0:00 tensorflow an open source machine
0:02 learning framework famous for powering
0:04 deep neural networks with high-level
0:06 code it was developed by the google
0:08 brain team and first released in 2015.
0:10 it's most commonly used with python but
0:12 can run in other languages like
0:14 javascript c plus plus and java at its
0:16 core it's just a library for programming
0:18 with linear algebra and statistics as
0:20 you know the word tensor describes a
0:22 multilinear relationship between sets of
0:24 algebraic objects within a vector space
0:26 aka a multi-dimensional array what makes
0:28 it special is its collection of apis for
0:31 data processing visualization model
0:33 evaluation and deployment that make deep
0:35 learning accessible to the average
0:37 developer it's extremely portable and is
0:39 able to run on tiny mobile cpus or
0:42 microcontrollers with tensorflow lite
0:44 can run in the browser with
0:45 tensorflow.js while the core library can
0:47 scale up to multiple gpus or run on
0:49 tensor processing units ships engineered
0:52 specifically to run tensorflow at a
0:54 massive scale it's used in medicine for
0:56 object detection and mri images by
0:58 twitter to sort your timeline by tweet
1:00 relevance by spotify to recommend music
1:02 by paypal for fraud detection in
1:04 addition to many other applications like
1:06 self-driving cars natural language
1:08 processing and so on to build your own
1:10 neural network right now create a python
1:12 file and install tensorflow next we'll
1:14 need some data like fashion mnist which
1:16 we can automatically import the goal is
1:18 to train a model that can predict the
1:19 clothing type of each image tensorflow
1:21 has a subclassing api for expert users
1:24 but also integrates with the
1:25 beginner-friendly keras library which
1:27 has a sequential api that can easily
1:29 build neural networks layer by layer we
1:31 start with a flattened layer that takes
1:33 the 28 by 28 pixel image as an input and
1:35 converts it into a one-dimensional array
1:37 this input layer is then fed into a
1:39 dense layer with 128 fully connected
1:42 neurons or nodes you can think of each
1:44 node like its own linear regression as
1:46 each data point flows through it it'll
1:47 try to guess the output and gradually
1:49 update a mapping of weights to determine
1:51 the importance of a given variable in
1:53 this case it uses a rectified linear
1:55 activation function that will output the
1:57 input if a certain threshold is met
1:59 otherwise it will just output zero and
2:01 the behavior of this layer can be
2:02 customized by tuning as hyperparameters
2:05 finally we have our output layer which
2:06 is also dense but is limited to 10 nodes
2:09 which corresponds to the total number of
2:11 clothing types in the data set now we
2:12 can compile the model and tell it to
2:14 optimize a certain loss function like
2:16 sparse categorical cross entropy as we
2:18 train the model for multiple epochs its
2:20 accuracy should gradually improve the
2:22 end result is a model that makes a
2:24 prediction with the likelihood that an
2:26 image is a certain type of clothing
2:27 congratulations you just built a neural
2:29 network this has been tensorflow in 100
2:32 seconds hit the like button if you want
2:33 to see more short videos like this
2:35 thanks for watching and i will see you
2:36 in the next one
Chrome Extension