Machine learning for kids

Machine learning for kid:

 I’d like to introduce machine learning for kids a free tool to help school children learn about artificial
intelligence and machine learning by letting them make things with it. I’m going to share some projects that I’ve seen school classes make with the tool and I’m going to talk a bit about what I saw them learned from doing it now first in case you’ve not seen this before.
This is scratch and it’s used around the world to teach kids about coding and  programming concepts are represented as colorful blocks on a pallet and kids drag them onto a canvas and snap them together to make their programs and it gives them a really friendly and accessible sandbox for them to experiment and learn in now machine learning for kids extends scratch it adds some additional blocks to the scratch palette that lets children use machine learning technologies in their own scratch projects let me show you what I mean now this is a project was made by a class in a local primary school and they were learning about how things like Amazon‘s Alexa work by making a virtual one for themselves.
So in this project they were going to train an AI to recognize the meaning of simple commands now the first lesson they learned was you know what are the main  stages of a machine learning project one collecting training examples two using those examples to train a machine learning model and then three making something with that machine learning model.

Example Machine learning for kids  #1

First collecting examples they created buckets for each of the commands that they wanted to train the computer to be able to recognize so for this project they wanted to create an assistant something that could turn a couple of devices on and off a lamp and a fan so they once they created these buckets they filled them with examples of how they’d give each of those commands so to train the computer to be able to recognize the command fan on they would type in things like please turn the fan on I’d like the fan on it’s too hot in here can we get some air in here and so on and so on now some of the kids who are maybe a bit slow at typing would come up with five or six examples in each of  their training buckets some of the kids are a bit faster at typing would come up with a dozen.

Example Machine learning for kids #2

The set of examples to train their own machine learning model next once they’ve done that it was time to make something with it in scratch now they’re machine learning model automatically gets imported into the scratch toolbox and it’s represented as this new set of blocks so for example they’ve now got a new block that they could give some text to and that block would use the machine learning model that they trained and use that to recognize which of the commands they taught it to recognize that text probably was now these new blocks combined with what the children already learn and already know about scratch means they could make all sorts of new and exciting projects so for example they could make this a virtual assistant that can respond to their commands given in natural language however they’re phrased and by making some like this you know suddenly. Amazon’s Alexa isn’t just this sort of magic black box to them anymore having made a simplified version for themselves it’s something that they’ve started to understand making this helps classes and helps children learn for themselves the steps that we go through to build a smarter system predictor the commands that we think users are going to give collect examples of those commands use those examples to train a machine learning model and then script what you want the system to do in response to what it’s recognized and that is essentially the process that is used in the real world to build this sort of AI project children also learn that this is often a very cyclical process you know they’ll try out their projects they’ll find things that they’re machine earning model doesn’t understand and then they go back and add additional examples to their training examples they train a new machine learning model and they’ll go round that loop repeatedly until they’re happy with how their system performs some of the children also experiment with confidence scores it’s a really interesting thing that you know they learn machine learning systems don’t just give you an answer but they give you a score showing how confident their system is that it’s correctly recognized their input. 
Example Machine learning for kids #3

The scratch program when they’re machine learning model has less than 50% confidence that it’s correctly recognized to command it replies saying I’m sorry I don’t understand so if they ask it something silly like make me a cheese sandwich instead of trying to turn the lamp or the fan on or off it says I don’t understand so–but children learn that there is one confidence for shot that makes sense for all projects or for all machine learning models they play with it they experiment with it and they get to know the impact that this has on a machine learning project what I mean is they’ll try setting the threshold very very high and they’ll see that their system their project will only take in action when it’s very very confident that it’s correctly recognized the command it very rarely takes the wrong action but it says I don’t understand very often even when it’s actually correctly understood what they’d asked it to do and it’s really interesting hearing the children describe the system like this they describe it as being you know too shy or too timid they describe it as not having enough self-confidence sometimes they’ll describe as being like when they don’t put their hand up in class to answer a question even when they actually know the answer because they weren’t confident enough that they had the right answer and it means that the children can talk about when different types of behaviors are appropriate you know for some ailing a patient’s you want to cross a system that doesn’t risk getting things wrong but for other AI applications it’s better that the system always tries when it has an answer it’s okay if it makes mistakes sometimes and by building systems like this for themselves and seeing how they can control the behavior like this it gives them a bit of insight into how these systems are created. A group of people it makes projects much quicker now that sort of planning logistics .

 Example Machine learning for kids  #4

A lot of successful AI projects  like  machine learning for  kid  and they also learned that successful projects will often look for training data that can be created as a byproduct of something people are already doing or that they’d enjoy doing anyway and that’s a really common strategy amongst successful AI projects this was a fun project this class were training a handwriting recognition system they wanted to train a machine learning model to recognize postcodes by drawing examples of postcodes directly onto an onscreen canvas now in the interest of time they didn’t do all the post codes in the country they just picked a few local cities and they didn’t do the whole post code they just picked the first couple of letters so they could do this pretty quick this helped them you know make a project they could do in a single lesson and once they’d use these examples to train a machine learning model they made projects like this a postman pat themed game where they can write a post code directly onto an onscreen envelope and then click on the stamp to send that letter now when they click on the stamp they’re machine learning model is being used to recognize what they’ve written so that their envelope can be sent to the right postman pattern now the great thing about projects like this isn’t just that they’re learning about the tech although that is really cool but by making simplified examples like this they’re learning about how a AI is used in the real world around them because this for example is essentially how large-scale sorting offices have automated the sorting of mail in this project the class for training a machine learning model to recognize sounds as specifically they came up with secure code words for four different places and the class had a lot of fun inventing sort of made-up gibberish noises to represent hotel stadium gas station and and they trained a machine learning model by recording.

Add a Comment

Your email address will not be published. Required fields are marked *