Research Blog: Google Brain

Wide & Deep Learning: Better Together with TensorFlow

Wednesday, June 29, 2016

Posted by Heng-Tze Cheng, Senior Software Engineer, Google ResearchWide & Deep Learningcategorical features

TF.Learn APILinear ModelsWide & Deep Learningresearch paperHow Wide & Deep Learning works.FoodIO^*queryitemlabelThe Wide model.wideAND(query="fried chicken", item="chicken and waffles")AND(query="fried chicken", item="chicken fried rice")memorizing

The Deep model.deepgeneralize

Combining Wide and Deep models.

query="fried chicken"item="chicken fried rice"Wider. Deeper. Together.GitHubLinear ModelsWide & Deep LearningAcknowledgement
* For illustration only. FoodIO is not a real app.^↩

Bringing Precision to the AI Safety Discussion

Tuesday, June 21, 2016

Posted by Chris Olah, Google ResearchConcrete Problems in AI Safety

Avoiding Negative Side Effects: How can we ensure that an AI system will not disturb its environment in negative ways while pursuing its goals, e.g. a cleaning robot knocking over a vase because it can clean faster by doing so?

Avoiding Reward Hacking: How can we avoid gaming of the reward function? For example, we don’t want this cleaning robot simply covering over messes with materials it can’t see through.

Scalable Oversight: How can we efficiently ensure that a given AI system respects aspects of the objective that are too expensive to be frequently evaluated during training? For example, if an AI system gets human feedback as it performs a task, it needs to use that feedback efficiently because asking too often would be annoying.

Safe Exploration: How do we ensure that an AI system doesn’t make exploratory moves with very negative repercussions? For example, maybe a cleaning robot should experiment with mopping strategies, but clearly it shouldn’t try putting a wet mop in an electrical outlet.

Robustness to Distributional Shift: How do we ensure that an AI system recognizes, and behaves robustly, when it’s in an environment very different from its training environment? For example, heuristics learned for a factory workfloor may not be safe enough for an office.

