M users received messages that Facebook is sunsetting the service on January 19th. Even though it was branded as AI, behind the scenes M relied on humans to answer the most complicated queries. For instance, you could book a table at a restaurant, order flowers or organize your next vacation.

Ben Goertzel, CEO and chief scientist at Sophia’s creator SingularyNET, said he has never pretended Sophia has human capabilities, but SingularityNET chairman David Hanson has described Sophia as “basically alive”.

This piece argues that current approach to AI regulation, pushed by people such as Elon Musk, is a mistake. We’ve already regulated AI it in the past. We just didn’t call it AI. 

This two-part series reviews what the Google Brain team has been up to in 2017. It’s also a good general overview of research advances from the past year.

A new conference, PyImageConf 2018 will be taking place August 26-28th at the San Francisco Hyatt Regency.

This post covers Reinforcement Learning (RL) basics and the popular Q-Learning, SARSA, DQN, and DDPG algorithms.

One of Ray’s goals is to enable practitioners to turn a prototype algorithm that runs on a laptop into a high-performance distributed application that runs efficiently on a cluster.

The recent Deep Learning: A Critical Appraisal paper has led to heated discussions on social media. This follow-up Q&A elaborates on some of the arguments from the paper.

Deep neural networks are typically designed and tuned for the problem at hand. Can we create a unified deep learning model to solve tasks across multiple domains?

Train neural network to code a basic a HTML and CSS website based on a picture of a design mockup. All the code is available on Github.

The goal of the MuJoCo Plugin and Unity Integration package is to combine the best of both worlds: use MuJoCo physics and Unity rendering within the same project.

A Tensorflow implementation of Hinton’s recent Dynamic Routing Between Capsules research paper.

The authors consider the task of program synthesis in the presence of a reward function over the output of programs, where the goal is to find programs with maximal rewards. Their PQT algorithm outperforms the baselines. By adding a program length penalty to the reward function, the authors can synthesize short, human-readable programs.

A Generative Adversarial Network (GAN) for orders made in e-commerce websites. Once trained, the generator in the GAN could generate any number of plausible orders.

Expected policy gradients (EPG) unify stochastic policy gradients (SPG) and deterministic policy gradients (DPG) for reinforcement learning. EPG integrates (or sums) across actions when estimating the gradient, instead of relying only on the action in the sampled trajectory.

The authors propose soft actor-critic, an off-policy actor-critic deep RL algorithm based on the maximum entropy reinforcement learning framework. In this framework, the actor aims to maximize expected reward while also maximizing entropy - that is, succeed at the task while acting as randomly as possible.