent technique to minimize an underlying loss function. Specifically, the underlying loss being minimized by the traditional AdaBoost is the exponential loss, which is proved to be very sensitive to random noise/outliers. Therefore, several Boosting algorithms, e.g., LogitBoost and SavageBoost, have been proposed to improve the robustness of AdaBoost by replacing the exponential loss with some designed robust loss functions. In this work, we present a new way to robustify AdaBoost, i.e., incorporating the robust learning idea of Self-paced Learning (SPL) into Boosting framework. Specifically, we design a new robust Boosting algorithm based on SPL regime, i.e., SPLBoost, which can be easily implemented by slightly modifying off-the-shelf Boosting packages. Extensive experiments and a theoretical characterization are also carried out to illustrate the merits of the proposed SPLBoost.