AI READY ASEAN HACKATHON - WORKSHOP 2

追加: 2026-05-13

[00:01:30]Ready?

[00:01:46][clears throat] That way on. Okay.

[00:02:36]Okay. Bye.

[00:03:38]Okay.

[00:03:38]>> Okay.

[00:03:39]>> All right. Okay.

[00:03:48]formator AI ready Build your own AI model with PTO block.

[00:04:36]Okay.

[00:04:48]Okay.

[00:05:01]Okay.

[00:05:06]I'm teacher P from college and I'm proud University.

[00:05:10]So today we want to conduct a second workshop on this AI ready assean hackathon. So today our speaker is Mr. Liipan from Chimpaka Maka. Okay, please allow me to share my screen.

[00:05:32]Okay, this AI ready as program is started since 2025. So this is the second year in conducting this AI ready ASEAN program in 11 ASEAN countries.

[00:05:45]Okay. Please um please >> all right let's get this logo moving >> just a little further.

[00:05:55]>> Done.

[00:05:56]>> AI Ready AIN empowering the future with AI. AI Ready AIN is an initiative by the Aian Foundation with support from Google.org or is set to empower 5.5 [music] million Azian citizens to thrive in the AI era. Running from 2024 to 2026, the program will train 2,000 master trainers, engage 800,000 individuals in AI learning through workshops, training sessions, and collaborative activities.

[00:06:22]AI Ready AIN has engaged participants across learning environments, fostering innovation and digital readiness.

[00:06:29]Educators were empowered with AI knowledge and tools enabling them to transform classrooms and inspire the next generation.

[00:06:36]AI ready aian program has been implemented proactively in Malaysia.

[00:06:41]Malaka Joango.

[00:07:20]Waha!

[00:07:30]Koala Lumpur.

[00:07:41]Pirul pina.

[00:08:15]Sarawa Saba.

[00:08:51]love one.

[00:08:57]A R I A R I A Ration AI Ready AIN empowering today's learners for tomorrow's digital future.

[00:09:28]Okay. So, AI ready ASEAN is an initiative by the ASEAN Foundation with support from Google.org. Okay. We aims to impact 5.5 million citizens. Okay. To empower them to thrive in the AI era.

[00:09:42]And in this initiative, we plan to train 2,000 master trainers, okay, to train more students. And there are 800,000 individuals engaged under this AI ready ASEAN program from 2024 to 2026.

[00:09:59]College Tingata and Namra University Tun Fatima Mahalaka was selected by ASEAN Foundation as one of the local implementing partner to implement AI ready ASEAN. So since last year we have conducted a lot of workshop on arrow of code on the AI awareness raising campaign. So this year we will proceed this program using the AI comic and also three sessions of the workshop to raise the AI awareness among the students uh teachers and parents. We also implement this AI literacy module in the AI class as young website. Okay. to engage the AI learning. Okay. Host regional and national policy convening, release a comprehensive regional AI research report and finally launch an e-learning platform serving as a central AI resource hub.

[00:10:51]The enc the objective of this AI ready ASEAN is to encourage teacher to participate in the AI training to actively enhance their teaching capabilities to integrate AI concept to equip students with indemand skills for carriers in STEM fields to promote the modernization of education in the ASEAN region by cultivating more AI skill teachers and students and engage in AI and digital literacy to enhance students competitiveness in higher education and in the job market.

[00:11:26]So the targeted individuals in this program includes youth okay from 15 to 35 years old. But we are welcome the primary school students to join us also educators, parents and underserved communities from 15 to 64 years old.

[00:11:44]The AI awareness raising campaign in 2026 start in May start in March until August 2026. So in March we have conducted the AI commit craft challenge and prompting AI workshop. Now this is the second workshop. Okay. We will learn about AI model and we will promote this AI data lab challenge. In July we will we will conduct the third workshop is about data science for innovation and the AI data lab challenge will close in August.

[00:12:19]Okay. The AI data lab challenge will use the platform like picto blocks Google teachable machine or any suitable platform that were introduced by Mr. D later. The categories is divided into three categories. primary school, secondary school and also preun university. Each each group must contain two students. For the primary ground uh preliminary round, we will select the champion and they will advance to the final round in September.

[00:12:55]This is the competition workflow. First the student need to find the insight and identify the problem and they start the data collection and sourcing. Next is data analyzing and dashboard correation.

[00:13:10]Okay. Then the student will learn the machine learning modeling with picto blocks today. And finally the students need to submit the video presentation and the solution pitch. So the submission requirements are the onepage storyboard in PDF which contain the problem and the AI solution and to show the data by using charts graph and the data source link. While for the video presentation, okay, is to explain how your project works and who it helps to prove the a machine learning training data and coding and to demo a real time test showing your AI working.

[00:13:56]Each participant group is allowed to submit only one entry and two participants must be in the same category. Combination of two different categories are strictly not allowed.

[00:14:08]Late submission after competition time will not be accepted. Any inappropriate violence, sensitive content and plagurism are strictly not allowed and the result will be announced on the AI ready ASEAN platform.

[00:14:23]I will explain more about this competition after the workshop. Now I would like to invite Mr. Liifan okay to conduct the workshop under Tik Tok blocks.

[00:14:38]Okay, Mr. Lee.

[00:14:40]>> Thank you. Thank you. Uh, Dr. Bang. So, now please allow me to share my screen.

[00:14:46]Give me a moment.

[00:15:00]Okay, cool. Hope you guys can see my screen.

[00:15:05]So today is our uh workshop number two where we will learn how to build AI model or machine learning model using a software called PTO blocks. Also in the meantime we also will introduce another platform is called Google teachable machine. Okay. Uh brief introduction about myself. My name is uh Tiffan or you can call me sir Lee or Mr. Lee. I'm from Chaka. So I want to before I start I want to say a huge thank you to the amazing organizers who made this event uh possible. A special shout out to uh PSTP, KPM, SSTP, JPMA and college nata Fatima uh for inviting me as a speaker today. So uh today students we are taking a huge step from just consuming technology right you are already uh familiar on using technology like on your phones now you are going to create using the technology instead of just playing games or using the apps of uh the other people made right what you are going to learn today is to build your very own AI brains AI model right from scratch like from be from totally zero until you build it. All right, cool. Now um let's go to the next slide. Just a brief introduction at about Chumbaka. At Chumbaka, we believe that technology is a very good tool or great tool to develop life skills or soft skills.

[00:16:41]Think about it. When we use a navigation app like Google Maps, we stop asking for direction. We just using it, right? And also when we play games, we let the game developers to do the thinking for us, right? But now today we are going to flip that around. Today you are the developers. You are the ones doing the thinking, right? You are the one doing the thinking and creating instead of just using the technology. So uh yeah actually Chimpaka have been doing this for more than 10 years. We partner with schools and organizations across Malaysia. So by participating today workshop you are actually joining a very large community of uh young innovators across the nation who are learning to solve uh real world problems by using technology. Right. Cool. Next.

[00:17:40]Now let's get started with a question, right? Imagine you want to build a smart camera. Okay, there's a scenario here.

[00:17:50]You want to build a smart system or a smart camera for your school's recycling bins, right? Or a chat box that stops cyber bullying. So the camera can detect uh whether a plastic bottle is uh put in the wrong pin or not. or a chat board or a chat box that can detect whether a student types a mean or bullying message. So, can you teach a machine to differentiate this questions?

[00:18:23]How does a computer know the difference without eyes or a brain? Right?

[00:18:29]Computers is just a machine. They don't have brains. They don't have eyes.

[00:18:34]Right? I mean eyes are are the cameras.

[00:18:37]I mean they don't have the real eyes, right? So how do they differentiate a plastic bottle and a banana peel or how they how the machine differentiate whether a text is a mean message or a bullying message or a very friendly message? Right. So now Oh, something is blocking my view. I'm going to minimize it.

