tensorflow dev summit 2017 요약

34
tf.Session() agenda = tf.summary.merge(dev.summit, 2017) sess.run(agenda)

Upload: jin-joong-kim

Post on 12-Apr-2017

1.151 views

Category:

Software


9 download

TRANSCRIPT

Page 1: TensorFlow Dev Summit 2017 요약

tf.Session()

agenda = tf.summary.merge(dev.summit, 2017)

sess.run(agenda)

Page 2: TensorFlow Dev Summit 2017 요약

KeynoteDistBelief -> TensorFlow

제품 뿐 아니라 연구 레벨에서까지 광범위하게 사용 가능

RNN 등의 복잡한 모델까지 커버 가능하다

CPU, GPU, TPU, Android, iOS, Raspberry Pi 등 다양한 플랫폼 지원

구Ú 클라우드에서도 사용 가능

Python, C++, Java, Go, Haskell, R ...

텐서보드 짱짱맨

Page 3: TensorFlow Dev Summit 2017 요약

Keynote

Page 4: TensorFlow Dev Summit 2017 요약

Keynote

Page 5: TensorFlow Dev Summit 2017 요약

Keynote

Page 6: TensorFlow Dev Summit 2017 요약

Keynote - TensorFlow 1.058배 빨라짐

XLA (Accelerated Linear Algebra), JIT

하이레벨 API 제공

Layers

Keras

Estimator

Canned Estimators

API 안정화

Page 7: TensorFlow Dev Summit 2017 요약

Keynote - TensorFlow 1.0

Page 8: TensorFlow Dev Summit 2017 요약

Keynote - TensorFlow 1.0

Page 9: TensorFlow Dev Summit 2017 요약

TensorBoardScalars

Images

Audio

Graphs

Distributions

Histogram

Embeddings

Page 10: TensorFlow Dev Summit 2017 요약

TensorBoard

Page 11: TensorFlow Dev Summit 2017 요약

Future for TensorBoardTensorFlow Debugger Integration

Plugins

Org-scale TensorBoard

Page 12: TensorFlow Dev Summit 2017 요약

TensorFlow at DeepMindChoosing a Platform

Flexibility

Usability

Scalability

Performance

Production Readiness

Page 13: TensorFlow Dev Summit 2017 요약

TensorFlow at DeepMindData Center Cooling

Gorila (DistBelief -> TensorFlow)

AlphaGo

WaveNet

Text to Speech

Music Generation

Learning to Learn

Page 14: TensorFlow Dev Summit 2017 요약

TensorFlow at DeepMind

Page 15: TensorFlow Dev Summit 2017 요약

TensorFlow at DeepMind

Page 16: TensorFlow Dev Summit 2017 요약

TensorFlow at DeepMind

Page 17: TensorFlow Dev Summit 2017 요약

Mobile and EmbeddedARM

CEVA

Movidius

IBM Power Systems

Intel

Qualcomm

Page 18: TensorFlow Dev Summit 2017 요약

Mobile and EmbeddedAndroid

tensor�ow/examples/android

TF Classify, TF Detect, TF Stylize

iOS

tensor�ow/contrib/ios_examples

Inception image labeling

Raspberry Pi

tensor�ow/contrib/pi_examples/label_image

tensor�ow/contrib/pi_examples/camera

Page 19: TensorFlow Dev Summit 2017 요약

Mobile and EmbeddedInception v3 is 93 MB!

Inception v1 quantized is just 7 MB

Exporting Models

Freeze graph

Graph Transform Tool

Quantize weights

Quantize calculations

Memory mapping

Page 20: TensorFlow Dev Summit 2017 요약

Mobile and Embedded12 MB increase, before tuning

2 MB increase, after tuning for Inception v3

Selective registration -DSELECTIVE_REGISTRATION

Page 21: TensorFlow Dev Summit 2017 요약

Distributed TensorFlow

Page 22: TensorFlow Dev Summit 2017 요약

Distributed TensorFlowwith tf.device("/job:ps/task:0/cpu:0"): W = tf.Variable(...) b = tf.Variable(...) with tf.device("/job:worker/task:0/gpu:0"): output = tf.matmul(input, W) + b loss = f(output)

Page 23: TensorFlow Dev Summit 2017 요약

Distributed TensorFlowcluster = tf.train.ClusterSpec({ "worker": ["192.168.0.1:2222", ...], "ps": ["192.168.1.1:2222", ...]}) server = tf.train.Server( cluster, job_name="worker", task_index=0) with tf.Session(server.target) as sess: ...

Page 24: TensorFlow Dev Summit 2017 요약

TensorFlow EcosystemData Processing (TFRecords)

Apache Beam : Native support

Hadoop MR and Spark

Cluster Manager

Kubernetes, Hadoop, MESOS, Slurm

Distributed Storage

Hadoop HDFS, Google Cloud Storage, AWS EFS

Serving

Page 25: TensorFlow Dev Summit 2017 요약

TensorFlow EcosystemPython -> Training

C

C++

Go

Haskell

Rust

Java

Page 26: TensorFlow Dev Summit 2017 요약

Serving Models

Page 27: TensorFlow Dev Summit 2017 요약

Serving Models

Page 28: TensorFlow Dev Summit 2017 요약

Serving ModelsOnline, low latency

Mutiple models in a single process

Mutiple versions of a model loaded over time

Compute cost varies in real-time to meet productdemand

auto-scale with CloudML, Docker & K8s

Aim for the ef�ciency of mini-batching at trainingtime ...

Page 29: TensorFlow Dev Summit 2017 요약

ML Toolkitsmodel = KMeansClustering(num_clusters=1000) model.fit( input_fn=numpy_input_fn(points, num_epochs=None), steps=1000) clusters = model.clusters() assignments = model.predict_cluster_idx( input_fn=numpy_input_fn(test_points))

Page 30: TensorFlow Dev Summit 2017 요약

ML ToolkitsLinearClassi�er

LinearRegressor

LogisticRegressor

KMeansClustering

WALSModel

SVM

TensorForestEstimator

DNN, RNN, LSTM, Wide & Deep, ...

Page 31: TensorFlow Dev Summit 2017 요약

ML Toolkitstensor�ow/contrib/learn/python/learn

classifier = tf.contrib.learn.DNNClassifier( feature_columns=feature_columns, hidden_units=[10, 20, 10], n_classes=3, model_dir="/tmp/iris_model") classifier.fit(x=training_set.data, y=training_set.target, steps=2000) accuracy_score = classifier.evaluate( x=test_set.data, y=test_set.target)["accuracy"]

Page 32: TensorFlow Dev Summit 2017 요약

ETCXLA: TensorFlow, Compiled!

Skin Cancer Image Classi�cation

Sequence Models and the RNN API

TensorFlow in Medicine

Wide & Deep Learning

Magenta: Music and Art Generation

Fast, Flexible, TensorBoard, Community

Page 33: TensorFlow Dev Summit 2017 요약

Science -> Engineering

Page 34: TensorFlow Dev Summit 2017 요약

sess.close()