"Large-Scale Deep Learning for Building Intelligent Computer Systems," a Keynote Presentation from Google
13 likes4,684 views
The document discusses the advances and applications of large-scale deep learning, particularly through Google's Brain Team, highlighting the importance of deep neural networks in various domains like speech recognition, image processing, and natural language understanding. It emphasizes the significance of TensorFlow, an open-source software framework designed to facilitate machine learning and enable broader access to deep learning technologies. Future prospects for deep learning are optimistic, with anticipated growth across multiple sectors including robotics and healthcare.
![Image Captions
W __ A young girl
A young girl asleep[Vinyals et al., CVPR 2015]](https://image.slidesharecdn.com/bim-01googledean-160516180952/85/Large-Scale-Deep-Learning-for-Building-Intelligent-Computer-Systems-a-Keynote-Presentation-from-Google-35-320.jpg)
"Large-Scale Deep Learning for Building Intelligent Computer Systems," a Keynote Presentation from Google
- 1. Large-Scale Deep Learning for Intelligent Computer Systems Jeff Dean Google Brain team g.co/brain In collaboration with many other people at Google
- 3. Building Intelligent Products Really hard without understanding Not there yet, but making significant progress
- 4. What do I mean by understanding?
- 5. What do I mean by understanding?
- 6. What do I mean by understanding?
- 7. Outline ● Why deep neural networks? ● Examples of using these for solving real problems ● TensorFlow: software infrastructure for our work (and yours!) ● What are different ways you can get started using these?
- 8. Google Brain project started in 2011, with a focus on pushing state-of-the-art in neural networks. Initial emphasis: ● use large datasets, and ● large amounts of computation to push boundaries of what is possible in perception and language understanding
- 9. Growing Use of Deep Learning at Google Android Apps drug discovery Gmail Image understanding Maps Natural language understanding Photos Robotics research Speech Translation YouTube … many others ... Across many products/areas: # of directories containing model description files Time UniqueProjectDirectories
- 10. The promise (or wishful dream) of Deep Learning Speech Text Search Queries Images Videos Labels Entities Words Audio Features Simple, Reconfigurable, High Capacity, Trainable end-to-end Building Blocks Speech Text Search Queries Images Videos Labels Entities Words Audio Features
- 11. The promise (or wishful dream) of Deep Learning Common representations across domains. Replacing piles of code with data and simple learning algorithms. Would merely be an interesting academic exercise… …if it didn’t work so well!
- 12. Speech Recognition Speech Recognition with Deep Recurrent Neural Networks Alex Graves, Abdel-rahman Mohamed, Geoffrey Hinton Convolutional, Long Short-Term Memory, Fully Connected Deep Neural Networks Tara N. Sainath, Oriol Vinyals, Andrew Senior, Hasim Sak Object Recognition and Detection Going Deeper with Convolutions Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott Reed, Dragomir Anguelov, Dumitru Erhan, Vincent Vanhoucke, Andrew Rabinovich Scalable Object Detection using Deep Neural Networks Dumitru Erhan, Christian Szegedy, Alexander Toshev, Dragomir Anguelov In Research and Industry
- 13. In Research and Industry Machine Translation Sequence to Sequence Learning with Neural Networks Ilya Sutskever, Oriol Vinyals, Quoc V. Le Neural Machine Translation by Jointly Learning to Align and Translate Dzmitry Bahdanau, Kyunghyun Cho, Yoshua Bengio Language Modeling One Billion Word Benchmark for Measuring Progress in Statistical Language Modeling Ciprian Chelba, Tomas Mikolov, Mike Schuster, Qi Ge, Thorsten Brants, Phillipp Koehn, Tony Robinson Parsing Grammar as a Foreign Language Oriol Vinyals, Lukasz Kaiser, Terry Koo, Slav Petrov, Ilya Sutskever, Geoffrey Hinton
- 14. Neural Networks
- 15. “cat” ● A powerful class of machine learning model ● Modern reincarnation of artificial neural networks ● Collection of simple, trainable mathematical functions ● Compatible with many variants of machine learning What is Deep Learning?
- 16. “cat” ● Loosely based on (what little) we know about the brain What is Deep Learning?
- 17. ConvNets
- 18. Learning algorithm While not done: Pick a random training example “(input, label)” Run neural network on “input” Adjust weights on edges to make output closer to “label”
- 19. Learning algorithm While not done: Pick a random training example “(input, label)” Run neural network on “input” Adjust weights on edges to make output closer to “label”
- 20. Plenty of raw data ● Text: trillions of words of English + other languages ● Visual data: billions of images and videos ● Audio: tens of thousands of hours of speech per day ● User activity: queries, marking messages spam, etc. ● Knowledge graph: billions of labelled relation triples ● ... How can we build systems that truly understand this data?