[00:19:03]Give me a moment. Yeah. Okay. Cool.

[00:19:07]Okay. So now what if you could teach your computer to make decisions like this like shown in the uh scenario above by giving them examples. Right?

[00:19:26]So if anyone is interested to guess how we are going to teach a machine without giving it a brain right and can maybe can someone try to relate has anyone here ever trained a pet before like your like cat at your home or a dog to do a trick before. So how how did you train your pet right uh think about it how do you train your pet to do something right so this is something that we are going to do the same with the computers okay so the learning objective today is that uh there are four discover we are going to discover how computers learn from data through interactive testing [clears throat] and then we are we are going to learn the theory behind of it how machine learning extract patterns it's just like we trend our pets, right?

[00:20:16]We give uh we keep giving them uh the same uh instructions, right? Keep giving them the same instructions and then they can extract the patterns from the instructions and then they can perform the tricks. After that, we will practice to collect the data, train a model. This is where you will start to do a hands-on to build your own learning model. Next, uh we will reflect our learning and then connect our learning to the competition.

[00:20:44]All right, next. So this is kind of like our agenda today is like around 60 minutes learning and then 15 minutes for Q&A maybe. Okay, next.

[00:20:57]So we will start with discover. So we are going to start with this discover phase right before I explain how AI or artificial intelligence. So machine learning is actually part of AI part of artificial intelligence. So I want you to figure out uh for yourself how do we play with it. So I have this activity where we can start exploring this machine learning using this teachable machine. So if you are using uh if you are joining using a laptop it will be wonderful. You can go to this uh website teachable machine with google.com. If it's too long to type, right, what you can do is that you just go to google.com and then you search for uh teachable machine with Google or Google teachable machine. Right? Let me zoom in a bit so that you can see. So this is the platform that we are going to use uh for the first activity to explore about machine learning. Okay. Starting from here, I'm going to go slowly a bit so that you can catch up. Right. Let's enter the website first. Once you enter the website, you will see something like this. Right? On the right hand side, you can see a picture moving. So, it can detect a different patterns, right? Tree or wings, sounds, uh images, right? On the right hand side, right?

[00:22:27]Cool. Let's go back to the slide and see. So what do you need to do next? So uh after the workshop we will share this slide with you. So no worry if you miss out like how uh if you miss any steps right don't worry you can refer back to this slide and uh uh practice or do the redo the activity again. So we will get started by clicking the get started button. Right. Where is the get started button? It's over here. Uh it's over here.

[00:22:57]Uh let me just there's no invitation to my screen. Sorry.

[00:23:07]Okay, cool. So I hope you guys can still see my screen. So if the get started button is over here, what we are going to do is that click on this get started button and then let's go back to the slide and look at the step number three. Uh step number three uh there actually here if you look at here right there are different projects that you can do you can do image you can do audio you can do post but today we will start exploring by using the image project. Okay let's click on the image project. Okay if you are ready if you're already uh on this website click get started and then go to this image project and then there are two options. The second one right is for microcontrollers. This one is for hardware. We are not going to touch this. We are going to choose the standard image model right for most users. So we are going to do this standard image model. Click on it and then you will see something like this.

[00:24:07]Uh there are three columns. The first column is uh is where we collect data.

[00:24:14]Second column is where we trend. You can see trending here. train and then the third one is for us to preview or to test. Okay. Now, um yes, when you first time use this website, your Google Chrome or your browser will prompt you to give access uh to give access to your camera so that this uh website can access your camera to collect the images because we chose the image model, right? So we want to train we want to train the machine using images. In order to capture images we use webcam or there are two option.

[00:24:56]Second option is that you have images ready. You can download the images and then you upload. But today we are going to use webcam. Okay we're going to use webcam. All right. Uh so what's next?

[00:25:11]Okay.

[00:25:13]So this is the activity.

[00:25:18]So you need to prepare two different things. Uh if you look at here the task you need to experience and see if the computer can tell the difference between two different things. So you can uh actually here I already stand by two different things. One is a remote control, one is a mouse. Uh if you can see it's a mouse and then a remote.

[00:25:41]Okay. So what we are going to do next is that let's see the instruction. Okay. So here they say pen. If you have pen with you, you can use pen. So we are going to capture using the webcam 20 to 30 images of object A. So I'll start with the remote control as my object A. So now let's go back to my um teachable machine and turn on the webcam for class one.

[00:26:10]This class one is actually the label uh how you want to label it. Okay. So now actually I'm trying to uh capture the remote control images. So I'll label this as remote control. Okay. And then I turn on my camera. Okay. From here you can see okay you can see my face and then the remote. So when I'm trying to record the images, right, um try to make sure your background is clean. Just shows the remote. Uh I'll do like this, right? And then uh don't forget we need to capture 20 to 30 images, right? So let's let's capture now. So while I'm doing this, you can do the same. You can capture the images for different object. You can try pen, handphone, anything. Right. Okay. And then when you capture the images, try to capture in different view like back of the remote, right? Side view and something like that.

[00:27:19]Okay.

[00:27:21]Actually what we are trying to do here is that we want to teach the teachable machine that uh these all these images uh you need to identify it as remote control. Okay cool after that I can close this remote control done. So what's next? Next same you need to capture again uh of uh 20 to 30 images for track B. Okay. Now you can use your phone or another thing. So I'm now I'm going to uh I'm going to use mouse. I have a mouse with me.

[00:27:56]Okay. Same as a mouse.

[00:27:59]Okay. I've turn my camera a bit so that I can capture a clean images for mouse.

[00:28:16]run two images after that and close it.

[00:28:21]Okay.

[00:28:23]Next, after we capture the images or we collect all the data, what do we do? We click the trend model. We now we can click the trend model to start training the machine. Okay, for this step is going to take some time. If you have a lot of images, the computer will take some time to process all your data. So for first activity, I only capture 30 around around 30 images, right? So it wouldn't it won't take long. Okay, it's quite fast running. It just took like 10 seconds. Okay. Now after you finish the training, boom, step number four, you can test lively uh with your machine learning model. So this three steps actually is uh representing the machine learning training.

[00:29:14]Okay, this one I'll explain later with the next slide. So let's continue with step number four first. Step number four is that we can now test the machine learning uh by holding the objects in front of the camera. As you can see on the right hand side here you can see the output. So in my output I have two objects. One is a remote control, one is mouse. for your case it might be different. Uh but if you can see right now it's showing that um the mouse is 99% uh something is wrong but it's okay.

[00:29:47]Let's just continue with the turn a bit and then now um okay I'm going to show the mouse. Okay, now it becomes 100%.

[00:29:56]The machine recognize this as a mouse 100% confidence. Okay, the percentage is actually the confidence level. Now let me show the remote. When I show the remote, okay, it says 100%. This is a remote control. 100%. Let's show something else. How about this? Okay. Uh something is wrong. It identify this as a remote control. Uh doesn't matter.

[00:30:26]Let's continue with something else.

[00:30:29]Okay. How about the pencil?

[00:30:32]Okay. It also recognize this as a remote control. Okay. How about different mouse? Let me show a different mouse.

[00:30:40]Just now is a wireless mouse. Right now I'm going to show a wire mouse. Okay, cool. It can identify this as a mouse perfectly. Even this is not the exact exact mouse like just now I use for the training, it still can recognize this as a mouse. That's cool. Okay.

[00:31:02]going to go back to our slides and do some reflection. So, uh here's a experiment like what happens to the confidence percentage, right? If you cover half the object or if you turn off the room lights, right? For example, okay, half of the object. Okay, remote. I cover half, I show half only. Okay, it still see as a remote. [laughter] That's okay.

[00:31:31]That's cool. Okay. How about this? I cover like this. Okay. It shows still shows as a remote. Okay. How about a mouse? If I cover, okay, it still can recognize as a mouse because uh as you can see right, we have two objects. The color are totally different. One is white, one is black.

