Hi, internet.

This is Andres Rodriguez.

I am a machine and deep learning solutions architect.

Deep learning is a branch of machine learning

that attempts to model various levels of abstractions

and data.

It is popular for vision, speech, text, and control

tasks, among others.

There are various frameworks or tools to train and deploy

deep networks, such as Caffe, TensorFlow, Theano, Torch,

and the list continues to grow every year.

Intel is actively working with the deep learning community

to optimize many of these frameworks,

to maximize CPU utilization, and significantly

improve computational performance on Intel's

architectures.

A series of deep learning videos will

be available on the Intel Developer

Zone, where we will discuss some of the frameworks

and how to set them up to efficiently train and deploy

deep networks on Intel's architectures.

We will also discuss what Intel is

doing to optimize these frameworks,

both for single node and for multinode distributed training.

Some examples of deep learning applications

that can benefit from Intel's commitment

to reducing the time to train are image classification.

In this example, the input is an image,

and the output is the class label of the image.

Object detection.

The input is an image, and the outputs

are the bounding boxes of the objects in the image

and their respective labels.

Image segmentation.

The input is an image, and the output

is a semantically segmented image

where each pixel produces a label

corresponding to some object.

Natural language object retrieval.

The input to the network is a query

in natural language, where the answer is

an object in the image.

The network retrieves the object from the image.

Visual and text question answering.

The input is a query in natural language about the image,

and the output is the answer in natural language.

Visuomotor control.

The input is an image, and the output

is a motor function performed by a robot.

Speech recognition.

The same network configuration is

used to train an English and a Mandarin speech

recognition system.

And natural language understanding.

In this example, the input is a short story followed

by a question about the story.

The output is the answer.

There are many other examples, as deep learning is quickly

expanding to various areas.

At Intel, we're committed to enable industries and research

institutions to effectively and efficiently use

deep learning for their applications.

To learn more about this topic and others

related to machine learning, please follow the links

in the description below.

Thanks for watching.

[MUSIC PLAYING]