in the paper

Research at Google and ICLR 2016

Sunday, May 01, 2016

Posted by Dumitru Erhan, Gentleman Scientist4th International Conference on Learning RepresentationsMachine LearningNeural NetworksDeep LearningGoogle Brain teamGoogle DeepMindblueOrganizing CommitteeProgram ChairsSamy Bengio, Brian KingsburyArea Chairs include:John Platt, Tara SanaithOral SessionsNeural Programmer-Interpreters(Best Paper Award Recipient)Scott Reed, Nando de FreitasNet2Net: Accelerating Learning via Knowledge Transfer Tianqi Chen, Ian Goodfellow, Jon ShlensConference Track Posters
Prioritized Experience Replay Tom Schau, John Quan, Ioannis Antonoglou, David Silver
Reasoning about Entailment with Neural Attention Tim Rocktäschel, Edward Grefenstette, Karl Moritz Hermann, Tomáš Kočiský, Phil BlunsomNeural Programmer: Inducing Latent Programs With Gradient Descent Arvind Neelakantan, Quoc Le, Ilya Sutskever
MuProp: Unbiased Backpropagation For Stochastic Neural NetworksShixiang Gu, Sergey Levine, Ilya Sutskever, Andriy Mnih
Multi-Task Sequence to Sequence Learning Minh-Thang Luong, Quoc Le, Ilya Sutskever, Oriol Vinyals, Lukasz KaiserA Test of Relative Similarity for Model Selection in Generative Models Eugene Belilovsky, Wacha Bounliphone, Matthew Blaschko, Ioannis Antonoglou, Arthur GrettonContinuous control with deep reinforcement learningTimothy Lillicrap, Jonathan Hunt, Alexander Pritzel, Nicolas Heess, Tom Erez, Yuval Tassa, David Silver, Daan WierstraPolicy DistillationAndrei Rusu, Sergio Gomez, Caglar Gulcehre, Guillaume Desjardins, James Kirkpatrick, Razvan Pascanu, Volodymyr Mnih, Koray Kavukcuoglu, Raia HadsellNeural Random-Access MachinesKarol Kurach, Marcin Andrychowicz, Ilya SutskeverVariable Rate Image Compression with Recurrent Neural Networks George Toderici, Sean O'Malley, Damien Vincent, Sung Jin Hwang, Michele Covell, Shumeet Baluja, Rahul Sukthankar, David MinnenOrder Matters: Sequence to Sequence for SetsOriol Vinyals, Samy Bengio, Manjunath Kudlur
Grid Long Short-Term MemoryNal Kalchbrenner, Alex Graves, Ivo Danihelka
Neural GPUs Learn AlgorithmsLukasz Kaiser, Ilya SutskeverACDC: A Structured Efficient Linear LayerMarcin Moczulski, Misha Denil, Jeremy Appleyard, Nando de FreitasWorkshop Track Posters
Revisiting Distributed Synchronous SGD Jianmin Chen, Rajat Monga, Samy Bengio, Rafal JozefowiczBlack Box Variational Inference for State Space Models Evan Archer, Il Memming Park, Lars Buesing, John Cunningham, Liam Paninski
A Minimalistic Approach to Sum-Product Network Learning for Real Applications Viktoriya Krakovna, Moshe LooksEfficient Inference in Occlusion-Aware Generative Models of Images Jonathan Huang, Kevin Murphy
Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning Christian Szegedy, Sergey Ioffe, Vincent VanhouckeDeep Autoresolution Networks Gabriel Pereyra, Christian SzegedyLearning visual groups from co-occurrences in space and time Phillip Isola, Daniel Zoran, Dilip Krishnan, Edward H. AdelsonAdding Gradient Noise Improves Learning For Very Deep Networks Arvind Neelakantan, Luke Vilnis, Quoc V. Le, Ilya Sutskever, Lukasz Kaiser, Karol Kurach, James MartensAdversarial Autoencoders Alireza Makhzani, Jonathon Shlens, Navdeep Jaitly, Ian GoodfellowGenerating Sentences from a Continuous Space Samuel R. Bowman, Luke Vilnis, Oriol Vinyals, Andrew M. Dai, Rafal Jozefowicz, Samy Bengio

Announcing TensorFlow 0.8 – now with distributed computing support!

Wednesday, April 13, 2016

Posted by Derek Murray, Software EngineerTensorFlowgRPCGoogle Cloud Machine Learningdistributed trainernception image classification

TensorFlow can speed up Inception training by a factor of 56, using 100 GPUs.

Kubernetesspeed up inference using TensorFlow Serving on Kubernetesnew librariesDistBelief synchronous SGD with backup workerson GitHub

TensorFlow training libraries - Jianmin Chen, Matthieu Devin, Sherry Moore and Sergio Guadarrama

TensorFlow core - Zhifeng Chen, Manjunath Kudlur and Vijay Vasudevan

Testing - Shanqing Cai

Inception model architecture - Christian Szegedy, Sergey Ioffe, Vincent Vanhoucke, Jonathon Shlens and Zbigniew Wojna

Project management - Amy McDonald Sandjideh

Engineering leadership - Jeff Dean and Rajat Monga

Machine Learning in the Cloud, with TensorFlow

Wednesday, March 23, 2016

Posted by Slaven Bilac, Software Engineer, Google Researchspeech recognition in the Google appsearch in Google PhotosSmart Reply feature in Inbox by GmailTensorFlowGCP NEXT 2016announced the alpha releaseCloud Machine Learning

TensorFlowCloud Machine LearningCloud Machine LearningTensorFlowdeep learningCloud Translate APICloud Vision APICloud Speech APISDCADeCAFGoogle ResearchGoogle Cloud PlatformGoogle Cloud Platform blog

Train your own image classifier with Inception in TensorFlow

Wednesday, March 09, 2016

Posted by Jon Shlens, Senior Research Scientistto classify images with TensorFlowImageNethuman performance

Image via Wikipedia

retrieverspotted salamanderfire salamanderherpetologyyou

Images via Wikipedia

Inception-v3Rethinking the Inception Architecture for Computer Vision

Schematic diagram of Inception-v3

ILSVRC 20122015 ImageNet Challenge

Exploration of different variants of this model architecture in order to improve the image classification system.