[00:31:52]So the machine learning is very good identify the patterns. This is what we are going to uh do some reflection in the next slide.

[00:32:04]Okay. Okay. Once u maybe I'll pause here for a while for you to finish the testing. Maybe you guys are still enjoying testing the uh machine learning, right? Okay. Cool. I'll give like 10 seconds before we go to the next slide.

[00:32:37]Okay. So that's all for the first activity. Now we are going to see what did you discover in the previous activity. Right. First of all, do you notice that you didn't even write a single program or single code, right?

[00:32:55]For the machine learning, uh you never write a single code to tell the machine like uh what a pen looks like, what a mouse looks like, right?

[00:33:06]And what a remote control looks like.

[00:33:08]Sorry. There's no programming involved in the machine learning process. Okay.

[00:33:16]Next, we also discovered that the machine learning learns from examples, right? You give a lot of pictures like 20 30 images, right? To the computer.

[00:33:29]You either you capture the images or you upload the images. The computer itself will then figure out the difference entirely on its own based on the data you feed them. Based on the images that you feed them. So this is second that you should discover. Next, you also should discover the model doesn't know what is a pen. Just like I show a pencil, right, just now in front of the camera and it detected as a remote control, right? It didn't understand. It doesn't understand the concept of the object like the pencil. It doesn't understand. It actually simply recognize the visual patterns like the colors, the edges, and the shapes, right? But somehow maybe we didn't give enough training or we didn't we didn't have the third class right just now it make mistake when I use a pencil right so in order to make your machine learn machine smarter what you can do is that you want the machine to identify more objects what should you do you should try more classes this is what we call classes add a class you can add a class and then you can name it pencil so that it won't identify the pencil wrongly again as a remote control. Right? Just now it made mistake. My machine made mistake, right?

[00:34:47]Because it doesn't know what is the pen.

[00:34:48]So you need to teach the machine. Uh the the machine hasn't learned about pencil, hasn't learned about pen. So they don't know how to identify it yet. So it will wrongly identify it as a wrong object.

[00:35:04]Okay. So this is the third thing you should discover. Next, sometimes the confidence will uh make mistakes, right?

[00:35:13]Uh just like if you change the background, if you manage to change the background, like for example, I use myself as a background and then I put this in front of me. Uh you can see the confidence level like go up and go down, go up and go down, right? Fluctuates and mistakes happen. So this is something that you need to take note about. Uh take note also when you trend your machine learning model. Normally what we do is that uh in order to not to get the AI get confused right we will try to use a very clean background color background to train the machine. Okay we just want to show the objects. If you show ourself inside the background right if the machine will think oh this thing including me this whole thing is the remote control that's why we don't want our background with noises the noises mean different kind of things uh in on in the screen when we train the machine so we want we want to make sure our background is clean mean there are no other things okay when you machine okay this is for the image images. Okay, next. And you also notice the training text time, right? I um for my my case, I only uploaded or I only capture 30 plus images. So, it quite fast like 10 seconds. But if you try yourself at home to increase the accuracy, right, you can train 100 images or 200 images and then it's going to take a longer time. Uh but you also notice that the prediction is instant. Uh so when we do when we put our object in front of it the prediction on the right hand side right the output will instantly change.

[00:37:03]Uh so this is the last thing that you should discover. The trending takes time but the prediction is like real time right like immediately.

[00:37:15]Okay. So basically the core idea is that the AI or the machine learning automatically discovers the visual features like the colors, edges and shapes. You don't need to tell it or you don't need to program it to uh to let the machine what to look for, right? The machine will figure it themsself with all the data you give to to it to the machine. All right? So these are the things that you should discover and learn about machine learning. So next uh let's learn about uh some theory uh behind of the AI actually how computer process the information uh in order to understand that uh I put this traditional programming as a comparison so that you can see more clearly what's the difference. So for a traditional programming right uh if you learn about programming before uh uh either Python programming in Python or Scratch or in other programming language any other programming language what you do is that you give the computer instructions and you give the computer data and then the computer will give you the answer.

[00:38:32]For example, uh you program you program like this.

[00:38:36]You say if pixel is red, if the images is red and the shape is brown, then the answer is apple. So this is traditional programming. Uh you let the computer knows the color and the shape so that it can identify the image as an apple. But is the problem is that is impossible to write instructions for every cases or every visual appearance right for every lightning condition or angle it's very is quite impossible to do that. So so this is traditional programming. So here we come with the machine learning. So basically in machine learning what we do is that we give the computer answers labels just like just like what we did just now we you tell the uh machine learning this is remote control of course you want to name this as something else like Apple also can but of course the label you know that is wrong right? Uh so label the answer we label it as uh remote control. So basically this is the label. This is the answer. We give the machine the answer also. Then we give the data we give the images. The data are the images is the images. After that the computer will finds the patterns by itself. The model after we train the model the computer will finds the patterns by itself.

[00:40:08]For example, uh we show 50 pictures of apples and then the computer will find the pattern by itself. The core idea is here is that the machine right it won't go and memorize the exact pictures. Uh they won't go and memorize oh this apple this apple they won't memorize. What they do is that they identify the features like I I keep saying right the shapes the color the color patterns the textures by identifying these features on the um right hand side you see here the key learning point here after they identify the features it will learns from the pattern and then once you show a new picture just like I show a new mouse right I show a new mouse here it still can take it as a mouse so This is the new mouse. This is the mouse that I use for training. Right? Even I show it different mouse, it can still identify it as a mouse because is already uh know the pattern of a mouse. So basically even if you show a brand new Apple or brand new pencil or brand new remote control that the machine never seen before, it still can guess it correctly.

[00:41:25]So this is what we call machine learning. All right, cool. I hope you guys can catch up with me.

[00:41:36]Now let's summarize the three steps of machine learning.

[00:41:42]The [clears throat] first one collect and label. Right? What you do?

[00:41:47]You what do you do first for a machine learning? First you need to collect the data. uh a good data is very important right if you get a very blurry image then the machine will learn it in a blurry way right okay so first you start with collect you collect the data and then you label the data you label it as as it named okay next you train the machine learning and you train the machine to recognize the patterns lastly you test and iterate right you test If you find out that the machine learning model is making mistakes, what you do? You can re upload the images. Uh delete the images that are not in good quality. Upload new images, tech new images, and then train again. Then test again. Make sure the machine learning can give you a very good result until it's satisfied.

[00:42:45]Right? Only then you can uh export and do and do some programming. This one we will explore in the next activity but not using teachable per machine. Uh we are going to use PTO blocks. Right.

[00:42:58]So these are the three steps for a machine learning. Basically start from left to right. You collect data, you train, then you export and not export.

[00:43:07]You preview first. You test it. You test whether it's accurate or not. If it's not accurate, you go back to the first step. You delete the images that you think is not good enough. then you take a new images. So here is very important.

[00:43:21]They tell you that good data is the most important step. Okay. Model failing.

[00:43:26]This means that uh so here's another hint. If the model is failing, so this means that it hasn't learned the pattern properly, right? So you must observe the output and see if there are any errors.

[00:43:40]Then give better data. Train again, test again, train again, and test again until you satisfy.

[00:43:48]All right, cool. That's all for the discovery part and the learning part.

[00:43:54]So, we have done the discovery and the theory. Now, we are going to do some hands on learning again.

[00:44:02]Here I have three challenges.

[00:44:05]I have three challenges. Let's go through one by one. Challenge number one, we are going to train our webcam to classify the whole frame. Whole frame, right? Including myself in the frame.

[00:44:18]Whether I'm holding a cup or bottle or is just my face. The the the purpose of this application is to uh to track whether I get enough drinking water or not. Okay. The second challenge is to identify whether a text is a mean message or a bullying message. So we are going to train an AI to classify the sentences as friendly or toxic. Okay. So the next one so uh I think if you have enough time we are going to go through all the challenges.

