How

Neural Networks
Work
 Learning Goals
Essential Question:
● What are neural networks and how do they work?

You will be able to:

● Describe the parts of a neural network.
● Explain what a neural network does.
 ACTIVITY SLIDE

● Choose any of the following activities and play with them for the next 5
minutes.
● As you play, consider:
○ What is the purpose or objective of the activity?
○ What do these activities have in common?
 ACTIVITY SLIDE
Scroobly
● Visit https://www.scroobly.com/.
● On this site you can choose a silly scribbled character to follow your
movements.
● Click “Start”.
● Go through the instructions, clicking “Next”.
● Choose a scribble or draw your own. Watch it follow you!
 ACTIVITY SLIDE
Lipsync
● Visit https://experiments.withgoogle.com/lipsync.
● This site uses a neural network to see how close your lips are to saying the
correct words. The information learned from it will help computers
understand people who have a hard time speaking.
● Click “Let’s do it”.
● Let the page load. Allow the computer to access your camera.
● Click “I’m ready”.
● When you can see your face, Lipsync!
 ACTIVITY SLIDE
Imaginary Soundscape
● Visit http://www.imaginarysoundscape.net/#/street.
● This site has a computer program imagines what different places would
sound like based on images and plays those sounds.
● Click “Street View” to use Google Street View images. If you prefer, you can
upload your own image.
● Click “Random” to go somewhere random.
● Put in a specific place you remember.
● Does the soundscape match what is in your head?
 DISCUSSION

What did the activity you played with do?

● How was it similar to the other activities?
Definitions
 Artificial Intelligence
Computer Science
● The activities you played with
are all examples of AI.
Artificial Intelligence
● Artificial Intelligence (AI) is a
branch of Computer Science Machine Learning
concerned with the designing of
computers that make
predictions and decisions. Neural
Networks
 DISCUSSION

● What did the activities you played with make

predictions about?
● What decisions were they making?
 Machine Learning
● They were also examples of Computer Science
Machine Learning.
● Machine Learning is a branch of Artificial Intelligence
Artificial Intelligence.
Machine Learning
● Using traditional methods, a
programmer would have to attempt
to describe how every possible
image might look in code. Neural
Networks
● With machine learning, the
programmer feeds the program a lot
of examples and the program learns
for itself what sorts of image data
means what.
 Neural Networks
● A lot of machine learning algorithms
Computer Science
use a technology called neural
networks.
● Neural networks are a collection of Artificial Intelligence
functions trained to solve a problem.
Machine Learning
● Neural networks find patterns in
data, save calculations that find
those patterns, and then use that
information to do the task they were Neural
built for. Networks
● We will be exploring how a trained
neural network works for the rest of
this lesson.
Neural Networks
 Definitions
Parts of a Neural Network
● Neuron: Essentially, a calculation. It is given input, does calculations on
that input, and outputs a number. Each layer of neurons uses each of the
numbers produced by the previous layer as its input.
● Input layer: All the starting information that can be used as input. In
pictures this is often pixel values; in NLP this is often lots of bodies of text.
● Output layer: Contains whatever the neural net was looking to produce -
what the image is, a new image, the most likely next word in a sentence,
etc.
● Hidden layers: All the layers between the input and the output layers.
● Activation: When a neuron finds what it is looking for, i.e. the number
output by a neuron is high enough to indicate that the input is what the
neuron was looking for.
● Neural Network: A collection of neurons trained to solve a problem.
But what is a neural network?
(Pt 1)
● Why is it hard to
program a
computer to
recognize a
handwritten digit?
But what is a neural network?
(Pt 2)

● What is a neuron?
But what is a neural network?
(Pt 3)
● In this example, what
is the input layer?
● In this example, what
is the output layer?
But what is a neural network?
(Pt 4)

● What do the
hidden layers do?
But what is a neural network?
(Pt 5)
● How do the layers of a
neural network interact
with each other?
But what is a neural network?
(Pt 6)

● In this example, what

is each layer doing?
● Why is this helpful?
 Video Overview
● One person’s handwritten ‘3’ often doesn’t look anything like
another person’s handwritten ‘3’, yet our brains can tell that
they’re all the same number. This is pretty awesome, actually.
● Neural networks are the main way we can get computers to do
similar tasks. Neural networks are made of layers of neurons.
● Neurons are calculations that produce numbers about the data.
● A neural network trained to find handwritten digits will have an
input layer of all the pixels of the image.
● The output layer will give the number that was most likely drawn.
● Hidden layers in between look for edges that put together make a
number.
Activity
● Let’s act like a trained neural network!
 Output
1a Hidden 1i 0
Layer
Layer
1 1
1j Hidden
1b
When you 2a
Layer 2g
are 2 2
activated, 1k
2b
place your 1c 2h
3
post-it note 1l
next to your
2c 2i 4
neuron 1d
name 1m

2d 5
2j
1e 1n
6
2e 2k
1o
1f 7

2f 2l
1p 8
1g

1q 9
1h
 DISCUSSION

Analyze the effectiveness of the neural network.

● What did the neural net do well?
● What didn’t it do well?
● How could we improve if we ran this again?
Neural Networks in
the Real World
AI & Medicine
Daphne Koller
Founder and CEO, Insitro
● Insitro uses machine learning to test molecules.
○ Find patterns in lots of data using Neural Networks
● Worked for Calico - Google’s extension-of-life branch
● Co-founder of Coursera and Engageli, both platforms to share college
content online
● Education
○ PhD at Stanford
○ BSc & MSc at Hebrew University in Jerusalem
● LinkedIn
AI & Astronomy
Brian Nord
Founder and Director of Deep Skies Lab
Scientist at FermiLab
● Deep Skies Lab uses AI to find information about the structure of
the universe, classifying galaxies, and other astronomy problems.
● At Fermilab he works on the origin of the universe.
● Also often discusses diversity and inclusion in STEM
○ Co-author of https://thisisblacklight.com/ , a curriculum for
learning about Black experiences
● Co-created an academic Strike for Black Lives
● Education
○ PhD in Physics at University of Michigan
● LinkedIn, His site
Closing
 Check for Understanding

How does a neural network work?

