Google AI Blog: October 2016

Supercharging Style Transfer

Wednesday, October 26, 2016

Posted by Vincent Dumoulin^*, Jonathon Shlens and Manjunath Kudlur, Google Brain TeamPasticheparodyonline forumsstyle transferLearning a representation for artistic styleA Learned Representation for Artistic Stylestyle interpolation

Credit: awesome dog role played by Google Brain team office dog Picabo.

using a single style transfer networkA Quick History of Style TransferA Neural Algorithm of Artistic Style

Content image: The Tübingen Neckarfront by Andreas Praefcke, Style painting: “Head of a Clown”, by Georges Rouault.

spectacular

Figure adapted from L. Gatys et al. "A Neural Algorithm of Artistic Style" (2015).

single100 separate style transfer networksOur Contribution: Learning and Combining Multiple Styles

Poppy Field (left) and Impression, Sunrise (right) by Claude Monet. Images from Wikipedia

same style transfer network

Pastiches produced by our single network, trained on 32 varied styles. These pastiches are qualitatively equivalent to those created by single-style networks: Image Credit: (from top to bottom) content photographs by Andreas Praefcke, Rich Niewiroski Jr. and J.-H. Janßen, (from left to right) style paintings by William Glackens, Paul Signac, Georges Rouault, Edvard Munch and Vincent van Gogh.

in real-timeTübingen

Magenta blogTensorFlowNat & Lo’s fantastic video explanationReferences
Image quilting for texture synthesis and transfer Image analogies A Neural Algorithm of Artistic StyleTexture Networks: Feed-forward Synthesis of Textures and Stylized Imageserceptual Losses for Real-Time Style Transfer and Super-Resolution
* This work was done during an internship with the Google Brain Team. Vincent is currently a Ph.D. candidate at MILA, Université de Montréal.^↩

Course Builder now supports scheduling, easier customization and more

Monday, October 17, 2016

Posted by Adam Feldman, Product Manager and Pavel Simakov, Technical Lead, Course Builder TeamCourse Builder v1.11deploying from WindowsImproved student access controls

Student Groups allow you to define which students can see which parts of a course. Want your morning class to see unit 5 and your afternoon class to see unit 6 -- while letting random Internet visitors only see unit 1? Student groups have you covered.

Calendar Triggers can be used to update course or content availability automatically at a specific time. For instance, if your course goes live at midnight on Sunday night, you don't need to be at a computer to make it happen. Or, if you want to unlock a new unit every week, you can set up a trigger to automate the process. Read more about calendar triggers and availability.

Easier visual customizationGraphQL serverNew course explorerDigital WorkshopNPTEL

releasing our first experimental codeforum

Equality of Opportunity in Machine Learning

Friday, October 7, 2016

Posted by Moritz Hardt, Research Scientist, Google Brain Teamsupervised learningdemographic parityEqual Opportunity
equality of opportunityLearn moreBig Picture teaminteractive visualizationfull version of our paperNIPS

Graph-powered Machine Learning at Google

Thursday, October 6, 2016

Posted by Sujith Ravi, Staff Research Scientist, Google ResearchMachine Learninggraph-basedreminders in Inboxsmart messaging in AlloGoogle Photos

Learning with Minimal Supervisiondeep learningsupervised learningunsupervised learningsemi-supervisedjointlyKnowledge GraphfirstthenGraph Learning: How It WorksPhotoReply in Allorelational datasparse

predict a color (“red” or “blue”) for every node in the graphthis research paperSmart Reply feature in Inbox

Semi-supervised learning on a graph

non-convexrealLarge Scale Distributed Semi-Supervised Learning Using Streaming Approximationstreaming algorithmmultimodalLanguage Graphs for Learning Humor emotion labelingSmart Reply for Inboxa priori

Learning emotion associations using graph constructed from word embedding vectors

solvesGraph-based Machine Intelligence in Actionremindersquestion answeringlanguage translationvisual object recognitiondialogue understandingrecent release of Alloannounced this past week on-device Smart Reply capabilityAcknowledgements

How Robots Can Acquire New Skills from Their Shared Experience

Monday, October 3, 2016

Posted by Sergey Levine (Google Brain Team), Timothy Lillicrap (DeepMind), Mrinal Kalakrishnan (X)natural language understandingspeech recognitioncloud roboticspreviously wroteLearning from raw experience with model-free reinforcement learning.successes with the Atari video game systemplaying GoLearning how the world works by interacting with objects.

Top row: robotic arms interacting with common household items.
Bottom row: Predicted future camera images given an initial image and a sequence of actions.

here

Learning with the help of humans.The experiments on learning by trial-and-error were conducted by Shixiang (Shane) Gu and Ethan Holly from the Google Brain team, and Timothy Lillicrap from DeepMind. Work on learning predictive models was conducted by Chelsea Finn from the Google Brain team, and the research on learning from demonstration was conducted by Yevgen Chebotar, Ali Yahya, Adrian Li, and Mrinal Kalakrishnan from X. We would also like to acknowledge contributions by Peter Pastor, Gabriel Dulac-Arnold, and Jon Scholz. Articles about each of the experiments discussed in this blog post can be found below:Deep Reinforcement Learning for Robotic ManipulationShixiang Gu, Ethan Holly, Timothy Lillicrap, Sergey Levine.video
Deep Visual Foresight for Planning Robot MotionChelsea Finn, Sergey Levine.videodataCollective Robot Reinforcement Learning with Distributed Asynchronous Guided Policy SearchAli Yahya, Adrian Li, Mrinal Kalakrishnan, Yevgen Chebotar, Sergey Levine.video
Path Integral Guided Policy SearchYevgen Chebotar, Mrinal Kalakrishnan, Ali Yahya, Adrian Li, Stefan Schaal, Sergey Levine. video