[00:44:55]If you don't have enough time, we will cover two challenges. And then the last challenge, you can uh do it at home at your self pace. And then the last challenge is the accessibility helper, right? It's like a sign language uh helper. Uh we will train an AI to identify sign language with uh it will identify thumbs up or a peace sign.

[00:45:21]So these three challenges are quite simple because we only have uh two classes or two labels, right? The AI will only uh learn to identify two classes or two labels. So uh because it's a very short workshop, we are not going to make it complicated. Of course, after the workshop, you can expand the applications. Okay. Now practice one. [snorts] So uh we will start with the uh hydration tracker. So first of all we are not going to use uh teachable uh Google teachable machine anymore. We are going to use a software called big top blocks. So I'm going to close this uh teachable machine.

[00:46:08]So uh just in case if you want to save your project right you can come here and save to your drive if you want. if you want. So I'm not going to demo this step. Uh today we will focus more on PTO blocks. So there are two ways to use PTOB blocks. The first way is use the uh web version. You can open it using your browser with this website ptos.ai.

[00:46:36]Okay. This is the first method. The second method is that you can go to Google and then you can search for download VTO blocks.

[00:46:47]You can use the desktop version as well.

[00:46:51]Uh if you want but today uh but the PTO blocks software right is actually quite large. So today we are going to use uh web version is more convenient. Okay. So let's go to Pikto blocks and then we choose blocks blocks or actually you can go uh we choose blocks first don't worry we choose blocks then you will see a user interface totally similar to scratch right it looks exactly like scratch uh the programming uh language that it use is block based just like scratch so it's not too complicated to understand. Okay, first of all, what we are going to do is that we need to add a ML environment.

[00:47:44]What is ML stand for? ML is actually stand for machine learning. So in order to uh add ML into our victory blocks, we need to go to extension. Uh if you if you try Scratch before, right, you know that where to get the extension, right?

[00:48:02]is over here left hand the left hand side bottom corner here. Click here and then from here you can see different uh extension that you can use face object but today we are going to focus on machine learning environment because under this machine learning environment it is uh quite strong enough to cover different kind of uh machine learning models. Let me show you. Okay. Once you entered the machine learning environment, it will show it will show at the bottom here after my blocks.

[00:48:39]After my blocks, you will see machine learning environment over here. Right left hand side here.

[00:48:47]Okay. After that you open ML environment. Click this open ML environment form. Okay.

[00:48:59]Okay. I have a project here but I'm going to delete this so that I can start over. So once you open the machine learning environment I think I should pause it here for a while so that you can manage to come here manage to open the website. Okay the website is Piktoblocks. Uh I'll type here Pikto blocks. Maybe I'll show here so that you can see uh the website is this pictoblocks.ai.

[00:49:32]Okay, I'll give some time you guys to go to the website pictoblocks.ai and maybe 15 seconds here.

[00:49:59]It's best if you're using a laptop. It's more convenient.

[00:50:08]Okay, cool. Times up. Hope that you already open the Pikto blocks website again. Let me repeat where to get the machine learning environment. First you need to go to extension here at the left hand side corner at the bottom here.

[00:50:23]Click add extension and then look for machine learning environment. Uh look for machine learning environment and then click it to add it. Okay, mine mine already uh so it shows the extension is already added for my case is already added. Once you add that, you will be able to see uh the machine learning environment blocks just be just after my blocks.

[00:50:51]Okay.

[00:50:54]Okay. Once you are ready, okay, I'll go a bit slower so that you can follow for the first project. So once you are ready, again, click this open ML environment. Click this.

[00:51:10]Okay, it will bring you to this uh user interface where you can create a new project uh where you can create a new project.

[00:51:21]So click this create new project form.

[00:51:25]Okay. After that it will ask you to enter project details. Yes, you have to enter the project name. If not you cannot proceed.

[00:51:34]So uh so I'll just put AI ready Ashure then I'll put image because uh we will start with the image model because here right they have different machine learning models.

[00:51:51]We can train the machine using image.

[00:51:53]That means right we can train the machine using images or we can train using pose uh or even hand pose or even audio the sound or numbers right CR this one if you want to learn more we are not covered this today but if you want to learn more you can try to Google they have this tutorial online I mean the top blogs got the tutorial online and then we have this text classifier so today We will we will cover image. We will cover text classifier. Text classifier where you can uh differentiate the sentences, right? Whether is friendly or toxic.

[00:52:35]This is what we are going to do in challenge two. And then in challenge three, we are going to use hand pose uh for the sign language. But now we are going to choose image classifier first because we are going to train the images first so that it can identify what again. Let me show you. You can identify me whether I'm holding a cup or a bottle or it's just with my face without anything. Right. Okay. Now, select image classifier. Of course, you need to put your project name and then lastly click this create project uh create project form, right? Then again, it will bring you to something similar to the Google teachable machine. They have three columns as well. Three columns stand for three steps. The first step collect collect the data. Right? You need to fit the machine with the data. Right? First step. Then second step you can train the model. Last step you can test the model.

[00:53:39]All right. Cool. Now um again same we can upload or we can do webcam. So now I'm going to use webcam to um yeah to take pictures on myself with a um with a what? With a bottle, right? Okay. I'm going to Yeah. No need to open that. I'm just simply do like this. Okay.

[00:54:00][laughter] Okay. Around 39 images and then go back.

[00:54:08]Okay. Yeah. The images are here. Okay.

[00:54:11]Okay, if you want to delete, you can see a dust bin icon here that you can delete individual images if you want. Okay, so this one, right, I'm going to uh what should I name this?

[00:54:23]With a photo, maybe. Okay, then this one I'll just uh with nothing maybe. All right, cool. So, this is just myself uh watching around, maybe doing my job, right? Looking at the computer, right? uh doing my job. So just picture of myself, right? Okay. So now I have two classes or or two labels. One with a bottle, one with nothing. Right. After that I'm going to train. Of course there's a advanced uh setting here. But we are not going to touch because the default setting is already good enough unless you are training a very complicated images that you want to tweak a bit. uh but today we are not going to touch this okay because we will cover the fundamental first so I have clicked the training button so yeah it's training training taking some time few seconds up to 1 minute okay it ask me to wait okay cool okay it says motor accuracy is good that's good now I'm going to test the image by webcam, right? So, where's the output? Okay, with nothing. Okay, let me look around.

[00:55:43]Okay, it's quite accurate when my hand go up. Okay, it fluctuates. Uh, don't worry about that. Okay, later if you're not satisfied, you can modify. Okay, this one. Okay, bottle pass. Okay, right. Quite good. Quite good. So, what we are going to do next is that let's go back to the slide and see the next step.

[00:56:06]Okay. Oh, so create class holding cup but I do it differently but I already done and then create class two with no cup I already done as well with different naming it's okay. Then uh we tren the model and test it. Okay I have done until this step but I haven't do the export model.

[00:56:26]So next we are going to export the model to our programming space so that we can do some programming. Okay, let's export the model here. Click here, export. And then we choose don't choose Python.

[00:56:40]We'll use block uh block block programming. They don't allow us to choose Python as well. So yeah, the only option is block based programming. So we'll do some simple programming here.

