How do we accelerate teaching young girls Machine Learning
Let’s face it! We need more women everywhere! One Kamala won’t cut it. We need more women like Kamala Harris, Jennifer Doudna, Mae C. Jemison, Tiera Guinn, and Fei Fei-le. I shouldn’t be able to count them on my fingers for one.
Data Science is what I know and what I have seen in my workplace and I see a big gap in how many women are represented! It’s an abysmal 24%. So, I hope you can use this playbook to help our younger generation get interested in this favorite subject of mine and change the course of our future.
Here is what I will be talking about
- Why is Data Science & Machine Learning Important?
- What to expect while teaching young teens? I have never dealt with teaching kids before so I have good learning lessons here.
- The content of the 3-hour workshop
Why is it Important?
I didn’t realize how hard it would be to navigate corporate America as a woman, much less as an immigrant woman in the tech industry. Women in the field of Machine Learning, Data Science, Artificial Intelligence are rare and I often find myself debating about tangible solutions with my male counterparts who manage everything from technical debt to strategic business decisions.
We need more women thinking through the solutions we use every day. We need more horsepower and movement into getting younger girls to become future leaders in Artificial Intelligence. Those who can question the moral ethics of what should be a data science solution and what is not a great idea to invest in. The most ‘Oh my god’ moment was this solution created by Target to predict when women could get pregnant(the hottest marketing target! Supposedly they are extremely profitable). Guess who developed that solution? A man. I wish there was somebody to question it.
That’s why I was grateful to my company, NCR when they offered to support my idea of conducting a workshop to inspire young underprivileged women from inner-city Atlanta schools to explore data science. I also wanted to get advice from strong women before me on how to lead a data science workshop for young women. I’m so lucky to have had 100 girls of code and Becky Tucker provide their support and feedback. Thank you.
I have never dealt with teaching kids before!
To preface, I was teaching Data Science to kids from 9–16. That is a huge age gap, and the learning curve of the group as a whole is highly varying. So, keeping the teaching sessions super interesting, engaging, and more hands-on was the key. More on that later.
It was hard to get most of these 75 kids to smile.
I’m one of those very tiny (5'0"), super happy, extremely positive people who’s always high energy and finding things to smile about. I thought that if I was enthusiastic and joked with the students, the girls would be smiling and laughing, too.
That didn’t happen easily, It took about an hour for most students to begin opening up and start smiling. I felt that this was a cautious group of students who didn’t open up or trust easily
You could lose them if you spoke to long with no hands on activity:
This was a learning lesson, I had 2 students napping and dozing off. I was very glad that I had the teachers reminding us to keep it short and precise and to make it a conversation rather than a lesson
These kids are super sharp and are like sponges:
Don’t keep it too light! They can handle complex information trust me! If you intrigue them enough you will get tons of interesting questions. I had kids raising their hands to either answer or ask questions and the enthusiasm was awesome.
The Content of the 3 Hour Workshop
I had a few conversations with industry leaders and experts to understand how to explain core concepts to kids easily. Becky Tucker was very gracious to offer her time and advice on how to structure the lessons. I have even picked a few of her examples!
- Where data science is used?
- What is Data Science vs Artificial Intelligence vs Machine Learning?
- Basic foundational machine learning concepts like Supervised and Unsupervised learning
- Hands-on example of the core concepts
- Conclusion
Explaining Data Science as Theory first
Our understanding of data science is really this:
But, we are not too far away from that idea: let’s take a look at the SPOT:
This ‘dog-like’ Boston Dynamics robot “SPOT” is being used in Singapore to encourage social distancing. Robots like these or self-driving cars are pushing the boundaries of AI & Data Science forward.
That’s not the only place where Data Science is used heavily. Think about the time when you last typed in google search, it is almost like google is reading your mind. It isn’t though, it just knows what most people search for and gives you suggestions on what it’s learned.
Google home, Netflix all of these folks are Data Science pioneers and they make that technology available to us at our fingertips!
Characters introducing core concepts! Adds fun to the classroom
What’s even cooler is that we are Data Science machines ourselves! Let us look at an example and Agnes can explain it (I am sure we all love Agnes from Despicable me!)
Agnes loves Art, Math, and Dogs. She is not very fond of History though. The question: Is Agnes going to be happy or sad doing the following tasks?
Some answers are easy for us to decide because our brain is telling us: “most probably Agnes will have more fun playing with the dog, and coloring her book than watching a history movie”.And our brain knows that because it is performing a Machine Learning task that we call “Classification”.
Pay attention to the question being asked here. “is Agnes going to be happy or sad?”
