We consider a statistical inverse learning problem, where we observe the image of a function $f$ through a linear operator $A$ at i.i.d. random design points $X_i$, superposed with an additive noise. The distribution of the design points is unknown and can be very general. We analyze simultaneously the direct (estimation of $Af$) and the inverse (estimation of $f$) learning problems. In this general framework, we obtain strong and weak minimax optimal rates of convergence (as the number of observations $n$ grows large) for a large class of spectral regularization methods over regularity classes defined through appropriate source conditions. This improves on or completes previous results obtained in related settings. The optimality of the obtained rates is shown not only in the exponent in $n$ but also in the explicit dependency of the constant factor in the variance of the noise and the radius of the source condition set.

Last November, Google opened up its in-house machine learning software TensorFlow, making the program that powers its translation services and photo analytics (among many other things) open-source and free to download. Now, the company is giving TensorFlow the machine learning equivalent of smart pills, releasing a distributed version of the software that allows it to run across multiple machines -- up to hundreds at a time. This sounds like an obvious way to improve TensorFlow, and, well, it is. Machine learning software only gets to be clever by analyzing large amounts of data -- looking for common properties and trends like facial features in photographs, for example. Letting TensorFlow run these sorts of operations on networks of computers simultaneously rather than individual machines means users can make smarter systems, and improve them faster.

Yuuta Banda poses for a photo with OriHime, his robot avatar. When Yuuta Banda was just four years old, he suffered a car accident that left him paralyzed, connected to a respiratory machine, and confined to bed for life. But almost two decades later, he's been able to experience different places, and even find a job thanks to OriHime, his robot avatar. "At first I couldn't understand what was so great about OriHime, but I gradually learned through using it that [the robot] afforded people with a sense of presence," Banda told me in an email. "I felt a greater sense of satisfaction as I spoke with people in different places to me through the robot."

Illinois Structured Learning Package (Illinois-SL) is a general purpose JAVA library for performing structured learning. It houses learning algorithms like averaged Structured Perceptron and Structured SVM with L2-Loss, and provides a minimal interface for your structured learning needs. The training algorithm employed for training SSVM is dual coordinate descent(DCD), which has been proven to have very good convergence properties. Illinois-SL comes with an efficient implementation of DCD with support for multi-threading. Illinois-SL provides a simple and neat framework for developing applications using structured prediction models.

In a world where digital tools support virtually every part of our lives, why is it that the full power of such tools has yet to be unleashed to those who might benefit most - educators and learners? In the latest of its series of publications concerning digital learning, Intelligence Unleashed: An Argument for AI in Education, Pearson, in collaboration with the UCL Knowledge Lab, maps out how artificial intelligence in education (AIEd) can be used to create learning tools that are more efficient, flexible and inclusive than those currently available; tools that will help learners prepare for an economy that is swiftly being reshaped by digital technologies. The authors, led by Professor Rose Luckin of the UCL Knowledge Lab, highlight existing and emergent technology that could be leveraged to address some of the most intractable issues in education, including achievement gaps. For example, technology available today could be applied to support student learning at a scale previously unimaginable by providing one-on-one tutoring to every student, in every subject. Existing technologies also have the capacity to provide intelligent support to learners working in a group, and to create authentic virtual learning environments where students have the right support, at the right time, to tackle real-life problems and puzzles.

Over the past few years the field of Machine Learning has entered the general parlance. From free massive open online courses to image recognition benchmarks being broken and decades of Atari games being mastered. During the same period developers have witnessed the release of several popular open source frameworks and libraries. The chart below shows different open source machine learning projects by initial commit date and programming language. The size represents the popularity of a project based on number of Github stargazers.

The assumption that a college education should take four years is baked into American culture. Colleges in the colonial days were founded on the premise of a four-year degree, a concept imported from Europe. Harvard University experimented with a three-year degree when it was founded in 1636, but the test was short-lived, and the four-year degree has been the standard ever since. We expect students to enter college at 18 and leave when they turn 22, and we worry about those who take a more circuitous route to graduation. But we need to reconsider that long-established, one-size-fits all model.

As we put the finishing touches on what promises to be another outstanding Hardcore Data Science Day at Strata Hadoop World in New York, I sat down with my co-organizer Ben Recht for the the latest episode of the O'Reilly Data Show Podcast. Recht is a UC Berkeley faculty member and member of AMPLab, and his research spans many areas of interest to data scientists including optimization, compressed sensing, statistics, and machine learning. At the 2014 Strata Hadoop World in NYC, Recht gave an overview of a nascent AMPLab research initiative into machine learning pipelines. The research team behind the project recently released an alpha version of a new software framework called KeystoneML, which gives developers a chance to test out some of the ideas that Recht outlined in his talk last year. We devoted a portion of this Data Show episode to machine learning pipelines in general, and a discussion of KeystoneML in particular.

Robots are taking all the jobs. But are we, the average, moderately skilled humans, screwed, or aren't we? Let me just get it out of the way now: We are, unless there are drastic, immediate changes to education and economic systems around the world. The dominant narrative going around today about Pew Research's new report on artificial intelligence and the future of jobs is that experts can't really decide whether automation is going to make working obsolete, that it's really a toss up whether robots will simply create new jobs in other sectors as they destroy ones in other. That's true, in one sense: The 1,896 futurists, CEOs, journalists, and university professors questioned for the report were split in half over robots will "displace significant numbers of both blue- and white-collar workers," with 52 percent of respondents agreeing that "human ingenuity will create new jobs, industries, and ways to make a living, just as it has been doing since the dawn of the Industrial Revolution."

Tree methods are commonly used in data science to understand patterns within data and to build predictive models. The term Tree Methods covers a variety of techniques with different levels of complexity but my aim is to highlight three I find useful. To set the problem up let's assume we have a census dataset containing age, education, employment status and so on. Given all this information we want to see if we can predict whether a person earns more than 50k per year. How can tree methods help us?