A professional highlight of 2017 has been teaching A Glance at Reinforcement Learning – an introductory course I've developed. You can find the course materials on GitHub. This one day course is aimed at data scientists with a grasp of supervised machine learning but no prior understanding of reinforcement learning. Course scope – introduction to the fundamental concepts of reinforcement learning – value function methods dynamic programming, Monte Carlo, temporal difference, Q-Learning, DQN – policy gradient methods score function, REINFORCE, advantage actor-critic, AC3 – AlphaGo – practical concerns reward scaling, mistakes I've made, advice from Vlad Mnih & John Schulman – literature highlights distributional perspective, auxiliary loss functions, inverse RL I've given this course to three batches at Data Science Retreat in Berlin and once to a group of startups from Entrepreneur First in London. Each time I've had great questions, kind feedback and improved my own understanding.

From machine learning to digital twins, the future of business has arrived. Today's pioneers synthesize an array of modern technologies to create solutions that scale. They begin with the end in mind, designing next-generation applications that fundamentally improve the user experience. They recognize that today's consumer demands high-quality products and services, and will accept nothing less. Register for this inaugural episode of Fast Forward to learn about this remarkable new class of information systems.

The problem descriptions are taken straightaway from the assignments. In this assignment a linear classifier will be implemented and it will be trained using stochastic gradient descent with numpy. To make things more intuitive, let's solve a 2D classification problem with synthetic data. As we can notice the data above isn't linearly separable. Hence we should add features(or use non-linear model). Note that decision line between two classes have form of circle, since that we can add quadratic features to make the problem linearly separable.

The course begins with a conceptual introduction to machine learning algorithms. This is followed by an introduction to the implementation of estimators in scikit-learn and best practices for using them. The rest of the course is focused around specific feature sources, and for each progresses through a short introductory lecture followed by three exercises of progressive difficulty, starting with standard and well-behaved cases, and ending with real-world and realistically problematic case studies. Throughout, the focus of the course is on building deep conceptual understanding, exhaustive practical experience, and covering common mistakes and edge cases. Intermingled in the machine learning material will be short discussions of helpful and diagnostic data visualizations.

Many people are finding their iPhone won't let them write one of the most popular words in the English language. Fresh from stopping people from writing'I', Apple's handsets now won't let some of their users write'it'. This time around, it corrects it to I.T. each time someone tries to write the pronoun. It's not clear why it's happening, or whether it's related to the previous time that Apple's autocorrect broke. But the same fix works, until Apple presumably updates its operating system to get around it.

We are excited to announce the Amazon ML Solutions Lab, a new program that connects machine learning experts from across Amazon with AWS customers to help identify novel uses of machine learning inside customers' businesses, and guide them in developing new machine learning-enabled features, products, and processes. Amazon has been investing in machine learning for more than 20 years, innovating in areas such as fulfilment and logistics, personalization and recommendations, forecasting, fraud prevention, and supply chain optimization. The Amazon ML Solutions Lab provides you access to the same talent that built many of Amazon's machine learning-powered products and services. Amazon ML Solutions Lab engagements are designed to take you through the full process of implementing machine learning, combining educational workshops and boot camps, advisory professional services, and hands-on help building custom models with a your own data. The Amazon ML Solutions Lab will partner with you to customize the engagement model to your needs, including hosting you at a dedicated facility at the AWS Headquarters in Seattle or sending machine learning model developers to work on-site with you.

My current research trajectory centers on what I call learning programming at scale. Decades of prior research have worked to improve how computer programming is taught in traditional K–12 and university classrooms, but the vast majority of people around the world--children in low-income areas, working adults with full-time jobs, the fast-growing population of older adults, and millions in developing countries--do not have access to high-quality classroom learning environments. Thus, the central question that drives my research is: How can we better understand the millions of people from diverse backgrounds who are now learning programming online and then design scalable software to support their learning goals? One critical prerequisite for improving how programming is taught is to understand why and how people are currently learning and what obstacles they face. To work toward this goal, I have been studying traditionally under-represented learner populations and non-traditional learning environments.

Labeling training data is increasingly the largest bottleneck in deploying machine learning systems. We present Snorkel, a first-of-its-kind system that enables users to train state-of-the-art models without hand labeling any training data. Instead, users write labeling functions that express arbitrary heuristics, which can have unknown accuracies and correlations. Snorkel denoises their outputs without access to ground truth by incorporating the first end-to-end implementation of our recently proposed machine learning paradigm, data programming. We present a flexible interface layer for writing labeling functions based on our experience over the past year collaborating with companies, agencies, and research labs. In a user study, subject matter experts build models 2.8x faster and increase predictive performance an average 45.5% versus seven hours of hand labeling. We study the modeling tradeoffs in this new setting and propose an optimizer for automating tradeoff decisions that gives up to 1.8x speedup per pipeline execution. In two collaborations, with the U.S. Department of Veterans Affairs and the U.S. Food and Drug Administration, and on four open-source text and image data sets representative of other deployments, Snorkel provides 132% average improvements to predictive performance over prior heuristic approaches and comes within an average 3.60% of the predictive performance of large hand-curated training sets.

Keras models follow Scikit-learn's interface of fit/predict with some notable extensions.

I recently completed all available material (as of October 25, 2017) for Andrew Ng's new deep learning course on Coursera. I found all 3 courses extremely useful and learned an incredible amount of practical knowledge from the instructor, Andrew Ng. Ng does an excellent job of filtering out the buzzwords and explaining the concepts in a clear and concise manner. For example, Ng makes it clear that supervised deep learning is nothing more than a multidimensional curve fitting procedure and that any other representational understandings, such as the common reference to the human biological nervous system, are loose at best. The specialization only requires basic linear algebra knowledge and basic programming knowledge in Python.