[00:56:53]If you are already familiar with Scratch, right, it will be very fun here because you can integrate AI in your in your Scratch project. If you love to build scratch animation, stories or even games, right? Uh now is your turn to integrate your a integrate AI in your project. So now we are going to cing the logic. Okay, let me explain the logic first before I uh program it. So now right once I export it right, you can see some blocks available over here. Uh a lot of blocks actually uh all these block is for the Okay, all this block is for the machine learning that you can use later. Ah, okay. Let's delete it. So, let's understand the logic before we start programming. So, now I'm going to build a virtual assistant that will remind me to drink water. uh maybe like after 30 I don't know maybe 20 minutes or 10 minutes it never detect me uh taking any bottle then it will remind me something like that. So example logic is given here. So yeah this is a pseudo code uh if you are uh I'm not sure maybe ASK student you learn about pseudo code. So this is pseudo code or the logic given when green flag is clicked then what we do we do we turn on the video on stage with 50% transparency blah blah blah forever then do this do this do this right okay now we are going to do step by step we'll start with when green flag click right so where do we find the green flag green flag green flag is over here when green flag it's just like scratch when you want to start your program a story uh you use green flag right so that you can press the green flag to start your project or your program. Next, uh we are going to turn on the video on stage with 50% transparency. So, it's over here. Oops.

[00:58:58]Turn on video on stage. Let me make this smaller a bit.

[00:59:04]How should I do it? Uh how about this?

[00:59:06]Okay. Turn on video on stage with Let's try zero transparency first. Okay. Zero transparency means that you can clearly see my face, right? But um it's look weird to see my whole face here. So I'm going to increase a bit the transparency.

[00:59:27]So with the transparency increasing uh you can barely see my face or you can even I can increase more 60. Uh if you if you don't like to see your face, you want to do it secretly uh you can put 100%.

[00:59:42]Right? But today we are going to use 50.

[00:59:45]Okay. Let's just follow the logic given.

[00:59:48]Okay. After that what should we do? B.

[00:59:52]So that's a forever loop. Okay. Forever.

[00:59:55]Where should we find the forever? It's is under control. Control. Okay.

[01:00:02]Let's take out this forever. And then inside we are going to do some uh if else logic to match some decision after the AI detect or predict something. So forever we are going to analyze the image from the camera. So this one we will take it from the machine learning. Okay. So analyze image.

[01:00:26]Please excuse me. I'm going to drink some water.

[01:00:34]Okay, analyze image inside the forever.

[01:00:38]What what what happen? Okay, it will analyze and then it will identify whether it's a bottle with a bottle or with nothing. So is identify as false, right? And then with nothing is true.

[01:00:52]But what if I hold my bottle I and then I analyze again? Analyze again. Let's check the result. Okay, if a proper is true then nothing is false. Okay. So this is something that we can use to program uh our project. Right. So after analyze we can do if else if else uh something like this. So I'm going to take the if else block.

[01:01:19]if else and then if it is identified with I'm going to make this smaller so that you can see clearly the program if is identified as with a bottle or else if identify as nothing will do something inside will do something here what should we do we can use some uh save block Maybe. Okay. So, it says uh great job. Staying hydrated.

[01:01:54]Okay. Good. Uh it say two seconds. Okay.

[01:01:58]Great job.

[01:02:02]Staying hydrated.

[01:02:05]So, so the program will encourage me to drink um water. Okay. I think I missed out something here. The image analyze also you don't forget about this. So it will keep analyze the image. Analyze the image. Okay. If you if you never drink the water like with nothing, right?

[01:02:25]Okay. What should you do? It says don't forget to drink water. Okay. Don't forget to drink water today for example.

[01:02:36]Okay.

[01:02:37]Something like this. So you have two program. The first program will identify if you're holding a bottle, you are drinking, you're drinking, it will say, "Oh, well done. Good job." Right?

[01:02:48]Staying hydrated for staying hydrated.

[01:02:51]If after long time, uh this one after long time, the logic is a bit different.

[01:02:56]But we'll just do a simple one like this. If it detects nothing, it will ask you remind you to drink water. Something like this. Okay. Now we are going to make this bigger so that we can uh run the program and test it out.

[01:03:12]Let's click on the green flag and test it out. So we have this cute uh Toby here. His name is Toby. Um yeah. Okay.

[01:03:22]It will say don't forget to drink water.

[01:03:24]Don't forget to drink water. Okay. It will keep remind me. Okay. Remind us.

[01:03:28]But now what if I holding my water and attack? Okay. He say, "Great job staying hydrated."

[01:03:39]Yeah, I need to take some water. Okay.

[01:03:41]Okay.

[01:03:43]Great job staying hydrated. Don't forget to drink water. Uh, of course, you can implement more uh complicated logic here. For example, you can set a timer for those who who know how to program using blockbased, right? Okay, you can set a timer here. Ah maybe you can keep counting like for example 10 minutes after 10 minutes then uh you keep checking checking uh then after 10 minutes if if you never identify once in within 10 minutes then you will give reminder so this is something that you can try to do but um um for this complicated programming I'm not going to cover this I'll let you to explore all right of course if you have any technical issues you can contact me after the workshop right uh I have my phone number leave at the end of the slide okay so is this is uh project number one we use image classifier next we are going to try the second classifier or the second machine learning model I hope you're still with me I hope I'm not going too fast okay of course if you want to save your project right Yeah, can click file and um say uh save to your computer, right?

[01:05:06]Because this is I I never sign into my account so it won't save uh uh in the cloud. So if you want you can create account and then try out to save your project uh just like scratch.

[01:05:19]Okay. All right. Cool.

[01:05:23]Let's take a deep breath.

[01:05:26]Yes. And then we are going to explore the second project.

[01:05:31]We are going to build a project right that can identify or classify attacks whether is uh friendly or toxic.

[01:05:39]Remember uh let me go back to the challenge so that you can recall again.

[01:05:44]So the second challenge we are going to train an AI to classify the sentences as friendly or toxic.

[01:05:56]>> [clears throat] >> Okay.

[01:06:00]Okay. Project two.

[01:06:03]So now um I'm going to like uh close I mean split take the program out. Just put one side. Okay.

[01:06:14]Um yeah. What should I do? Um yes I need to go to the machine learning environment. Then I need to choose text classifier.

[01:06:25]Let's see if I can do it in the same project. Open the machine environment.

[01:06:32]Okay, here the uh image classifier that I built just now. Oh, it didn't capture the symbol. So, actually we cannot name the project with symbols. Um that's fine. So AI already assign then I'll put text classifier right and then we are going to choose text classifier in this classifier right we don't need camera okay let's choose text classifier and then create create a new project yeah okay this is something that uh I wanted to test as well And uh it proved that uh my assumption is actually correct. We cannot build two machine learning in one project. So what we are going to do is unfortunately uh we have to delete the previous uh block first.

[01:07:29]Okay. Please wait.

[01:07:33]Okay. Let's delete everything and try one more time.

[01:07:38]Create new project. Uh I text classifier then create again. Yes.

[01:07:50]So this is something that you guys need to take note as well. Unfortunately they cannot support uh two models running at the same time. Okay. But it's fine.

[01:08:01]Okay. So if you remember right the two labels uh the first one is friendly message friendly. Okay. friendly and then the second level is uh toxic toxic messages. So um for this one right we don't the data we wanted to collect is not uh is not image anymore is text uh you can type text like this text okay so uh yeah so how to uh what are the data that you can provide for example I have a few examples here for example right for the friendly class You can put you are doing a great job. Okay. Let's try you are doing a great job for example.

[01:08:46]Okay. And then you press enter. Uh you press enter. Then this is the first uh data already. Okay. Is is one. Okay. I don't think one is enough. Let's try train the motor and see. It says friendly should have minimum five samples. So for this text classifier you should have minimum five data. Okay, you should have minimum five data. So actually here I have a few more examples. I love this project for example. I love this project.

[01:09:21]Okay. What else? Let's be friends.

[01:09:28]Okay. What else? Can I help you with that? Uh, can I help you with that? What else? This is awesome.

[01:09:42]Enter. What else? Have a great day.

[01:09:49]Uh, what else? You are so smart. For example, any Oh, that's all the examples. Next, we're going to the toxic class.

[01:10:00]So toxic class for example, you are ugly. Oh, so toxic.

