Name: TensorFlow Hub。再利用可能な機械学習 (TF Dev Summit '19) (TensorFlow Hub: Reusable Machine Learning (TF Dev Summit '19))
Uploaded: 2021-01-14T10:37:36.000Z
Duration: 7 min 36 s
Description: VoiceTubeの動画で発音を聞きながら英語表現を覚えよう！学べる英語：

ANDRE SUSANO PINTO: Hello, my name is Andre,

I'm the technical lead for the TensorFlow Hub project.

TensorFlow Hub is a library and a repository

We open-sourced it last year, and today I'll

Let's start by talking about when you'd want to use it.

If you have problems collecting enough data

then transfer learning is a technique for you.

You can use Hub to make it easy to reuse parts of models that

Additionally, since it's so easy to reuse computations

and weights, it becomes possible to leverage features

without having to learn how to fit them into neural networks.

and experimentation becomes slower over time.

By defining an artifact that does not depend on code,

Hub allows for more maintainable systems,

similar to how libraries have helped software engineer.

So the main concept is these pre-trained building blocks

Typically, we start by training a large model

interested then into a part of it, which is reusable.

This is typically a bottleneck layer or some other distributed

Then we can package them into a SavedModel that

defines this computation and the weights that were trained,

and no longer depends on the original code.

Then you can bring this module back as a piece of a new model

Since the model is defined by TensorFlow primitives,

you can fine tune its weights and adjust them

This concept of saving a part of a model and then loading it

back is getting integrated right in the core of TensorFlow.

that make it possible to share more than just signatures.

And with eager execution, it becomes even easier

to select the part of a network that gets exported.

In TensorFlow 2.0, we can load a module with hub.load.

if the model is already on our file system.

Additionally, due to the new capabilities,

you can now share any object which is composed

So in this case, text_features has two members--

__call__, which is a TF function, and embeddings,

We're also really excited that we added support

for polymorphic functions when we serialize TF functions

This will provide a more natural interface

For example, here we see an image representation module

being loaded and then being used in inference mode,

or being used during training mode where batch norm is on.

And all of this is just baked by TF graphs,

so we no longer need to be selecting things

We just get all the API that looks very--

Additionally, we have added a new symbol--

This makes integrating hub modules into a Keras model

In this example, we see how it is to build a text sentence--

The top layer-- this Keras layer, which is the NNLM--

is a layer that receives sentences as inputs

Then we have a dense layer and a classification layer.

this NNLM layer-- includes text preprocessing on it,

we can just feed sentences straight into our model.

Additionally, if we wanted to try other text models,

They'll be as easy to try like the latest research.

字幕リスト動画再生

TensorFlow Hub。再利用可能な機械学習 (TF Dev Summit '19) (TensorFlow Hub: Reusable Machine Learning (TF Dev Summit '19))

grab

pattern

concept

typically