Techfitlab Breaks Down Tesla Autopilot, AI, ML, and DL Complexities

Key Takeaways

this is what's ai and i share artificial intelligence news every week if you are new to the channel and want to stay up to date please consider subscribing to not miss any further news today i have a guest speaker for you his name is gaurav he has been working in the data science field for nine years now and is also a fellow youtuber here's how he sees artificial intelligence he will also be explaining the difference between artificial intelligence machine learning and deep learning then he explains what machine learning is and the different ways an algorithm learns such as supervised learning unsupervised learning semi-supervised learning and reinforcement learning he then explains deep learning a bit deeper and how it works using some examples finally he shows a real world example of ai which is tesla's autonomous driving chips in this case and uses this to explain what a neural network is i will now let him introduce himself in his channel enjoy guys and i hope you learned something new from giraffe today thanks luis for having me on your channel uh this is garf just to give you a bit of introduction about myself so i've been working in this data science field for past nine years since when i started a channel as lewis mentioned which is techfit so within my channel it is divided into two parts which is tech and fitness again within tech i do cover different data science topics and also looking at different uh programming or markup languages and again give you some introduction about those and also be looking at different uh projects as well and kind of demonstrating those projects with yourself and also kind of like sharing my experience in this field and also trying to learn from my peers and the other half of the channel is fitness again i'm a qualified gym instructor and personal trainer hence why my channel kind of like shows those two areas such as tech and fitness so the topic we are going to cover today is artificial intelligence again within artificial intelligence when you look at on youtube or you're trying to search on google there are different terms that comes up as well which is such as machine learning or deep learning so what i want to kind of clear within this is kind of what exactly is artificial intelligence and where does deep learning and machine learning sit and maybe demonstrate with one of the examples to put it in really simple terms like so audition intelligence is a simulation of human intelligence especially done by machines so that's what exactly is artificial intelligence now you would wonder where does actually machine learning or deep learning sit so to kind of elaborate it more artificial intelligence is your umbrella term where you have your architectural intelligence and under that you have your machine learning so machine learning is basically some set of article intelligence and deep learning is the subset of machine learning so if you kind of like talk about it on the reverse order so you have your deep learning it's just subset of machine learning and then you have your machine learning which is subset of article intelligence all the machine learning is counted as ai but not really all the ais candidates machine learning because again i could have really a really good rule based system which could be uh just collect trying to answer the questions that a customer is asking and would be like just working as an ai model but it's not really there's no machine learning applied to that it's just a rule based engine so let's talk more about machine learning machine learning is very you're giving the ability to the machines to actually learn and improve from the experience without being explicitly programmed so within machine learning you have different forms of studies such as supervised learning so supervised learning is where you have label data so we can take an example of uh building a churn model so if you look at the historic patterns of the customers you know exactly what customer left what customer didn't left so you do have that label data saying that this person left or not so this is kind of like your supervised learning where you try to build different types of models then you can use different types of models such as logistic regression or naives or support vector machines to predict what person will likely to be leaving in future and you can have those probability scores against that so that is your supervised type of learning second is unsupervised learning with an unsupervised learning you you don't really have any label data in your database so you're trying to see what patterns comes out of the data one so one time example would be clustering technique where you're trying to build the clusters which comes out of the database itself which is a unstructured form of the data and the models you could use maybe k means or mean shift etc third one is semi-server-wise so the name kind of like suggested all where your model is first trying to build a structure in their data set and then predicting against that structure it's kind of called that semi-supervised learning so fourth one is reinforcement learning so reinforcement learning is quite similar to human brain where you're trying to mimic that human brain activity such as if your human brain is learning something it does not really need constant supervision in terms of answering something so to put in a really basic examples if your brain looks at a picture of a cat it exactly knows if it looks at a different picture of cat who would know that this is the cat so the same way would be for machines as well to actually look at a picture of cat and learn from it and so it doesn't really