At its F8 developer conference in San Francisco today, Facebook demonstrated its latest artificial intelligence (AI) research efforts. Not surprisingly, they are about video. Video implies a whole bunch of individual images put together. So it logically flows from Facebook's progress around object recognition and image caption generation using AI. "You can imagine us building image search on steroids," Joaquin Quiñonero Candela, Facebook's director of Applied Machine Learning, said onstage today.

Machine learning is considered by many to be tech's next frontier. To help researchers and developers who might otherwise not have access to machine learning's benefits, Google on Wednesday announced a new distributed version of TensorFlow, the artificial-intelligence engine the Web giant uses to add capabilities such as speech and object recognition to its products. The new version of TensorFlow will let researchers perform large-scale machine learning across hundreds of computers, shrinking the training process for some models from weeks to hours. Google already uses machine learning algorithms to deliver search results, help translate languages and identify objects in photos. Google open-sourced TensorFlow in November, allowing developers to build on its framework, contribute source code and provide feedback.

The battle for the future of computing is a battle to bring artificial intelligence to the mainstream – and Google is quietly overhauling a machine learning tool used to improve some of its most popular services including Google Translate and Google Photos. TensorFlow can be used to help teach computers how to process data in ways similar to how the human brain handles information. It is also open source, meaning Google has published and shared the code online so that external developers can use and improve it. The latest version, released by Google on Wednesday, adds a feature many TensorFlow users have asked for since the tool made its public debut in late 2015: the ability to operate on multiple devices. Instead of being limited by the processing capabilities of a single computer, it can use distributed networks to handle more complicated tasks – as if TensorFlow will now be able to use many brain cells instead of being confined to just one.

Google has employed machine learning on its Calendar application in an effort to help its users better keep track of, and complete, long-term goals. Users simply add a personal goal -- like hitting the gym three times a week, for example -- and Google Calendar will help them find the time and stick to it. Setting a goal requires the user to answer a few questions, specifying duration and times. From there Calendar will look at the user's schedule and find the best windows to schedule time to help complete the goal. It's another example of a major technology company using machine learning -- the concept of pattern recognition and computational learning theory -- to make its users' lives easier.

Last week, machine learning took a big leap forward when Google's AlphaGo, a machine algorithm, beat the world champion, Lee Sedol, in the game Go. An ancient Chinese board game that dates back nearly 3,000 years, Go is played on a 19-by-19 square grid, with each player trying capture the opponent's territory. Unlike Western chess that has around 40 turns in a game, Go can go up to 200. It was thought it would take at least another 10 years before a machine could beat a human in Go. What's most remarkable is that AlphaGo turns out to be a machine that can improve its performance every day, without the direct supervision of a human programmer.

Consider how Google generates answers to questions. For the past several years, Google has relied on its Knowledge Graph, a large-scale effort to collect information about objects in the digital world and map relationships among things. Ask for information about "Leonardo da Vinci" and Google displays a panel next to the search results list summarizing his basic biographical information. Included are thumbnail links to some of his most significant pictures as well as several important artists that other people have searched for. Behind the scenes is a semantic infrastructure.

Google today announced the launch of version 0.8 of TensorFlow, its open source library for doing the hard computation work that makes machine learning possible. Normally, a small point update like this wouldn't be all that interesting, but with this version, TensorFlow can now run the training processes for building machine learning models across hundreds of machines in parallel. This means training complex models using TensorFlow won't take days or even weeks, but often only hours. The company says distributed computing has long been one of the most requested features for TensorFlow and with this, Google is essentially making the technology that powers much of its recently announced hosted Google Cloud Machine Learning platform available to all developers. Google says it's using the gRPC library to manage all of these machines.

According to OECD predictions, exceeding budgets on health spending remains an issue for OECD countries. Maintaining today's healthcare systems and funding future medical advances will be difficult without major reforms. Public expenditure on health and long-term care in OECD countries is set to increase from around 6% of Gross Domestic Product (GDP) today to almost 9% of GDP in 2030 and to 14% by 2060. This will be the harsh reality unless governments and private companies change the structure of how healthcare is funded. In countries with private health insurance, certain treatments such as cancer care are so expensive that only the privileged with good insurance plans can afford them.

Cameras and audio equipment are getting better all the time, but mostly through brute force: more pixels, more sensors, and better post-processing. Mammalian eyes and ears beat them handily when it comes to efficiency and the ability to only focus on what's interesting or important. Neuromorphic engineers, who try to mimic the strengths of biological systems in manmade ones, have made big strides in recent years, especially with vision. Researchers have made machine-vision systems that only take pictures of moving objects, for example. Instead of taking many images at a steady, predetermined rate, these kinds of cameras monitor for changes in a scene and only record those.

Recurrent Neural Networks (RNN) make use of sequential information. Unlike traditional neural networks, where it is assumed that all inputs and outputs are independent of one another, RNNs are reliant on preceding computations and what has previously been calculated. RNNs can be conceptualized as a neural network unrolled over time. Where you would have different layers in a regular neural network, you apply the same layer to the input at each timestep in an RNN, using the output, i.e. the state of the previous timestep as input. Connections between entities in a RNN form a directed cycle, creating a sort of internal memory, that helps the model leverage long chains of dependencies.