- 21. Important Property of Neural Networks Results get better with more data + bigger models + more computation (Better algorithms, new insights and improved techniques always help, too!)
- 22. What are some ways that deep learning is having a significant impact at Google?
- 23. “How cold is it outside?” Deep Recurrent Neural Network Acoustic Input Text Output Reduced word errors by more than 30% Speech Recognition Google Research Blog - August 2012, August 2015
- 24. ImageNet Challenge Given an image, predict one of 1000 different classes Image credit: www.cs.toronto. edu/~fritz/absps/imagene t.pdf
- 25. The Inception Architecture (GoogLeNet, 2014) Going Deeper with Convolutions Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott Reed, Dragomir Anguelov, Dumitru Erhan, Vincent Vanhoucke, Andrew Rabinovich ArXiv 2014, CVPR 2015
- 26. Team Year Place Error (top-5) XRCE (pre-neural-net explosion) 2011 1st 25.8% Supervision (AlexNet) 2012 1st 16.4% Clarifai 2013 1st 11.7% GoogLeNet (Inception) 2014 1st 6.66% Andrej Karpathy (human) 2014 N/A 5.1% BN-Inception (Arxiv) 2015 N/A 4.9% Inception-v3 (Arxiv) 2015 N/A 3.46% Neural Nets: Rapid Progress in Image Recognition ImageNet challenge classification task
- 28. Good Generalization Both recognized as “meal”
- 29. Sensible Errors
- 30. “ocean” Deep Convolutional Neural Network Your Photo Automatic Tag Search personal photos without tags. Google Photos Search Google Research Blog - June 2013
- 35. Image Captions W __ A young girl A young girl asleep[Vinyals et al., CVPR 2015]
- 36. Model: A close up of a child holding a stuffed animal. Human: A young girl asleep on the sofa cuddling a stuffed bear. Model: A baby is asleep next to a teddy bear.
- 38. Combined Vision + Translation
- 39. What do you want in a machine learning system? ● Ease of expression: for lots of crazy ML ideas/algorithms ● Scalability: can run experiments quickly ● Portability: can run on wide variety of platforms ● Reproducibility: easy to share and reproduce research ● Production readiness: go from research to real products
- 40. TensorFlow: Second Generation Deep Learning System
- 41. http://tensorflow.org/ and https://github.com/tensorflow/tensorflow If we like it, wouldn’t the rest of the world like it, too? Open sourced single-machine TensorFlow on Monday, Nov. 9th, 2015 ● Flexible Apache 2.0 open source licensing ● Updates for distributed implementation coming soon
- 45. Motivations DistBelief (1st system) was great for scalability, and production training of basic kinds of models Not as flexible as we wanted for research purposes Better understanding of problem space allowed us to make some dramatic simplifications
- 46. TensorFlow: Expressing High-Level ML Computations ● Core in C++ ○ Very low overhead Core TensorFlow Execution System CPU GPU Android iOS ...
- 47. TensorFlow: Expressing High-Level ML Computations ● Core in C++ ○ Very low overhead ● Different front ends for specifying/driving the computation ○ Python and C++ today, easy to add more Core TensorFlow Execution System CPU GPU Android iOS ...
- 48. TensorFlow: Expressing High-Level ML Computations ● Core in C++ ○ Very low overhead ● Different front ends for specifying/driving the computation ○ Python and C++ today, easy to add more Core TensorFlow Execution System CPU GPU Android iOS ... C++ front end Python front end ...
