Transcription
Introduction to Artificial IntelligenceDeep Learning - Tensor FlowJanyl JumadinovaDecember 2, 2016Credit: Google Workshop
Neural Networks2/24
Neural Networks3/24
Neural NetworksA fully connected NN layer4/24
Implementation as Matrix Multiplication5/24
Non-Linear Data Distributions6/24
7/24
Deep LearningIEach neuron implements a relatively simple mathematicalfunction.Iy g (w · x b)8/24
Deep LearningIEach neuron implements a relatively simple mathematicalfunction.Iy g (w · x b)IThe composition of 106 109 such functions is powerful.8/24
Deep LearningBook: http://www.deeplearningbook.org/Chapter 5“A core idea in deep learning is that we assume that the data wasgenerated by the composition of factors or features, potentially atmultiple levels in a hierarchy.”9/24
Results get better with:Imore dataIbigger modelsImore computation10/24
Results get better with:Imore dataIbigger modelsImore computationBetter algorithms, new insights and improved methods help, too!10/24
11/24
Adoption of Deep Learning Tools on GitHub12/24
Tensor FlowIOperates over tensors: n-dimensional arrays13/24
Tensor FlowIIOperates over tensors: n-dimensional arraysUsing a flow graph: data flow computation framework13/24
Tensor FlowIIOperates over tensors: n-dimensional arraysUsing a flow graph: data flow computation framework13/24
Tensor FlowI5.7 ScalarINumber, Float, etc.14/24
Tensor Flow15/24
Tensor Flow16/24
Tensor FlowITensors have a Shape that is described with a vector17/24
Tensor FlowITensors have a Shape that is described with a vectorI[1000, 256, 256, 3]I10000 ImagesIEach Image has 256 RowsIEach Row has 256 PixelsIEach Pixel has 3 values (RGB)17/24
Tensor FlowComputation is a dataflow graph18/24
Tensor FlowComputation is a dataflow graph with tensors19/24
Tensor FlowComputation is a dataflow graph with state20/24
Core TensorFlow data structures and conceptsIGraph: A TensorFlow computation, represented as a dataflowgraph:- collection of ops that may be executed together as a group.21/24
Core TensorFlow data structures and conceptsIGraph: A TensorFlow computation, represented as a dataflowgraph:- collection of ops that may be executed together as a group.IOperation: a graph node that performs computation on tensors21/24
Core TensorFlow data structures and conceptsIGraph: A TensorFlow computation, represented as a dataflowgraph:- collection of ops that may be executed together as a group.IOperation: a graph node that performs computation on tensorsITensor: a handle to one of the outputs of an Operation:- provides a means of computing the value in a TensorFlowSession.21/24
Tensor FlowIConstants22/24
Tensor FlowIConstantsIPlaceholders: must be fed with data on execution.22/24
Tensor FlowIConstantsIPlaceholders: must be fed with data on execution.IVariables: a modifiable tensor that lives in TensorFlow’s graphof interacting operations.22/24
Tensor FlowIConstantsIPlaceholders: must be fed with data on execution.IVariables: a modifiable tensor that lives in TensorFlow’s graphof interacting operations.ISession: encapsulates the environment in which Operationobjects are executed, and Tensor objects are evaluated.22/24
Tensor Flow23/24
Tensor Flowplayground.tensorflow.org24/24
Core TensorFlow data structures and concepts I Graph: A TensorFlow computation, represented as a data ow graph: - collection of ops that may be executed together as a group. I Operation: a graph node that performs computation on tensors I Tensor: a handle to one of the outputs of an Operation: - provides a mean