I don’t hide that I love to teach kids. If I wasn’t a biologist or a data scientist, teacher would probably by my 3rd choice. There something fun in seeing the eyes of the kids spark when they get a concept and learn a new idea.
Recently I’ve been challenged by friends on how to explain data science concepts to very young kids. Sure the mathematical and computacional knowledge they have is pretty limited but that doesn’t mean it’s impossible to teach them new concepts. You just have to change the way you teach them.
Challenge accepted! Get your junk drawer of office supplies! It’s time to teach the kids about classification !
So what are classification problems?
Classifying objects, animals, books, food…..is part of the human experiment. Explaining to the kid that the computer first defines the rules with a training set and then when it receives new objects classifies them according to the rules defined previously will become easier when we show them the process in real life.
Time to face the junk drawer or box. Everyone of us has it. A place where office supplies go to wait for their use. Or maybe be forgotten, I don’t know 🙂
A classification algorithm follows rules and those rules may differ from algorithm to algorithm. So our exercise is to show that. Handle some of the objects to the child and give some to you and ask them to separate them however he/she wants. Do the same.
What was the rule for your first sorting ? Maybe, use ?
Was the rule the child used the same as yours? Discuss what rule was used and what extra rules can be used to classify the objects.
Maybe materials are possible right?
Now can we separate further the groups formed with one rule? Sure we can! Post it’s will help you understand the rules used in each iteration.
Let’s start with the use of each object.
Can we extend the classification of the writing objects? How about dividing them by major types such as pencil, marker and pen?
Can we specify the classification of pens even more? Maybe by colour? Let’s try!
It’s time to assess the results of the little experiment with the child. What does it mean for the blue pens? How many attributes did they earn during the experiment ?
Could we try another way to classify the objects? Would another person classify the objects in a different way? Which rules would we find in the process? Let the kids experiment and encourage them to register their finds.
Time to wrap up the experiment
We’ve just done manually what could be called a decision tree in Machine Learning. By questioning our data for one specific attribute (such as colour, material, use, etc) we created a rule to separate the objects and we found out that we can use several questions in a row to further separate the objects, thus creating a tree shaped diagram such as the one below.
We also found that our decisions differ based upon the question asked in each branch of the tree and even who asks the question. Time to explain the kids that computers use the same method to classify data and that the results may differ from algorithm to algorithm.
Wait, have we just found out a way to explain a machine learning algorithm to a kid? Yep =)