need a constant supervision what exactly that picture is so which is kind of like so say for an example within your level data you're providing the label data as a machine to learn from it and it doesn't really need any constant supervision afterwards it can actually predict if it sees a new picture of a cat it can actually predict if it's a cap or not so this kind of like learning is your reinforcement learning so deep learning is basically an ai function or you can call it as a class of machine learning which basically mimics human brain so deep learning is also classed as deep artificial neural networks or deep reinforcement learning some of the examples of deep learning would be image recognition sound recognition or looking at recommender systems or natural language processing so the idea behind deep learning here would be that machines are capable to make decisions without any human supervision it's only done based on the experience of the data provided and even if the data is unstructured or unlabeled they can actually make decisions so now we talked about deep learning machine learning and ai again you can take analogy which is again as russian doll where you have your deep learning and then you have another encapsulated one which is your machine learning and which is your ai on top let's look at one of the real life example of ai so we'll talk more in detail about tesla autopilot ai chip so as you might be aware of tesla autopilot ai chip which is again tesla is trying to reach a level five autonomous driving very there is no human supervision is required anymore where you just sit in the car you click the button from point a to point b and the car brings you to that point so within this ai chip the idea behind this is to basically mimic a human brain so the technique which is used in the background is neural net so to give you a really basic example of neural nets again the idea is that you have your input layer where your all the input comes in then you have your hidden layer which is again where you make your decision in terms of the weight edges given to each of the input layers and to your activation functions then you have your output layer which is the actual decision so the idea is that it is the forward propagation where your model is basically predicting what exactly value it is but based on the learning because you have your data which is again labeled data based on the correct answer so the model basically does the backward propagation which is again done based on the greater decent but again we don't we're not going to get into a lot of these terms well because i'm trying to leave it more higher level to just kind of give you the understanding of how the model is training or what exactly is neural nets so once we have a basic understanding of neural nets let's look at what exactly is the autopilot air chip is trying to do so autopilot airchip is trying to actually mimic a human brain so same way for example if my eyes are looking on the road while i'm driving the car it's actually making the decisions oh should i turn right or should i turn left or to drive on the road markings exact same way machine is trying to do the same work where through my vision what i'm trying to do is i'm trying to take that vision through my retina which actually goes back into my cortex where i'm actually making the decision within my brain to turn right or left so exact same with the airship is trying to do as well where the ai chip is the neural network's brain is there so what it's trying to do is just trying to take the input to the cameras what exactly the vision is and based on the the label data is trying to learn within that ai chip and making the decision to turn right or left at a certain point or drive on their own markets so kind of like to go in more detail exactly what ai chip is trying to do and how does it make decisions so if you put it in this way as well to collect really explain it on a really granular level so if you look at this chip so same example as on the right you can see where there's video inputs where the actual input of the data is coming in from then you have your two chips that are sitting in the middle which is actually where the brain is which is again where the neural network is sitting so where it makes predictions of turning right or left then you have your output signals which is again the decision and based on the decision you have your chip which actually makes the decision of turning the steering wheel right or left it does not just collect camera vision data it does collect a lot more different types of data such as uh gps data map data or ultrasonic sensor data p data or steering angle data so these are different types there that has also been collected within the chip to make those decisions within the neural net model this is how the chip tends to learn to how to draw the curve which is exactly mimicking the human brain so i hope you enjoyed this topic thanks luis for having me on your channel i really appreciate it and i'll pass it on to you thank you for this valuable information gareth this was really clear and well explained if you guys enjoyed it too go check out his channel it is the first link in the description box below as usual i will see you on my channel for a new video this saturday where i will be covering the eccv 2020s best paper please leave a like if you went this far in the video and since there are over 90 of you guys watching that are not subscribed yet consider subscribing to the channel to not miss any further news clearly explained if you would like to start or improve with machine learning i've linked all the best online courses in a reporter in the description thank you for watching [Music] you