Today OpenAI, a non-profit artificial intelligence research company, launched OpenAI Gym, a toolkit for developing and comparing reinforcement learning algorithms. It supports teaching agents everything from walking to playing games like Pong or Go. OpenAI researcher John Schulman shared some details about his organization, and how OpenAI Gym will make it easier for AI researchers to design, iterate and improve their next generation applications. John studied physics at Caltech, and went to UC Berkeley for graduate school. There, after a brief stint in neuroscience, he studied machine learning and robotics under Pieter Abbeel, eventually honing in on reinforcement learning as his primary topic of interest.

Always wanted to compete in a Kaggle competition, but not sure you have the right skill set? We created a free interactive Machine Learning tutorial to help you out!Together with the team behind Kaggle, we have developed a free interactive tutorial on how to apply Machine Learning Techniques that can be used in your Kaggle competitions. Step by step, through fun coding challenges, the tutorial will learn you how to predict survival rate for Kaggle's Titanic competition using R and Machine Learning. Start the Machine Learning with R tutorial now!This free R tutorial is provided by DataCamp, an online interactive education platform that offers courses in data science and R programming. Each course is built around a certain data science topic, and combines video instruction with in-browser coding challenges so that you can learn by doing.

The last few years have seen an explosion of interest in Machine Learning (ML) technology and potential applications. Machine Learning is the unsung hero that powers many applications, systems, sensors, devices, and products. Today, Machine Learning is so pervasive that we can often assume its presence in most of the applications and systems without having to specifically call it out. In simple terms, machine learning is a computer's ability to learn from data, and it is one of the most useful tools we have to develop intelligent systems and applications. Machine learning is used widely today for all kinds of tasks, from churn prediction in large companies, to web search, to medical diagnostics, to robotics.

In my last post on the subject of AI, values,and socially beneficial simulation, I looked at one way of approaching the problem of AI values, society, and institutions through simulation and modeling. It occurred to me while writing it that there was also another way that might be interesting to think about, and it stems from a rejected piece I wrote last year on the subject of the computer as a "moral mirror." Though that piece was rejected for a reason (it wasn't that well-written), it has come into my mind again over the last week as the Internet has collectively reflected on my Slate article about AI values. I have some commonalities and differences with Stuart Russell's view of AI value alignment. I've looked at them here, here, here, and here.

This book is a high quality statistical text with R as the software of choice. If you want to be comfortable with fundamental concepts in parallel with learning R, then this is the book for you. Having said this, you will love this book even if you have studied advanced statistics. The book also covers some advanced machine learning concepts such as support machine learning (SVM) and regularization.

They are becoming more capable at a faster pace than people and therefore will effectively outsmart us in a short amount of time," Mr Thrun now believes that education is the best way to tackle the big upheavals that are likely to spring from the widespread adoption of artificial intelligence and robotics. But not education as you might know it. "We are still living with an educational system that was developed in the 1800s and 1900s," he says. "Needs have shifted in the modern age and what's also shifted is our ability to use digital media. We can now deliver a top-notch education at home in a way that was never possible before."

In a recent interview with The Verge, the Microsoft founder and long-time education proponent explained how advancements in AI will make personalized learning more effective. At present, personalized-learning software allows students to move through course content at their own pace: Students who master material quickly advance to the next subject, while students who require extra practice can revisit modules as needed. Although studies -- such as those conducted by the Bill and Melinda Gates Foundation -- show that the approach can positively impact learning outcomes, Gates thinks there's room for improvement, especially when it comes to human-software interactions. He describes the technology's failings in his interview with The Verge: Today's interactivity is OK: I answered a few questions wrong, so then it repeats the lecture. You know, part of the reason I'm so willing to tackle new subjects is that for each of those subjects, if I get utterly confused, I know somebody I can send an email to, and they'll straighten me out.

We present algorithms to effectively represent a set of Markov decision processes (MDPs), whose optimal policies have already been learned, by a smaller source subset for lifelong, policy-reuse-based transfer learning in reinforcement learning. This is necessary when the number of previous tasks is large and the cost of measuring similarity counteracts the benefit of transfer. The source subset forms an `$\epsilon$-net' over the original set of MDPs, in the sense that for each previous MDP $M_p$, there is a source $M^s$ whose optimal policy has $<\epsilon$ regret in $M_p$. Our contributions are as follows. We present EXP-3-Transfer, a principled policy-reuse algorithm that optimally reuses a given source policy set when learning for a new MDP. We present a framework to cluster the previous MDPs to extract a source subset. The framework consists of (i) a distance $d_V$ over MDPs to measure policy-based similarity between MDPs; (ii) a cost function $g(\cdot)$ that uses $d_V$ to measure how good a particular clustering is for generating useful source tasks for EXP-3-Transfer and (iii) a provably convergent algorithm, MHAV, for finding the optimal clustering. We validate our algorithms through experiments in a surveillance domain.

SAN MATEO, California – April 28th, 2016 – Movidius, the leader in low-power machine vision technology, today announced both the Fathom Neural Compute Stick – the world's first deep learning acceleration module, and Fathom deep learning software framework. Both tools hand-in-hand will allow powerful neural networks to be moved out of the cloud, and deployed natively in end-user devices. The new Fathom Neural Compute Stick is the world's first embedded neural network accelerator. With the company's ultra-low power, high performance Myriad 2 processor inside, the Fathom Neural Compute Stick can run fully-trained neural networks at under 1 Watt of power. Thanks to standard USB connectivity, the Fathom Neural Compute Stick can be connected to a range of devices and enhance their neural compute capabilities by orders of magnitude.

We call it "AI-assisted grading," where machines assist humans in grading consistently and quickly. For many questions, Gradescope's AI will be able to learn how to grade all student submissions from a small number of answers graded by the instructor, such that an instructor would only have to grade about ten answers out of a hundred submissions. For known questions, Gradescope's AI will be able to grade instantly, without any human input. This seems to be the evolving narrative of man and machine, collaborating together for the greater good, software and robots becoming "partners" in work rather than usurpers of it. And for the time stressed teacher or teaching assistant, having to "grade" ten essays instead of hundreds is certainly appealing.