Scribd
Upload
443 views20 pages

Slideshare Grokking Deep Learning 170314155452

This document summarizes how neural networks make predictions using simple examples. It explains that neural networks take in input data, multiply it by weights, and output a prediction. A single input network is shown making one prediction based on the number of toes on a baseball team. A multiple input network with three inputs (toes, wins/losses, fans) is shown making one prediction. Finally, a network with one input but multiple weights and predictions (injuries, wins, sadness) is demonstrated. The document concludes by combining these examples into a network with multiple inputs and outputs.

Uploaded by

LADU PRASAD
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
443 views20 pages

Slideshare Grokking Deep Learning 170314155452

This document summarizes how neural networks make predictions using simple examples. It explains that neural networks take in input data, multiply it by weights, and output a prediction. A single input network is shown making one prediction based on the number of toes on a baseball team. A multiple input network with three inputs (toes, wins/losses, fans) is shown making one prediction. Finally, a network with one input but multiple weights and predictions (injuries, wins, sadness) is demonstrated. The document concludes by combining these examples into a network with multiple inputs and outputs.

Uploaded by

LADU PRASAD
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 20

How do neural

networks make
predictions?
Save 42% off Grokking Deep
Learning with code grokdl at
manning.com
Let’s start with the simplest neural network that we
can, to illustrate how neural networks make
predictions.

This network takes in one datapoint at a time


(average number of toes -on the players’ feet- on
the baseball team) and outputs a single prediction
(whether or not it thinks the team will win).
Let’s break down the
process:

Here we have entered our single input datapoint (#


of toes = 8.5).
Now we’ve got to do a bit of
math:

Our input is going to be multiplied by the weight


(0.1).
Out comes the prediction:

We’ve shown how a basic neural network predicts


something - 8.5 toes x 0.1 = 0.85
You might be asking: what exactly is weight
in our neural network?
Our neural network accepts an input variable as
information, and a weight variable as knowledge and
outputs a prediction. Every neural network you will ever
see works this way. It uses the knowledge in the
weights to interpret the information in the input data.
Basically, a neural network's weight is a measure of
sensitivity between the input of the network and its
prediction – like a volume knob.
Now let’s see what happens when we have
multiple inputs:

This example is like our first but it has multiple


inputs, rather than just one (# of toes, win/loss, # of
fans).
We’re going to set our three inputs and
corresponding weights:

As you can see, the three datapoints have been


entered (# of toes, win/loss, # of fans) and the
weights set.
It’s time for
math! We’re
going to
multiply each of
our inputs by
our weights
(notice that the
weights are
different) and
add them
together.
Voila! We have our prediction:

After filtering our inputs (# of toes, win/loss, # of


fans) through our weights (knowledge) we now have a
prediction (0.98).
That’s pretty cool, right? But, what else can
we do with our neural network?
Neural Networks can also make multiple predictions
using only a single input. This is perhaps a simpler
augmentation than multiple inputs. Instead of using
more inputs for a single prediction, we are going to
make multiple predictions about a single input.

The prediction occurs in the same way as if there were


3 disconnected single-weight neural networks – like in
the first example.
It’s like our first example x 3:

This time, we’re going to apply multiple different


weights to our input (win/loss, happy/sad, and
injuries sustained in %).
We insert our datapoints and multiply them
with their weights, one at a time:

It is important to note that these are, in fact, three


separate predictions – unlike our second example,
where the multiple inputs were linked to one another.
It’s time for
math again!
Here we are
going to
perform an
elementwise
multiplication –
like in our first
example, but for
each weight
(0.3, 0.2, and
0.9).
We’ve got our 3 predictions!

By filtering our input through our various weights


(volume knob) we have predicted player injuries, win
%, and probability of sadness!
What would happen if we combined the
methods used in the second and third
examples?
Yes, neural networks can predict multiple outputs given
multiple inputs!
The way in which we built a network with multiple
inputs or outputs (examples 2 and 3) can be combined
together to build a network that has both multiple
inputs AND multiple outputs.
Just like before, we simply have a weight connecting
each input node to each output node and prediction
occurs in the usual way.
Let’s multiply each of our three inputs by our
three weights:

We’re going to perform three independent, weighted


sums of each input.
You can think of this in one of two ways: either
as 3 weights coming out of each input node; or 3
weights going into each output node.
Here, each
output
node has taken
its own weighted
sum of the input
and made a
prediction.
That’s all! Hopefully you found this
presentation fun and informative!
Save 42% off Grokking Deep
Learning with code grokdl at
manning.com

Check out
some of
our other
deep
learning
titles:

You might also like