Comparison of optimization algorithms and hardware setups for training this model faster or to a higher degree of predictive performance.

Retraining/fine-tuning the Inception-v3 model on a distinct image classification task or as a component of a larger network tasked with object detection or multi-modal learning.

transfer learningobject detectionzero-shot learningimage captioningvideo analysiswe are releasing libraries and code for training Inception-v3

Training an Inception-v3 model with synchronous updates across multiple GPUs.

Employing batch normalization to speed up training of the model.

Leveraging many distortions of the image to augment model training.

Releasing a new (still experimental) high-level language for specifying complex model architectures, which we call TensorFlow-Slim.

Demonstrating how to perform transfer learning by taking a pre-trained Inception-v3 model and fine-tuning it for another task.

instructionstrainevaluatefine-tune

Model Architecture – Christian Szegedy, Sergey Ioffe, Vincent Vanhoucke, Jon Shlens and Zbigniew Wojna

Systems Infrastructure – Sherry Moore, Martin Wicke, David Andersen, Matthieu Devin, Manjunath Kudlur and Nishant Patil

TensorFlow-Slim – Sergio Guadarrama and Nathan Silberman

Model Visualization – Fernanda Viégas, Martin Wattenberg and James Wexler

Deep Learning for Robots: Learning from Large-Scale Interaction

Tuesday, March 08, 2016

Posted by Sergey Levine, Research Scientistthis robotKAISTDARPA robotics challengeconvolutional neural networkrecent work by Pinto and GuptaThe research on robotic hand-eye coordination and grasping was conducted by Sergey Levine, Peter Pastor, Alex Krizhevsky, and Deirdre Quillen, with special thanks to colleagues at Google Research and X who've contributed their expertise and time to this research. An early preprint is available on arXiv

Exploring the Intersection of Art and Machine Intelligence

Monday, February 22, 2016

Posted by Mike Tyka, Software Engineerpublished a storystrange, wonderful and oddly compelling imageswe released the source codewitnessed a tremendous interestAmanda PetersonMemo AktenSamim WinigerKyle McDonaldGene Kogan

“GCHQ”, 2015, Memo Akten, used with permission.

A Neural Algorithm of Artistic StyleexperimentingTwitter botsexplorationsexperiments

The style transfer algorithm crosses a photo with a painting style; for example Neil deGrasse Tyson in the style of Kadinsky’s Jane Rouge Bleu. Photo by Guillaume Piolle, used with permission.

GitXivfurther

“Saxophone dreams” - Mike Tyka.

Machine learning and arta similar programArtificial IntelligenceGray Area Foundation for the Arts

Friday Feb 26th: DeepDream: The Art of Neural Networks, an exhibit consisting of 29 neural network generated artworks, created by artists at Google and from around the world. The works will be auctioned, with all proceeds going to the Gray Area Foundation, which has been active in supporting the intersection between arts and technology for over 10 years.

On Saturday Feb 27th: Art and Machine Learning Symposium, an open one-day symposium on Machine Learning and Art, aiming to bring together the neural network and the creative coding communities to exchange ideas, learn and discuss. Videos of all the talks will be posted online after the event.

Running your models in production with TensorFlow Serving

Tuesday, February 16, 2016

Posted by Noah Fiedel, Software Engineerspeech recognition in the Google appSmart Reply in Inboxsearch in Google Photosnew and interesting challengesTensorFlow ServingTensorFlow

TensorFlow Serving

model lifecycle management

experiments with multiple algorithms

efficient use of GPU resources

run experiments

validation

remote procedure callgRPC

machinequestionsfeature requestsgithub.com/tensorflow/servingtutorial@googleresearch+ResearchatGoogleJeff DeanGCP Next 2016

Teach Yourself Deep Learning with TensorFlow and Udacity

Thursday, January 21, 2016

Posted by Vincent Vanhoucke, Principal Research ScientistDeep learningTensorFlowrecently releasedGitHubenthusiasts around the globeDeep Learning CourseUdacity

convolutional networksrecurrent neural networksUdacity blog postregister for the course