- 49. MatMul Add Relu biases weights examples labels Xent Graph of Nodes, also called Operations or ops. Computation is a dataflow graph
- 50. with tensors MatMul Add Relu biases weights examples labels Xent Edges are N-dimensional arrays: Tensors Computation is a dataflow graph
- 51. with state Add Mul biases ... learning rate −=... 'Biases' is a variable −= updates biasesSome ops compute gradients Computation is a dataflow graph
- 52. Device A Device B distributed Add Mul biases ... learning rate −=... Devices: Processes, Machines, GPUs, etc Computation is a dataflow graph
- 53. Automatically runs models on range of platforms: from phones ... to single machines (CPU and/or GPUs) … to distributed systems of many 100s of GPU cards TensorFlow: Expressing High-Level ML Computations
- 54. Mobile and Embedded Deployment Desires ● Execution efficiency ● Low power consumption ● Modest size requirements
- 55. Quantization: Using Low Precision Integer Math ● Train using 32-bit floats, and after training, convert parameters to quantized 8-bit integer representations ● We have used this in many different applications: ○ Very minor losses in overall model accuracy. E.g.: ■ ~77% top-1 accuracy for image model with 32-bit float, ~76% top-1 accuracy with 8-bit quantized integers ○ 8-bit math gives close to 4X speedup and 4X reduction in model size ○ Saves considerable power, as well
- 56. gemmlowp Support for this open-sourced in gemmlowp package: https://github.com/google/gemmlowp Efficient GEMM implementations for ARM and x86 Ongoing performance work to make it even better
- 57. TensorFlow and Quantized Models Support for quantized integer kernels Automated tool coming shortly to tensorflow/contrib/quantization/quantize_graph Converts TensorFlow model/graph trained using 32-bit floats ● Emits new graph and associated parameter checkpoint ● Uses quantized ops where equivalent ops exist ● Falls back to float when no equivalent quantized op exists
- 58. TensorFlow and Mobile Execution ● Android support already there ○ Example app in TensorFlow GitHub repository under: ■ tensorflow/examples/android/... ● iOS support coming shortly: ○ https://github. com/tensorflow/tensorflow/issues/16 ○ https://github.com/tensorflow/tensorflow/pull/1631
- 59. To Learn More ● Attend Pete Warden’s talk tomorrow (Tuesday, May 3), 10:30 to 11:15 AM “TensorFlow: Enabling Mobile and Embedded Machine Intelligence” http://www.embedded-vision.com/summit/tensorflow-enabling-mobile-and-embedded-machine-intelligence
- 60. How Can You Get Started with Machine Learning? Four ways, with varying complexity: (1) Use a Cloud-based API (Vision, Speech, etc.) (2) Run your own pretrained model (3) Use an existing model architecture, and retrain it or fine tune on your dataset (4) Develop your own machine learning models for new problems More flexible, but more effort required
- 61. (1) Use Cloud-based APIs cloud.google.com/translate cloud.google.com/speech cloud.google.com/vision cloud.google.com/text
- 62. (1) Use Cloud-based APIs cloud.google.com/translate cloud.google.com/speech cloud.google.com/vision cloud.google.com/text
- 63. Google Cloud Vision API https://cloud.google.com/vision/
- 64. Google Cloud Vision API demo
- 65. (2) Using a Pre-trained Image Model with TensorFlow www.tensorflow.org/tutorials/image_recognition/index.html
- 66. Using a Pre-trained Image Model with TensorFlow on Android https://github.com/tensorflow/tensorflow/tree/master/tensorflow/examples/android
- 67. (3) Training a Model on Your Own Image Data www.tensorflow.org/versions/master/how_tos/image_retraining/index.html
- 68. (4) Develop your own machine learning models https://www.tensorflow.org/versions/master/get_started/basic_usage.html
- 69. What Does the Future Hold? Deep learning usage will continue to grow and accelerate: ● Across more and more fields and problems: ○ robotics, self-driving vehicles, ... ○ health care ○ video understanding ○ dialogue systems ○ personal assistance ○ ...
- 70. Combining Vision with Robotics “Deep Learning for Robots: Learning from Large-Scale Interaction”, Google Research Blog, March, 2016 “Learning Hand-Eye Coordination for Robotic Grasping with Deep Learning and Large-Scale Data Collection”, Sergey Levine, Peter Pastor, Alex Krizhevsky, & Deirdre Quillen, Arxiv, arxiv.org/abs/1603.02199
- 71. Conclusions Deep neural networks are making significant strides in understanding: In speech, vision, language, search, … If you’re not considering how to use deep neural nets to solve your vision or understanding problems, you almost certainly should be Pre-trained models or pre-trained APIs are a low overhead way of starting to explore TensorFlow makes it easy for everyone to experiment with these techniques ● Highly scalable design allows faster experiments, accelerates research ● Easy to share models and to publish code to give reproducible results ● Ability to go from research to production within same system
- 72. Further Reading ● Le, Ranzato, Monga, Devin, Chen, Corrado, Dean, & Ng. Building High-Level Features Using Large Scale Unsupervised Learning, ICML 2012. research.google. com/archive/unsupervised_icml2012.html ● Dean, et al., Large Scale Distributed Deep Networks, NIPS 2012, research.google. com/archive/large_deep_networks_nips2012.html. ● Sutskever, Vinyals, & Le, Sequence to Sequence Learning with Neural Networks, NIPS, 2014, arxiv.org/abs/1409.3215. ● Vinyals, Toshev, Bengio, & Erhan. Show and Tell: A Neural Image Caption Generator. CVPR 2015. arxiv.org/abs/1411.4555 ● Szegedy, Vanhoucke, Ioffe, Shlens, & Wojna. Rethinking the Inception Architecture for Computer Vision. arxiv.org/abs/1512.00567 ● TensorFlow white paper, tensorflow.org/whitepaper2015.pdf (clickable links in bibliography) research.google.com/people/jeff research.google.com/pubs/MachineIntelligence.html Questions?