On May 16, via a video link, US President Donald Trump "addressed" a conference in Tianjin from Washington and floored the audience with his almost flawless Chinese. Trump highlighted the big leaps made by artificial intelligence or AI, but what impressed the audience more was the US president's tone - his Chinese intonations, inflections and pitch were near perfect. Well, as it transpired, the voice was not really Trump's, after all, but that of an AI-enabled voice technology developed by iFlytek Co Ltd. And, for the record, unlike his granddaughter, Trump hardly knows any Chinese. The iFlytek technology demonstrated its speech synthesis capability - it can produce an unbelievably human-like voice.

Deep learning has created a perfect dichotomy. On the one hand, we have data science practitioners raving about it, and every one and their colleague jumping in to learn and make a career out of this supposedly game-changing technology in analytics. And then there is everyone else wondering what the buzz is all about. With a multitude of analytics technologies projected as the panacea to business' problems, one wonders what this additional'cool thing' is all about. For people on the business side of things, there are no easy avenues to get a simple and intuitive understanding.

I've written a number of times in the past about the growing use of Artificial Intelligence in the drug discovery process, whether it's in terms of identifying molecules for analysis or predicting potential side effects. A recent paper from a team at the University of North Carolina at Chapel Hill Eshelman School of Pharmacy suggests that AI can now go one step further and design new drug molecules from scratch. Their method, which is known as Reinforcement Learning for Structural Evolution (ReLeaSE), consists of two neural networks that the researchers refer to as the teacher and the student. The teacher component of the system understands the syntax and linguistic rules behind the chemical structures of around 1.7 million biologically active molecules. The student component they learn from these in order to propose molecules that can be used in new medicines.

Artificial intelligence (AI) is clearly a growing force in the technology industry. Chatbots and virtual assistants are becoming a key part of new products, and robots are taking center stage at conferences and showing potential in their roles in various industries like retail and manufacturing. Meanwhile, companies such as Google, Microsoft, and Salesforce are integrated AI as an intelligence layer into the entire tech stack. Yes, AI is now having its moment. This isn't the AI that pop culture has conditioned us to expect; it's not sentient robots or Skynet, or even Tony Stark's Jarvis assistant.

The CIFAR deep learning summer school in Toronto has been training the top AI researchers entering or finishing Ph.D. programs since 2005. Over 1,200 students from 60 different countries applied, of which 200 were selected to attend. Attendees represent some of the leading AI labs in the world, Montreal Institute of Learning Algorithms (MILA), University College London, University of Toronto, University of Alberta, Berkeley, NYU, Columbia, CMU, MIT, ETH Zurich, and Stanford. Every year, the school has trained the next generation of top AI researchers which now hold top posts at AI companies like Google, Facebook, Tesla, and Uber. During an intense 10-day period, students learn the tricks of the trade from top AI researchers like deep learning pioneers Yoshua Bengio (MILA), Geoff Hinton (UofT), and reinforcement learning pioneer, Richard Sutton (University of Alberta, Google Deepmind).

Interest in the use of Artificial Intelligence (AI) in schools is growing. More educators are participating in important conversations about it as understanding develops around how AI will impact the work of teachers and schools. In this post I want to add to the conversation by raising some issues and putting forward some questions that I believe are critical. To begin I want to suggest a definition of the term'Artificial Intelligence' or AI as it is commonly known. What do we mean by'Artificial Intelligence'?

A world renowned pioneer in social robotics, Cynthia Breazeal splits her time as an Associate Professor at MIT, where she received her PhD and founded the Personal Robots Group, and Founder and Chief Scientist of Jibo, a personal robotics company with over $85 million in funding. While Breazeal's work has won numerous academic awards, industry accolades, and media attention, she had to fight early skepticism in the 1990s from other experts in robotics and AI. At the time, robots were seen as physical and industrial tools, not social or emotional companions. Her first social robot, Kismet, was unfairly called out in popular press as "useless". Breazeal bucked the trend with a very different vision: "I wanted to create robots with social and emotional intelligence that could work in collaborative partnership with people. In 2-5 years, I see social robots helping families with things that really matter, like education, health, eldercare, entertainment, and companionship." She hopes her work and influence will inspire others to create robots "not only with smarts, but with heart, too."

To learn Chinese in this room, talk to the floating panda head. The Mandarin-speaking avatar zips around a 360-degree restaurant scene in an artificial intelligence-driven instruction program that looks like a giant video game. Rensselaer Polytechnic Institute students testing the technology move inside the 12-foot-high, wrap-around projection to order virtual bean curd from the panda waiter, chat with Beijing market sellers and practice tai chi by mirroring moves of a watchful mentor. "Definitely less anxiety than messing it up with a real human being," says Rahul Divekar, a computer science graduate student working on the project. "So compared to that anxiety, this is a lot more easy."

We show that many machine learning goals, such as improved fairness metrics, can be expressed as constraints on the model's predictions, which we call rate constraints. We study the problem of training non-convex models subject to these rate constraints (or any non-convex and non-differentiable constraints). In the non-convex setting, the standard approach of Lagrange multipliers may fail. Furthermore, if the constraints are non-differentiable, then one cannot optimize the Lagrangian with gradient-based methods. To solve these issues, we introduce the proxy-Lagrangian formulation. This new formulation leads to an algorithm that produces a stochastic classifier by playing a two-player non-zero-sum game solving for what we call a semi-coarse correlated equilibrium, which in turn corresponds to an approximately optimal and feasible solution to the constrained optimization problem. We then give a procedure which shrinks the randomized solution down to one that is a mixture of at most $m+1$ deterministic solutions, given $m$ constraints. This culminates in algorithms that can solve non-convex constrained optimization problems with possibly non-differentiable and non-convex constraints with theoretical guarantees. We provide extensive experimental results enforcing a wide range of policy goals including different fairness metrics, and other goals on accuracy, coverage, recall, and churn.

Robot manipulation is increasingly poised to interact with humans in co-shared workspaces. Despite increasingly robust manipulation and control algorithms, failure modes continue to exist whenever models do not capture the dynamics of the unstructured environment. To obtain longer-term horizons in robot automation, robots must develop introspection and recovery abilities. We contribute a set of recovery policies to deal with anomalies produced by external disturbances as well as anomaly classification through the use of non-parametric statistics with memoized variational inference with scalable adaptation. A recovery critic stands atop of a tightly-integrated, graph-based online motion-generation and introspection system that resolves a wide range of anomalous situations. Policies, skills, and introspection models are learned incrementally and contextually in a task. Two task-level recovery policies: re-enactment and adaptation resolve accidental and persistent anomalies respectively. The introspection system uses non-parametric priors along with Markov jump linear systems and memoized variational inference with scalable adaptation to learn a model from the data. Extensive real-robot experimentation with various strenuous anomalous conditions is induced and resolved at different phases of a task and in different combinations. The system executes around-the-clock introspection and recovery and even elicited self-recovery when misclassifications occurred.