[01:10:07]I had to. Okay. All these bullying messages. Nobody likes you. Oh no.

[01:10:17]Or else you are stupid. All these toxic toxic uh uh sentences, right? So you need to provide these examples so that the machine can identify. Okay. Stop talking to me. Uh, it's very mean. Okay, this sentence is very mean. No good. You are the worst. Oh, no.

[01:10:39]What else? Give up. Now, what else? Okay.

[01:10:46]Okay, cool. Of course, you can give more more and more uh data to your machine.

[01:10:52]But uh now I'm going to train first with all this data. Oh, it trained like like instant. Oh, then done. Okay, after that I thought it were going to take a few seconds. No, but it train like for instance.

[01:11:10]Okay, now we can input here so that we can see the output. What if I say good morning? Ah, good morning is a friendly.

[01:11:20]Okay. Uh, what else can we try?

[01:11:24]Uh uh what should we try? Uh how are you?

[01:11:34]How are you also friendly? Okay. What else can we try?

[01:11:40]The project looks bad. For example, oh this one they also uh identify as friendly means that when you do your testing right on your machine learning model with different different data or input you find that the testing give you a not accurate result. What should you do? Ah you can add more data to your machine learning. So the project looks bad. Okay. So yeah add more data then tren again and then try one more time.

[01:12:13]Ah, after I try again. Okay, now it goes to to do. Right. So, this is what you should do. Uh, you train, you test. Uh, after you test, then you test with different sentences, examples, right?

[01:12:27]And then, um, if you find it not accurate or not cover enough, add more data. All right. Add more data. Okay.

[01:12:37]Now I'm going to export now to the blocks so that I can do some programming.

[01:12:57]Okay, it's uh taking some time to be exported. Let's be patient.

[01:13:08]Okay, motor model loaded successfully and [clears throat] only one test classifier is given.

[01:13:18]Okay, I'm going to turn off my camera.

[01:13:21]You can actually turn on and turn off here using the camera icon here because this uh model they don't need a camera.

[01:13:31][clears throat] Okay. Now, yeah, I've done train, I've done the testing, and then the export.

[01:13:37]So, now I'm going to do some coding with the uh machine learning model.

[01:13:44]Okay, use block. Uh uh I'm not sure if you if you are familiar with scratch, right?

[01:13:53]You should know about the ask block. Uh many of you use this for different application, right? Like for example, you use this for your quiz game. You use this for your um uh story project, right? To ask a question. All right.

[01:14:11]So this one right uh they will ask user to type a message then AI to check for it whether is a a friendly message or a toxic message. Okay. So this is the pseudo code or the example logic.

[01:14:27]Again, it starts with a green flag.

[01:14:30]Okay, the very first code. And then they have this forever. Okay, let's take this out. Uh green flag is under events.

[01:14:38]Under events and then under control, you tag the forever.

[01:14:47]Okay.

[01:14:50]After that, okay, ask type a message for the chat room. Okay.

[01:14:59]chat room. We are actually trying to simulate a chat room uh in this video blocks. Of course, if you are uh advanced enough, you can actually integrate this in your uh telegram. Uh if you uh want to explore, you can go to YouTube and try to find out how, right? Okay. And then analyze the text. uh the text is actually from the answer. So this ask and answer is a question answer p whenever you ask a question right for example let's turn this on turn on the answer you can turn on and off here with the thick chat box on and off [clears throat] let's uh type let's run this program and simply type anythingah okay the answer will be store inside it okay if you ask again and then you give different answer uh good for it will replace. So [clears throat] this answer will only be able to store one value.

[01:16:06]Okay. So whenever you ask a question and the user type for an answer then you will get this and then you will try to um run it with the machine learning model to identify whether it's a friendly or a toxic message.

[01:16:26]After [clears throat] that you use if else just like just now.

[01:16:30]Uh okay let's take out if else and then if [clears throat] then comparison equal.

[01:16:42]So it could be toxic, could be toxic, it could be uh friendly.

[01:16:56]Friendly typo and then we put okay toxic output at the bottom then I put here. So now we have two program one one is friendly one is toxic. Okay.

[01:17:16]So what should we do for friendly? Uh we will display the user message.

[01:17:22]Users say for example right user say and then we join with the answer. Try answer.

[01:17:38]Okay, it's friendly. Then we display the message. We displayed it like this. Uh let me zoom it and cancel this. Okay, we will display it. Users say good. Let me put at the middle so you can see. Users say good. So means that is t friendly.

[01:17:55]But just how I test it is it actually get to We have to improve our machine learning and it says good is a toxic work word that means something is wrong but it's okay let's just put first and then if the sentence is detected as toxic what we'll do what we do we do this one message block please use kind words okay we'll say message block please use kind Okay, now we are going to test it out. Okay, so this is a very simple program. If you like not able to manage, if you are not managed to follow, you can take a screenshot then you can uh practice again or rewatch the recording. Okay.

[01:18:49]Now type a message for the chat room.

[01:18:52]Let's type something. Hi, how are you?

[01:18:58]Okay, it runs too fast until we cannot see the result. So, probably right. Uh oh, it says for two seconds, but I'm going to put like uh here. Wait for it go for the second round because it's in a forever loops, right? Before it go to the second round, I'm going to wait for 5 seconds. Okay, run one more time. Hi, how are you?

[01:19:23]Message block. Please use kind words. Uh this means that something is wrong with our uh machine learning model. So if you see if you after you test it out uh for your project and you find something is wrong, you want to modify your machine learning, what should you do? You should come to this machine learning environment and then go to edit the training. Uh let's edit the training a bit.

[01:19:51]Okay. So just now it's like uh how are you? Right. Uh let me test here. Hi, how are you?

[01:19:59]Why they detect this as toxic? Uh I don't know. But I'm going to uh add more text. Can I add more text?

[01:20:07]Um it's here. How are you enter? Yeah, let's try one more time. Train and then try uh still detect. Is it because of the high? Let me delete the high.

[01:20:24]It's still display as toic. Maybe let me run the trend model one more time.

[01:20:29]Trend. Okay. How are you? Huh? It's weird. Is this something wrong here?

[01:20:39]I should trend high as well. Hi. Okay, let me trend high. Okay, high is good. Working fine.

[01:20:47]But I when I put how are you? Okay, now it's fine. How are you? Sorry. Okay, now it's fine. Okay.

[01:20:56]Okay, cool. Okay, no more. No more bug.

[01:20:59]Okay, I'm going to No need to export. I think we need I don't think we need Yes, because we already exported just now, right?

[01:21:18]Okay. For your process.

[01:21:36]Okay, cool. Let's [clears throat] try one more time. Hi, how are you?

[01:21:42]message block again. Okay. Um, hi. How about hi? Message block again. Okay.

[01:21:48]Moral of the story. I think we need to export the model one more time.

[01:21:54]>> Okay. Lesson learned. The model is not up uh updated immedi uh by themsel. What you need to do after you retrain your model, you need to export one more time.

[01:22:07]The export again.

[01:22:23]Okay, hold your horses. It's running.

[01:22:26]Running.

[01:22:33]Okay, cool. Let's try this. Wow.

[01:22:35]Running. Hi. Okay, finally. How are you?

[01:22:41]Okay, user say how are you?

[01:22:45]Okay, I hope you guys still can hear me because it's rains very heavily uh just outside my office. Okay, u that's all for the uh text classifier. Okay, now try to imagine where else you can apply this, right? Not only uh friendly or toxic word uh you can also like uh if you are in the telegram group you will notice that there are a lot of spamble spammers right they send spam messages so you can try to detect those spam messages and then uh you can delete it if you want to try to integrate with the telegram maybe this one is for secondary school level or form six level you can try to explore Okay. [clears throat] Okay. Um yeah, that is all for the challenge number two. Um seems like we still have some time. So I'm going to uh demo for the sign language our uh our last model that we will cover today.