How to Classify Images with TensorFlow

Monday, December 07, 2015

Posted by Pete Warden, Software EngineerJetpacdeep learningTensorFlowan annual competitionImagenetsource codelabel_imageREADME

Image via Wikipedia

the TensorFlow Inception tutorial

Image via U.S History Images

nematodefound that even someone who was trained at the job could only achieve a slightly-better 5.1% error doing the same task manuallythis simple tutorial on recognizing hand-drawn digits from the MNIST data set

NIPS 2015 and Machine Learning Research at Google

Sunday, December 06, 2015

Posted by Sanjiv Kumar, Research Scientist29^th Annual Conference on Neural Information Processing Systemsmachine learningdeep learningbluePROGRAM ORGANIZERSGeneral ChairsCorinna Cortes, Neil D. LawrenceProgram Committee includes:Samy Bengio, Gal Chechik, Ian Goodfellow, Shakir Mohamed, Ilya SutskeverORAL SESSIONSLearning Theory and Algorithms for Forecasting Non-stationary Time SeriesVitaly Kuznetsov, Mehryar MohriSPOTLIGHT SESSIONSDistributed Submodular Cover: Succinctly Summarizing Massive DataBaharan Mirzasoleiman, Amin Karbasi, Ashwinkumar Badanidiyuru, Andreas KrauseSpatial Transformer NetworksMax Jaderberg, Karen Simonyan, Andrew Zisserman, Koray KavukcuogluPointer NetworksOriol Vinyals, Meire Fortunato, Navdeep JaitlyStructured Transforms for Small-Footprint Deep LearningVikas Sindhwani, Tara Sainath, Sanjiv KumarSpherical Random Features for Polynomial KernelsJeffrey Pennington, Felix Yu, Sanjiv KumarPOSTERSLearning to Transduce with Unbounded MemoryEdward Grefenstette, Karl Moritz Hermann, Mustafa Suleyman, Phil BlunsomDeep Knowledge TracingChris Piech, Jonathan Bassen, Jonathan Huang, Surya Ganguli, Mehran Sahami, Leonidas Guibas, Jascha Sohl-DicksteinHidden Technical Debt in Machine Learning SystemsD Sculley, Gary Holt, Daniel Golovin, Eugene Davydov, Todd Phillips, Dietmar Ebner, Vinay Chaudhary, Michael Young, Jean-Francois Crespo, Dan DennisonGrammar as a Foreign LanguageOriol Vinyals, Lukasz Kaiser, Terry Koo, Slav Petrov, Ilya Sutskever, Geoffrey HintonStochastic Variational Information MaximisationShakir Mohamed, Danilo RezendeEmbedding Inference for Structured Multilabel PredictionFarzaneh Mirzazadeh, Siamak Ravanbakhsh, Bing Xu, Nan Ding, Dale SchuurmansOn the Convergence of Stochastic Gradient MCMC Algorithms with High-Order IntegratorsChangyou Chen, Nan Ding, Lawrence CarinSpectral Norm Regularization of Orthonormal Representations for Graph TransductionRakesh Shivanna, Bibaswan Chatterjee, Raman Sankaran, Chiranjib Bhattacharyya, Francis Bach Differentially Private Learning of Structured Discrete DistributionsIlias Diakonikolas, Moritz Hardt, Ludwig SchmidtNearly Optimal Private LASSOKunal Talwar, Li Zhang, Abhradeep ThakurtaLearning Continuous Control Policies by Stochastic Value GradientsNicolas Heess, Greg Wayne, David Silver, Timothy Lillicrap, Tom Erez, Yuval TassaGradient Estimation Using Stochastic Computation GraphsJohn Schulman, Nicolas Heess, Theophane Weber, Pieter AbbeelScheduled Sampling for Sequence Prediction with Recurrent Neural NetworksSamy Bengio, Oriol Vinyals, Navdeep Jaitly, Noam ShazeerTeaching Machines to Read and ComprehendKarl Moritz Hermann, Tomas Kocisky, Edward Grefenstette, Lasse Espeholt, Will Kay, Mustafa Suleyman, Phil BlunsomBayesian dark knowledgeAnoop Korattikara, Vivek Rathod, Kevin Murphy, Max WellingGeneralization in Adaptive Data Analysis and Holdout ReuseCynthia Dwork, Vitaly Feldman, Moritz Hardt, Toniann Pitassi, Omer Reingold, Aaron RothSemi-supervised Sequence LearningAndrew Dai, Quoc LeNatural Neural NetworksGuillaume Desjardins, Karen Simonyan, Razvan Pascanu, Koray KavukcuogluRevenue Optimization against Strategic BuyersAndres Munoz Medina, Mehryar MohriWORKSHOPSFeature Extraction: Modern Questions and ChallengesDmitry Storcheus, Afshin Rostamizadeh, Sanjiv KumarJeffery Pennington, Vikas SindhwaniNIPS Time Series WorkshopMehryar MohriCorinna CortesNonparametric Methods for Large Scale Representation LearningAmr AhmedMachine Learning for Spoken Language Understanding and InteractionLarry HeckAdaptive Data AnalysisMoritz HardtDeep Reinforcement LearningDavid SilverSergey LevineAdvances in Approximate Bayesian InferenceShakir MohamedDanilo RezendeCognitive Computation: Integrating Neural and Symbolic Approaches Ramanathan V. Guha, Geoffrey Hinton, Greg WayneTransfer and Multi-Task Learning: Trends and New Perspectives Mehryar MohriAndres Munoz MedinaLearning and privacy with incomplete data and weak supervisionFelix YuAlexander Blocker, Krzysztof Choromanski, Sanjiv KumarNando de FreitasBlack Box Learning and InferenceAli EslamiGeoff HintonQuantum Machine LearningHartmut NevenBayesian Nonparametrics: The Next GenerationAmr AhmedBayesian Optimization: Scalability and FlexibilityNando de FreitasReasoning, Attention, Memory (RAM)Alex Graves, Ilya SutskeverExtreme Classification 2015: Multi-class and Multi-label Learning in Extremely Large Label SpacesMehryar MohriSamy BengioSamy BengioMachine Learning SystemsJeff DeanSYMPOSIABrains, Mind and MachinesGeoffrey Hinton, Demis HassabisDeep Learning SymposiumSamy Bengio, Phil Blunsom, Nando De Freitas, Ilya Sutskever, Andrew ZissermanMax Jaderberg, Sergey Ioffe, Alexander GravesAlgorithms Among Us: The Societal Impacts of Machine LearningShane LeggTUTORIALSNIPS 2015 Deep Learning TutorialGeoffrey E. Hinton, Yoshua Bengio, Yann LeCun
Large-Scale Distributed Systems for Training Neural NetworksJeff Dean, Oriol Vinyals