● What do the different parts of a neural network do?
○ Neuron
○ Input layer
○ Output layer
○ Hidden layers
 Learning Goals
Essential Question:
● What are neural networks and how do they work?

You will be able to:

● Describe the parts of a neural network.
● Explain what a neural network does.
Project
 Vocabulary

● Audience: the people giving or likely to give attention to

something. In this case, our audience is the community
we've been interviewing.
● Audience-centered message: a message that is tailored
to a specific audience
 Teaching Project - Timeline
● Brainstorm the audience you’d like to create your project for
● Collect information about the audience through interviews
● Plan your lesson
● Run the plan by experts on your audience
● Create lesson
● Get feedback from classmates
● Edit lesson
● Run lesson
● Iterate as desired
● Rubric
 Teaching Project
● Now that you know how neural networks work and what they are good
for, you can explain to someone else.
● You will be choosing your audience. Some possible audiences:
○ Adults in your family
○ Other students
○ Friends
○ Policy-makers
● Choose one application your audience uses that uses neural networks.
Explain to your audience how that application works.
● You will have time to develop, collect feedback, and edit it.
 BRAINSTORM
Spend a few minutes thinking about a community you are part of, of
connected to. This can include people who make policy for your
community like city councilors.

How would this community benefit if you taught them more about AI?

After a few minutes, pick one that you are most interested in teaching.
 Teaching Project - Content
● Choose a technology your audience likely uses that probably uses
neural networks.
● Create goals for your audience.
○ What do you want them to know after your lesson?
● Create an assessment that ensures the audience learned what you
wanted them to learn.
● Explain to the audience how that technology works, keeping the
learning goals in mind.
 Teaching Project - Materials
● You can make your project anything, such as:
○ A video explanation
○ Slides / Prezi
○ A model
○ A play where you act out the different parts of the network.
● Make sure the presentation medium you choose is appropriate for your
audience.
 Teaching Project - Assessment
● There are lots of ways to assess what somebody knows:
○ Ask 3-5 multiple choice questions.
○ Ask an open ended question.
■ Make sure you know what qualities an answer should have in
order to show they know the content.
 Question Brainstorm
Work with your peers to brainstorm a variety of questions that you can ask people from
your chosen audience. (You may also want to ask experts within your audience. For
example, if you want to teach first graders you can also talk to first grade teachers.)
● What personal information (if any) do we want from our interviewees?
○ Possible examples: first name, age/age group, occupation
● In small groups, create a list of three questions to ask community members that
will help you gauge their understanding of neural networks, and how they like to
learn. For example:
○ Describe a lesson that really worked for you. Why did it work?
○ What sorts of AI do you use? What do you know about how they work?
 DISCUSSION
Share with the class:
● What questions are you going to ask and why?
Class feedback:
● Can they ask fewer questions? Does this cover what
they need to know?
 INTERVIEW PREP
Now that we've prepped for our interviews we can start collecting
responses from the community.
You will be working with a partner to conduct your interviews. Team up
with one of your classmates and plan your approach.
● Make sure you have your list of questions ready.
● Decide how you will divide up the interview. For example, one team
member can ask the questions while the other records the answers.
● Each team member should have a chance to take on a speaking role.
 THE INTERVIEW
Try to get at least 3 people.
Use this template for your interviews:
1. Introduce yourself.
2. Give a brief explanation of what you are doing.
3. Ask the interviewee for any necessary personal information (first name, age,
occupation).
4. Ask each of the questions on your list.
a. Give the interviewee enough time to think about and answer the question.
b. Try to avoid giving your own answers or opinions.
c. At least one team member should be recording answers.
d. It's okay to accept "I don't know" for an answer. That will be useful to us later!
5. Thank the interviewee for their time.
 Tips for understanding an
audience
● Identify the listeners. Who are they? Where are they from?
● Find common ground. What do you have in common with
your audience?
● Analyze the audience's established attitudes and beliefs
about your topic.
As we think about how to teach our community more about
neural networks, we can start by using the information we
gathered in our interviews to build a profile of our audience.
 UNDERSTANDING
YOUR AUDIENCE
What have you learned about the people you will teach?
● What are the primary characteristics of our audience?
● What interests them?
● What kinds of AI do they use?
● What does our audience know already?
 PRESENTATIONS
Some questions to ask yourself while you're preparing your presentation:
● What does the community already know about AI and in particular Neural
Networks? Were there any common misconceptions?
● How can you communicate your lesson effectively?
PRESENTATION FEEDBACK
For each of the presentations:
● Name at least one thing they did well.
● Name at least one thing they could do better.
● Consider:
○ What is the audience for this presentation?
○ Was there anything that was particularly clear or unclear?
○ What about the presentation was engaging?
 EDITS
Look at the feedback you received.
● What would you like to change about your lesson based on the feedback?
● What are you not going to change? Why?
 PRESENTATIONS
● Share your projects with your audience.
 REFLECTION
What was this experience like?
● What did you learn about audience-centered communication?
● What was the process of editing like?
● Did the assessment show that your audience understood your project?