In Data Science, it is important to ask the right question to solve the problem intelligently.
3 words are used all the time! What do they really mean?
Artificial Intelligence: This is the superset of all the other sets. To build something like Eve (from WALL-E) we need all the tools that will create a complex program. For example: if you need Eve to identify a glass full of water and pick it up and give it to Wall-E(Wall-E can then water the plant he so lovingly holds with it :D), we are saying Eve is artificially intelligent.
Machine Learning: For Eve to identify a glass is a Machine Learning task. she first needs to know what ‘glass’ is and then she needs to understand what ‘full’ means. This whole process can be done by showing Eve lots and lots of full and non-full glasses.
Data Science: Data Science is part of Machine Learning(identifying a full glass) and part analytics which becomes part of decision making. Example: do we want Eve to learn how to pick a glass full of water or do we want her to understand how to pick up a plant? This decision needs to be made before building a machine learning model and most of the time we make this decision based on our experience with the world.
The next part is understanding the simple foundational concepts of Machine Learning.
Hands on examples are key to explain the concepts in easy way.
Everything about Machine Learning is learning, the way we humans learn from things around us. Sometimes, we already know what we are looking at and have all the answers. Other times we are unsure where we can place that item. This is the subtle difference between Supervised
Supervised: (I printed our many such examples and asked the class to guess the correct answer)
We already know the answer here and when the machine answers a question we ask, we can tell if it guessed it right or wrong.
Dog & Cookies
We all know that the picture on the left is a dog and the picture on the right is a cookie. Now, we have to tell a machine to recognize when it sees a dog to correctly point out this difference. One method is to differentiate by saying, dogs have 3 dots(2 eyes and 1 nose) and cookies have more.
Easy! right? WRONG
The machine is going to be like we have 20+ dots here, so it ought to be cookies! However, we know this is not cookies but a bunch of dogs together.
Building a ‘Dog or Cookie’ program is called a Supervised problem because we know how to check whether we got it right. In other words, we can supervise the results. With supervised learning, we have a set of metrics that we can measure to tell us whether our Machine Learning model is correct or not. The most common way is the Confusion Matrix.
Try to not skip the math part. I used confusion matrix as a easy way to introduce few concepts, but there are many more things to cover here.
Confusion Matrix:
Let’s say we decided on using the same feature (# of dots) to make our decision. We give the machine 100 pictures to figure out which of them are dogs and which of them are cookies.
We can start building a matrix like the one shown below and ask these questions:
- How many of the dogs did the Machine guess right: 40
- How many of the cookies did the Machine guess right: 50
- How many times did the machine think the dog was actually a cookie: 4
- How many times did the machine think the cookie was a dog: 6 (we could have many cookies with just 3 dots.. Missed the chocolate chip on them.)
This matrix is used very often to decide how much better we are doing at teaching the machine… Ideally, we want the machine to guess all of the answers right.
Un-Supervised Learning: (I asked the class to give me differences between dogs and cats- both 4 legged animals with whiskers)
Here we are a little unsure and want help from the machine itself to help us sort and categorize things in buckets so that we can come back and make a better assessment. Let’s walk through another example.
We have 100s of puppies handed to us and the person says “I don’t know what breed they are”. You say wait a minute let me first run an Unsupervised ML model. You start noting their color, height, weight, ear length, paw shape, bark sound, etc., and tell the algorithm- can you put all the puppies that are similar together?
That technique is called clustering and it is Unsupervised because we do not know the right answer.Because we do not know the right answer we do not have metrics to measure if we are right or wrong. We only know some puppies belong together because they have similar features.People have come up with mathematical ways of figuring out if the answer is right but a lot of times in practice it is done manually.
CONCLUSION
We have so far learned the difference between AI, ML, and Data Science.
We have learned about some core fundamentals of Machine Learning starting with the Supervised and Unsupervised learning distinction.
Within Supervised we learned about Classification and Confusion Matrix. We also learned about an unsupervised machine learning example
This is foundational Machine Learning. We are just looking at the tip of the iceberg and there is so much more to explore in this space. I hope more and more young kids join in this journey!
Credits:
Becky Tucker- Thank you for generating the curiosity of exploring ideas right down to their core foundational concepts. The example of cookies and dogs is borrowed from a presentation Becky was giving at Netflix Data Science for a women's meetup.
100 Girls of Code- My fellow data science explorers please reach out to these folks! They are doing an awesome job of teaching code, technology, and advanced concepts to underprivileged kids by exposing them to real-world companies.
(P.S: Feel free to comment below and let me know if you disagree with some explanations! I hope we all can contribute and make this learning experience far better for future generations.)