TensorFlow - Google’s latest machine learning system, open sourced for everyone

Monday, November 09, 2015

Posted by Jeff Dean, Senior Google Fellow, and Rajat Monga, Technical LeadDistBeliefneural networksconcepts like “cat”the Google appin Google PhotosLarge Scale Visual Recognition Challenge in 2014automated image captioningDeepDreamTensorFlowwhitepaper examplesauto-differentiation

Inspecting a model with TensorBoard, the visualization tool

example model architecturessignals derived from deep neural networksmagic features of tomorrowwww.tensorflow.org

Google Research Blog

Wide & Deep Learning: Better Together with TensorFlow

Bringing Precision to the AI Safety Discussion

Research at Google and ICLR 2016

Announcing TensorFlow 0.8 – now with distributed computing support!

Machine Learning in the Cloud, with TensorFlow

Train your own image classifier with Inception in TensorFlow

Deep Learning for Robots: Learning from Large-Scale Interaction

Exploring the Intersection of Art and Machine Intelligence

Running your models in production with TensorFlow Serving

Teach Yourself Deep Learning with TensorFlow and Udacity

How to Classify Images with TensorFlow

NIPS 2015 and Machine Learning Research at Google

TensorFlow - Google’s latest machine learning system, open sourced for everyone

Labels

Archive

Feed

Company-wide

Products

Developers