Hi,
I am a teacher in a secondary school and I would provoke an interest in AI in my students. Are there any projects that are accessible and interesting that I could use? It doesn’t have to be complicated, but if it had a competitive element that would help retain interest.
Thanks in advance.
Hi @JWalker
That’s a great idea! I suppose you’re thinking of secondary school “kaggle”! 
So… Welcome - and thanks for starting a great conversation
Just off the top of my head I can think of a number of things that would be quite straightforward - and have a competitive element through e.g. some metric such as Accuracy/Loss, or hitting a certain target Accuracy/Loss with as few neurons as possible…
For example, recognising MNIST digits is classic and straightforward, and would provide a good pedagogical introduction to certain concepts such as the forward pass, back-propagation, convolution - without needing to get into heavy/math detail (unless to stretch more interested students).
I think the way that PL shows what is going on in each component and during training would be great for students - pure code is not as engaging.
Let me think about some specific examples/datasets and I’ll reply again later, but in the meantime I’m sure one of the PerceptiLabs guys will have some great input for you.
You could also consider some of the many Live Coding examples that @robertl has shown on the PerceptiLabs YouTube channel… or have you already considered those and are looking for something else?
PS I don’t know anything about teaching, but I have heard of something called a “lesson plan” - if you already have some ideas about the points you want to discuss/illustrate we could help work out how best to use PerceptiLabs for that.
And… what IT would the students be using? Obviously they need to be able to make reasonable progress on something in the time & with the computational resources available.
Hi @JWalker,
That’s an awesome idea!
I think @JulianSMoore already covered most follow-up questions and good suggestions, I just wanted to chip in and say that there are a few different datasets which gets downloaded with PerceptiLabs that are intuitive and some providing more challenges than others (Mnist is included in those).
Also let us know if we can assist with anything specific
Hi Julian and Robert,
Thanks for your answers.
They would very likely be using laptops because we ask them to bring those to school. I can tell them that they code will optimise faster on a PC, but they are of a generation where everything is mobile so they are likely to have a laptop, phone and playstation/xbox.
I haven’t played very much with PerceptiLabs just yet because I wasn’t sure if it was suitable for me. You have mentioned Mnist and Accuracy/Loss so the first port of call is for me to go and learn those myself before thinking of a way to present them to students. They are bright kids and they will pick it up easily enough.
I would be running this as a club and not as a lesson per se, hence all that I need is for them to have their own copies of the software and then for me to demonstrate the concept and to set them the task. The difficulty will arise when I will have to be IT support or a code debugger when it inevitably “doesn’t work”.
Thank you for the responses. It sounds like it is a great way to introduce the kids to AI. We are a school who is trying trying to teach the students to work with AI because in the future that is likely to be their reality.
Hi again @JWalker
I admire your - and the school’s - forward thinking on making sure the students are AI literate; I certainly benefitted from having access to computers a long time before they were mainstream. I think I learned more about computing at school than at uni (because I’m an old school Physics BSc)
Just so you know, I’m just another enthusiastic user, but I have been engaged with PerceptiLabs for quite some time now and I’m keen to get the word out on AI and tools like this because, like you, I see it as the future… So, what next?
In case you didn’t notice it at the top of the page this link gives you access to the PerceptiLabs Community on Slack - I find it slightly friendlier for interactive chatting as opposed to sending/receiving posts/replies. If you want to pop over there we could do the following via Direct Messages
And I’m sure the rest of the PL crew will back us up as needed!
Have a look around at the datasets that are included with PL - or available elsewhere - and let us know where you’d like to start.
Over to you!
Hi Julian,
Thank you, that is great. Give me a couple of days and I will get myself on slack so that we can chat more there.
Hi @JWalker, it is a great idea, as an engineering student, I can tell you that I would have liked to be taught this kind of subjets at school.
An interesting project that could work for you is a fruits classifier (bananas vs apples for example), it is easy to implement, and it can make it easier for your students to interpret the metrics, as we say in spanish “first you learn how to walk, then you can run”.
Thank you Camilo.
[20 chars]
Hi @JWalker - I guess you have had other things on your plate, but don’t hesitate to ask for whatever you need whenever you want to throw out a question
Sorry Julian,
I just moved country, so time has been a little challenging and your email came through to my spam so I didn’t know that you had replied. This is on my todo list, but life is a bit busy right now. Thank you for checking in.
Hi @JWalker
I spent nearly twenty years living in and moving around Europe (Budapest, Frankfurt, Alicante, Barcelona, Brussels) and moving never gets easier… in fact I moved back to UK from Belgium Dec 20th, 2020 at one of the lets-close-the-borders-and-cancel-trains times of Covid-19…
I feel you - hope it’s going smoothly and we’ll see you when you’re feeling settled.
Thanks.
I just moved to Malaysia and working whilst in quarantine. The good thing is that quarantine lets you deal with jetlag very easily
Ok, I built and optimised my own model using the perceptilabs data.
I was chatting with Robert and I need to use Gradio to run my model.
It was fun.
I have been playing with this and it is becoming quite frustrating. Do I need to learn a lot of ML so that I can implement the models? I have the model, but I cannot encourage it to accept an image and give me a prediction.
Hi @JWalker
Great to hear you’ve been able to build and train a model! I haven’t yet tried gradio myself so I can’t say much except, as far as I know, no, you don’t need to learn a lot of ML!
PL has two key strengths: it is a visual editor of the computation graph that has always been the basis of TensorFlow, so it’s easier to conceptualise your model as it evolves in the first place, and secondly, the “data wizard” knows about common model types and should construct a viable basic skeleton for you.
From there you can play with parameters
And then structure (more layers, more “features” a.k.a. “kernels” for convolutions etc.
For the final part of what you are doing - if you can provide screenshots or other output to clarify the situation I’m sure @robertl will be able to get you over the last hurdle.
Thank you, I still have a bit to learn with regards to this.
I have to be mindful that the students would be taking this as an optional co-curricular activity, hence I need to make it fun with a tangible end result. This is why I like the sign language idea because they can train it themselves and then spell their own names and it see if it picks it up. Afterwards, we could move onto a more difficult task.
I am still working through tutorials etc, it is interesting for me at least.
Sounds like you are making steady progress. I would be really interested to hear what you (and your students) think are the easiest & hardest parts of ML etc.
I personally find the terminology & jargon a bit of an obstacle - especially the names given to various metrics which I think are too ambiguous/confusing! Every discipline has its history and quirks though and unfortunately there’s no shortcut…
Which makes me wonder whether there are any interesting ML/AI mnemonics or other aides memoires.
I think we could all - I certainly would, I think - benefit from knowing how you approach teaching this.
Hi Julian,
The memory aids would be something that I would ask them to create. You would find that the students are great absorbing new information, you only need to tell them once; however, they are just as great at forgetting it just as quickly. Hence, we need to encourage them to use that information shortly after they have assimilated it.
Teachers usually follow a look - see - do model. i.e learn the theory, see an example using that theory and then practice using the theory. Personally, I think that there is another step; people need to consolidate the information by either creating a mental model (this is better for qualitative information) or justifying the logical steps in a sequence. Essentially, they need to explain the process to themselves.
Consolidation - good point: I think I’ve seen in self-study recommendations next day, next week & maybe a 3rd review for consolidation.
I knew that kids were great at 1-shot learning (and maybe we are too - still - just less aware of it?) but had never really thought about the fragility of memory…
I’ve created a few mnemonics etc. over the years… but with a whole group at it there should be some memorable stuff! Looking forward to hearing all about it eventually