[01:23:50]Okay, which is the hand post classifier.

[01:23:54]Okay, so again uh I'm going to delete everything. Say bye-bye. So yeah, I'm going to go to uh ML environment, machine learning environment and I'm going to create a new project and then yes, I'll name this as hand post and then I'll choose this uh right hand pose classifier and this one they need a camera. Okay, actually is very fun to play. Let's see what we get. Okay, hold our horses.

[01:24:39]Okay, cool. Of course, uh we want to collect data, we can use upload, we can use webcam. Okay, let's see what uh classes we want to create. Okay, thumbs up. Okay, let's name this as thumbs up.

[01:24:54]Okay, webcam.

[01:24:57]Okay. Okay. See this. Ah, they can detect 21 joints, right? Is it 21 4 5 30? Yes. 21 points or joints of your hand, right? So now we are going to do thumbs up. Okay.

[01:25:16]We use hand because my right hand have to press the button.

[01:25:22]Okay. Thumbs up. Thumbs up. Thumbs up.

[01:25:24]Thumbs up. Okay. With different angle.

[01:25:28]Okay. If you if it lost detection, right, of the drawings, you open your palm again. And uh yeah, come over.

[01:25:36]Sometimes if your lighting uh your lighting in your room is very dark, right, it cannot detect very well. Like my room now is quite quite quite not that bright.

[01:25:50]Where's my thumbs up?

[01:25:56]Okay. Is 32 enough? Okay. Yes. They need only minimum 20 samples. Okay. Thumbs up is done. What's next?

[01:26:09]They need a P. Okay. P.

[01:26:21]Peace.

[01:26:29]Okay. Okay. Done. Peace. Right. Cool.

[01:26:33]Let's train.

[01:26:37]Okay. It's going to take some time.

[01:26:47]Right. Cool. Okay. Again, we have to testify. Thumbs up. Yeah, 100%.

[01:26:55]100%. But sometimes uh they they cannot detect the joint right. Thumbs up.

[01:27:02]That's why sometimes right when you use real life application you have to do multiple times. Right. Okay. Okay. I'm I'm curious what if I use right hand what will right hand thumbs up.

[01:27:22]Okay, the piece still working. Thumbs up.

[01:27:42]Oh, the thumbs up still working. That's good. I I put some light on it so that you can identify, right? Because it's too dark. So, I put some light. Okay.

[01:27:50]Cool.

[01:27:52]Okay. Everything is done. Okay. Again, same same process. We export the model to our programming space so that we can do some programming using the the model that we trained just now. Cool. Uh yeah, we are going to uh Oh No, sorry.

[01:28:08]My bad. My bad.

[01:28:11]The logic. Ah okay. Okay. Can pose uh AI with text to speech. Uh text to speech is this one. If you want to make some sound uh text to speech, text to speech. Yeah. If you play with uh Scratch before, actually I think you should familiar with this where it can use uh some uh text to I mean the function to read out the text with different languages as well. Okay. Um but um to make it easy to for you guys to see I'm going to use the same block the say the say block and then I'm going to turn the answer off.

[01:28:53]Okay. Again the logic. Okay. Okay, when green flag click again you start the green flag. Okay, the video because this one they need the camera right again with 60%. And then don't forget about the green flag. This one don't deliver us later. We're going to use it and then of course they have this forever right uh it's quite similar to the image classifier. I think they need this right analyze. You need to analyze first only then you have the if else right? If else. Where's the control? If else if else. If so, what are the classes available?

[01:29:34]We have two. If thumbs up, if is identified as thumbs up and if it is identified as peace, what you do? The program is actually very similar to the image classifier just now. Very very similar. Actually almost the same.

[01:29:54]Okay, thumbs up is good job. So this one good job.

[01:30:00]Thumbs up is identify. We will say good job. Okay, then peace. Peace.

[01:30:07]Peace.

[01:30:09]Okay, cool. So now we are going to test it in full screen. Oh, it's a bit laggy.

[01:30:17]Okay, I'm going to put this toy on my shoulder.

[01:30:21]Okay, run.

[01:30:24]Peace.

[01:30:26]Okay, good job.

[01:30:29]Good job everyone. Hope you can follow.

[01:30:32]Hey, you cannot detect and it detected SP is something wrong means that you need to improve your machine learning model. Okay, good job please. Good job.

[01:30:42]Good job.

[01:30:47]10.

[01:30:50]Peace.

[01:30:52]Good job.

[01:30:55]Peace.

[01:30:57]That's cool. Okay. My motors I think is working fine. Just that sometime it makes mistake. Okay. So what should you do? You should uh trend more if you want to avoid mistakes, right? Trend more uh tren the machine learning more with more samples with more data. Okay. With more images.

[01:31:16]Okay, I think that's all for the machine learning model. For the application part, uh for this one, right, uh you have to uh let us show a few examples that you can think about where you can apply all these machine uh learning models. Okay, now let's go back to our slides. That's all for the practices.

[01:31:37]Ah, right. Cool. So, now it's time for reflection, right?

[01:31:45]Okay. Um we have tried three different machine learning models. Of course it's not just uh there are more.

[01:31:54]Of course we don't have enough time to cover all. So what you can do is that after the workshop you can try the sound if you're interested like like if you have iPhone or whatever uh Google devices right you may try uh hey Siri uh something like that right or hey Google turn on the light of for example anything like anything similar to that uh you can apply it using audio classifier okay you can try this at home right it's actually quite fun as well okay um yeah Let's come back to this reflection, right?

[01:32:32]Okay.

[01:32:34]Um again uh for this workshop is to prepare you for this AI uh data lab challenge competition where you need to identify a problem and then uh you need to find the data and then display the train to prove that the problem is exist. Right? That one will be covered in workshop three for example right if you are saying that uh uh you are talking about accidents right you need to find data about accidents like uh then display it using graph and chart to prove that the problems uh to to prove that the problem is after that you use machine learning model uh to solve the problem okay how can we do it first you need to think about um what specific problem are you trying to solve.

[01:33:29]After that you need to think about which machine learning model is best for your problem. Do you need image? Do you need uh text? Do you need hand pose? Uh for example, hand pose. You are helping uh uh uh some somebody that cannot uh talk right with uh these difficulties. people with these difficulties, right? You're trying to build sign language to help them. Uh probably you don't need image, you don't need text, you need hand pose, right? Okay. What else? You can also try sound, right? Like just now I said the uh hey Siri to do some uh application using the sound control.

[01:34:11]Okay. So the second step is that you need to choose the base uh base model for your problem. Then number three, uh how are you going to collect enough data to make sure that your model works in real life, right? Because like for example, just now when we train our image classifier, right, we want to train it using a empty background, a very clean background. But in real life, right, when you're trying to point the camera to the c uh to the mouse, for example, to the remote controller, you may uh have different background. Am I right? It's not it's impossible the the mouse will be uh in the same background as what you uh as what in your images, right? So, you need to collect enough data to ensure your model works in different scenario, right? or different objects like for example mouse mouse yes you use the mouse to train but that what is your mouse how but how do you make sure the model also works on other people mouse as well something like that >> and then uh what virtual action because in our competition we just cover software we didn't cover hardware so in your software you don't need to use any hardware so in your software what virtual action will your pto block if you are using pto block software. Uh, what virtual action will your Pikto block script trigger when the AI detects something like for example display a warning or uh changing sprite, playing sound, reading text, uh something like that. Okay, that is four things that you should think uh to relate the machine uh machine learning model to your competition project.

[01:36:08]Okay. So basically in your competition right you need to solve a real life problem uh using machine learning model.

[01:36:16]Whatever problem it is uh try to look around of your community of your uh uh the place that you stay try to observe any any problems that you think is worth to solve or is possible to solve using machine learning or not. Okay.

