We consider binary classification problems in the presence of a teacher, who acts as an intermediary to provide a learning algorithm with meaningful, well-chosen examples. This setting is also known as curriculum learning [1, 2, 3] or self-paced learning [4, 5, 6] in the literature. Existing practical methods [4, 7] that employ such a teacher operate by providing the learning algorithm with easy examples first and then progressively moving on to more difficult examples. Such a strategy is known to improve the generalization ability of the learning algorithm and/or alleviate local minima problems while optimizing non-convex objective functions. In this work, we propose a new method to quantify the notion of easiness of a training example.

Scholars have examined the lawmaker's choice between rules and standards for decades. This paper, however, explores the possibility of a new form of law that renders that choice unnecessary. Advances in technology (such as big data and artificial intelligence) will give rise to this new form – the micro-directive – which will provide the benefits of both rules and standards without the costs of either. Lawmakers will be able to use predictive and communication technologies to enact complex legislative goals that are translated by machines into a vast catalog of simple commands for all possible scenarios. When an individual citizen faces a legal choice, the machine will select from the catalog and communicate to that individual the precise context-specific command (the micro-directive) necessary for compliance.

As you're choking down your latest serving of Trump Clinton Brexit Racism Terrorism Wealth Gap Climate Change Casserole, you could use some good news. Let's start with The Inevitable, the new best-seller by Kevin Kelly, one of our wisest technological prognosticators. "This is the moment that folks in the future will look back at and say, 'Oh to have been alive and well back then!'" Kelly writes. "There has never been a better time with more opportunities, more openings, lower barriers, higher benefit/risk ratios, better returns, greater upside than now. In the mid-2010s, we're getting the first sneak peeks at a bouquet of technologies that can vastly improve the lives of most people on the planet and solve some of our hardest problems--even climate change. Just consider for a moment how much everyday life has been transformed since 2007, when smartphones, social networks and cloud computing took off at about the same time.

When Stanford University student Joshua Browder began accumulating numerous parking tickets in London for minor violations last year, he realized he needed to do something. "After the fourth ticket, my parents said to me, 'You're on your own. We're not going to help you anymore,'" the Britain native told NBC News. Not wanting to pay, and upset with "local governments trying to get away with murder," the computer science and economics student wrote an appeal letter. To his, and his parents', surprise, it worked.

There are different ways an algorithm can model a problem based on its interaction with the experience or environment or whatever we want to call the input data. It is popular in machine learning and artificial intelligence text books to first consider the learning styles that an algorithm can adopt. There are only a few main learning styles or learning models that an algorithm can have and we'll go through them here with a few examples of algorithms and problem types that they suit. This taxonomy or way of organizing machine learning algorithms is useful because it forces you to think about the the roles of the input data and the model preparation process and select one that is the most appropriate for your problem in order to get the best result. When crunching data to model business decisions, you are most typically using supervised and unsupervised learning methods.

Look no further than Google's Tensor Processing Unit (TPU), SoftBank's acquisition of ARM (SoftBank hopes to be a big player in AI), and now a venture-backed startup rolling out a family of "Deep Learning" computers. That startup is Wave Computing, based in Campbell, Calif. The six-year-old company came out of stealth mode Thursday (July 21), revealing its design of a massively parallel dataflow processing architecture called the Wave Dataflow Processing Unit (DPU) for deep learning. Derek Meyer, Wave Computing CEO, told EE Times, "In order to accelerate deep learning, the world needs a new computing architecture." Traditional computer architectures are designed for control flow-oriented applications.

As a Stanford student, Zeshan Hussain found many ways to take part in public service projects near and far – on campus, at a high school on the other side of San Francisco Bay and at a tropical disease hospital in India. In January 2016, along with other members of the Muslim Student Union (MSU) and other student groups, Hussain helped organize Syrian Refugee Awareness Week, which included a teach-in about the crisis, a benefit dinner to raise funds for the charity United Muslim Relief and a clothing collection drive in student residence halls. The organization brought in Sana Khatib, a Syrian-American activist whose father is a former political prisoner and whose family fled Syria and the Assad regime when she was young. Through a clothing drive the MSU also collected 500 pounds of clothing just on campus from students and faculty, an accomplishment Hussain described as "very heartening." "We wanted to raise awareness about the crisis and its history, and about the personal struggles of students who may be refugees, or students who have families that are refugees," he said.

"Mastering a field of knowledge involves not only'learning about' the subject matter but also'learning to be' a full participant in the field." This quote is a passage from the readings this week that really resonated with me. Social learning is a way to really engage in the subject matter because you are not only passively participating in the subject matter by reading it but also engaging in the subject by discussing it with your peers and in some instances with the instructor. The chapter also discusses surface learning in comparison with deep learning. Surface learning is when you know the basic facts of the material but don't understand it enough to use it out of direct context, or at least that was my understanding.

The traditional definition of artificial intelligence is the ability of machines to execute tasks and solve problems in ways normally attributed to humans. Some tasks that we consider simple--recognizing an object in a photo, driving a car--are incredibly complex for AI. Machines can surpass us when it comes to things like playing chess, but those machines are limited by the manual nature of their programming; a 30 gadget can beat us at a board game, but it can't do--or learn to do--anything else. This is where machine learning comes in. Show millions of cat photos to a machine, and it will hone its algorithms to improve at recognizing pictures of cats.

Distributed learning is the problem of inferring a function in the case where training data is distributed among multiple geographically separated sources. Particularly, the focus is on designing learning strategies with low computational requirements, in which communication is restricted only to neighboring agents, with no reliance on a centralized authority. In this thesis, we analyze multiple distributed protocols for a large number of neural network architectures. The first part of the thesis is devoted to a definition of the problem, followed by an extensive overview of the state-of-the-art. Next, we introduce different strategies for a relatively simple class of single layer neural networks, where a linear output layer is preceded by a nonlinear layer, whose weights are stochastically assigned in the beginning of the learning process. We consider both batch and sequential learning, with horizontally and vertically partitioned data. In the third part, we consider instead the more complex problem of semi-supervised distributed learning, where each agent is provided with an additional set of unlabeled training samples. We propose two different algorithms based on diffusion processes for linear support vector machines and kernel ridge regression. Subsequently, the fourth part extends the discussion to learning with time-varying data (e.g. time-series) using recurrent neural networks. We consider two different families of networks, namely echo state networks (extending the algorithms introduced in the second part), and spline adaptive filters. Overall, the algorithms presented throughout the thesis cover a wide range of possible practical applications, and lead the way to numerous future extensions, which are briefly summarized in the conclusive chapter.