[01:36:34]Next um here are some real life applications idea that we can use picto blocks to uh to to apply uh that you can take it as a inspiration or reference. Okay. So basically uh again we don't use extra hardware you can uh build amazing software solution even without extra hardware. For example, virtual study buddy, right? Uh this is a virtual assistant where it can detect whether a students is falling asleep or not or Yeah. And then it will place a friendly reminder, right? So the camera so maybe the students are walking listen to the music in front of the laptop, right? Uh so if it detects students feeling asleep uh falling asleep then it will uh play a reminder sound something like that.

[01:37:27]Next. Okay. Mental health journal. Okay.

[01:37:30]Text classifier. So, uh it's a diary app. Okay. Uh basically is a diary app that can analyze uh users daily entry whether it detects any uh uh abnormal patterns like for example sad or anxious right. So that uh the sprite can offer something or take some action to convert the the the people or the students or the users uh or even uh or even uh send a message to the parent or something like that.

[01:38:04]Okay. Next hand post classifier. This one is sign to speech, right? uh an app that uses the webcam to read sign language and then use the text to speech to speak the words for people who don't know the sign language. Uh this one is for accessibility.

[01:38:22]Okay. Okay. Next uh this one very famous project uh for competition recycling camera app right a camera that can identify the materials that uh uh this is actually very good project that we should uh implement in every school so that we can teach the students how to identify whether an object is recyclable or not. Yeah something like that. Okay so this is could be a education app.

[01:38:51]Okay. So this is an example that you can take inspiration uh you can refer you can refer right is that all yeah that is all that is all for the ideas that you can refer. Next uh we will uh that is the end for our workshop. Now uh we open for Q&A but before that I will share a few uh common or common questions or common troubleshootings questions that you may face when you use the Vtop. For example, some of the students may face this issue. Uh but now it's uh it's not longer exist. I'm not sure but I'll just show it here. Uh sometime right they will ask you to log in. Uh they will ask you to log in before you use their machine learning environment. Lock in means that you need to sign up for an account. You need to sign up for an account. Uh but don't worry it's free.

[01:39:44]It's free. It's totally free.

[01:39:47]just that sometime uh my students will get this notification they have to sign up before they use uh it's very weird uh this happens to some of the students but not all so I just put it here just in case you face this issue you just proceed with the account sign out and then uh if you are using desktop version you may face this issue they will say that the uh the directory is is failed to load or whatever right because the folder is locked You have to change your workspace path to your documents. It's actually here. If you're using desktop, right, is over here under this setting.

[01:40:27]There's a option here that you can choose to uh to to change your directory. The directory means where you save your machine learning model.

[01:40:37]Sometimes, right, the default folder don't uh you don't have access to the default folder. What you do? You go to setting and then you change the directory. You can change to desktop.

[01:40:50]You can change to download folder.

[01:40:51]Whatever folder that you can access.

[01:40:53]Sometime the default folder will link to your open drive will link to your Google drive. Uh if it links to open uh uh not open drive, one drive sorry one drive and Google drive. If they link to your drive sometime you will face the issues that you cannot access. So what you do if you this only applies to desktop version. What you do you go to setting and change the directory. Okay this is number second.

[01:41:24]And then um for the webcam if your webcam shows a black screen uh it's probably you don't you didn't you forget to give access. So basically at the uh browser here if you're using the test uh the browser version you have to allow access from here. Uh make sure you allow access for the camera. Okay. Lastly why is hand pose better than image classifier for sign language? Uh so this is a question that asked by a few students right because the hand post classifier can can detect the 21 joints.

[01:42:05]If you if you try again right you go to open your uh hand and see it can track 21 joints of your hand and then it can find the pattern more accurately other than the uh compared to the image classifier.

[01:42:21]Okay.

[01:42:23]Next. Um yeah that's all for today. Yeah and we can um now open for the Q&A session. I mean you can uh keep asking questions uh in in in the YouTube channel uh I'm sorry YouTube or yeah and then we will try to answer uh your questions. And then uh for this one, this QR code and the link, this one is a a registration form for you to complete if you wish to receive uh our Chumbaka workshop completion certificate.

[01:42:58]Uh just a gender reminder, this certificate doesn't uh uh recognize as for your PHSK score. It's just a technical certificate to show that oh you already completed this machine learning workshop or you already learned how to use the machine learning tools okay it's not for P ASK so for PJsk right uh later Dr. P will share more about that. Okay.

[01:43:27]Um yeah and then I have my content leave at the last slide. If you have any issue you want to ask about technical about the machine learning or about the programming you can feel free to text me or email me. All right cool I think that is all for today. Um yeah thank you everyone for joining and now I will pass it back to um Dr. Pang. Am I right?

[01:43:53]Okay, thank you Mr. Lee. Okay, such a wonderful workshop. I hope everyone can get u knowledge on this machine learning and join the competition which will open today until 9 August. Okay, let me share some uh competition guidelines with you.

[01:44:25]Okay. So this is the AI data lab challenge the preliminary round.

[01:44:32]Okay. So you are encouraged to use the picto blocks. Okay. Or Google teachable machine or any suitable platform. Okay.

[01:44:40]That can perform the machine learning.

[01:44:43]Okay. So start from today you can start start to collect the data. Okay. And we will close this submission on 9th August and we will choose the champion from each state to the final round in September.

[01:45:00]Okay. This competition is open to primary school, secondary school and also preun university.

[01:45:08]Okay. So uh to uh to enable you to perform the uh mission in this challenge, you need to attend the workshop which is build your own AI model with PTOB blocks today and also the second workshop which is the data science for innovation. Okay, which is scheduled on 1st July 2026.

[01:45:33]Okay.

[01:45:35]So this competition you are challenged to develop impactful data analytic storyboards, interactive dashboards and AI power solution to address the real world issue align with the 17 sustainable development goals. So I believe that every issue okay must be related to one of these 17 SDG. Okay. So just choose one or more SDG. Okay.

[01:46:05]Relate the issue with the sustainable development goals and submit the video to us. Okay. The pitching video is 3 to 5 minutes. So these are the faces. Okay.

[01:46:17]First you need to collect the data.

[01:46:19]Okay. You can collect your own data or you can get the sources. Okay. Online from for example World Bank open data.

[01:46:27]Okay. or ASA website data.

[01:46:31]Okay. Phase two is data analytic and dashboard creation. So for the phase one, phase two and phase three, okay, we will cover it in the following workshop in 1 July.

[01:46:45]So today, okay, we are in phase four of the machine learning modeling and the final stage is the video presentation and solution pitching.

[01:46:56]Okay. So this are the judging criteria.

[01:47:00]Okay. I will share this document in our uh link tree. Okay. Any information you can get it in our link tree. Okay. So you need to submit a one page storyboard. Okay. Or dashboard and a three to five minute pitching video.

[01:47:20]Okay. So before we end this session, okay, please allow me to share a wonderful platform which is developed by google.org.

[01:47:35]Okay. So from this AI class website, okay, you can learn more about AI. There are total 15 modules for youth, educators and parents. If teacher would like to join us as master trainer, master trainer, okay, please uh message me or email me. I will send an invitation email to you as a master trainer. Okay, if you are master trainer, you then you can monitor your students in your class. Okay, you have your own uh referral code. Okay.

[01:48:13]Yeah. So that's all for today's sharing.

[01:48:17]Okay. So we hope that you can join us in the next workshop on 1st July.

[01:48:24]Okay.

[01:48:30]Okay. So for the certificate uh we will assign the certificate based on the Google form field uh that we post in the chat box. Okay.

[01:48:42]in the YouTube chat box.

[01:48:49]>> Okay.

[01:49:18]Okay.

[01:49:32]Okay. Thank you, Mr. Lee. Okay. See you.

[01:49:37]Bye.

[01:49:38]>> Thank you everyone. Bye. Bye.