Deep learning in 5 minutes | What is deep learning?

deep learning powers most of the technologies that we rely on every single day this could be machine translation that translates to websites that we're visiting it could be facial recognition that lets us log into our phones or it could even be recommendations that we rely on to show us the next best tv series or show on netflix to watch so in this video we will look into deep learning in more detail and really understand what it is this video is part of the deep learning explained series brought to you by assembly ai a company developing a state-of-the-art automatic speech to text api if you'd like to have a free api token follow the link in the description first and foremost a

description of deep learning is a group of techniques that are based on neural networks that have the capacity to learn complex patterns directly from the data so let's break this down and go at it step by step first thing neural networks

are algorithms that were created inspired by our brains you can see in this example we have layers of neurons stacked one after each other in neural networks we have three different layers one of them is called a input layer where we accept the input the other one called an output layer at the end of the network where it gives the actual prediction or the outcome of the network or the algorithm and in between we have the hidden layers all of these layers consist of neurons depending on the input or the output of the probe or the problem you're solving you're going to have different number of neurons there is always one input and one output layer but the more hidden layers that you have in between those the deeper network that you're going to have and that's where the keyword deep comes from in deep learning whereas learning comes from machine learning or the fact that this network is able to learn different complex patterns neural networks and deep learning has been heavily impacting our lives in the last 10 to 15 years but actually they've been around for longer than that the first neural network was suggested all the way back in 1943 and since then there have been waves of interest that eventually died out and you can actually think that we are currently living on a new wave of interest when it comes to deep learning and this time people think that it's not actually going to go away because first of all we have a lot of data that we can deal with and the computing power is getting better and better every day well not to mention the research that is being done into deep learning currently to make it better and faster every day so let's talk about how this all relates to machine learning and also how machine learning compares to deep learning let's start all the way from the top we know that we have computer science about everything right and computer science has different branches this could be computer security software engineering or distributed systems for example and one of these branches is called artificial intelligence in artificial intelligence the general goal is to have the computers perform tasks that are normally typical to humans in a way that is accurate and also efficient there are different approaches to artificial intelligence of course and one of these approaches is called machine learning in summary machine learning aims for the computer to learn how to do a task directly from the data well deep learning is part of machine learning it's again in itself a group of techniques that are in machine learning so when you say machine learning you are actually including deep learning inside the picture but when we want to compare

machine learning and deep learning of course what we're trying to compare is the traditional machine learning algorithms with the deep learning algorithms that have been improved in the last decade all right so let's look into how they're different traditional machine learning algorithms on one side and deep learning on the other side so the first and most prominent difference is that when you are training machine learning algorithms you need features extracted from the data manually and when you're doing deep learning training you don't have to do that so just to give you an example let's say you're trying to classify cats or dogs the most common example on the internet so what happens is when you want to um train your model to understand if a given picture is of a cat or a dog you have to extract features such as uh how many pointy ears it has what does the nose look like what is the color of the nose what is the pose that the animal is giving for example this is just a abstract example of what kind of features that you can extract whereas for deep learning you can just give the image as it is to your deep learning algorithm and it will itself understand what the pattern is and what the features are that are separating the two animals from each other in these photos

and this is very nice right because when you're training deep learning models you do not have to do future engineering at all but of course this nice little feature comes with a cost when you're training deep learning algorithms you have to have much more data than you need for traditional machine learning algorithms to be able to train your model to be accurate and also due to all the computation that needs to be done in deep learning algorithms you have to have a stronger machine which has higher processing power and it's going to take a longer time but that's kind of like a trade-off of course the stronger the machine that you have the less time it's going to take but on average it's always going to take longer to train deep learning algorithms that compare to the traditional machine learning algorithms but there is one other difference actually that is kind of hard to put on paper and that is that deep learning algorithms are actually able to capture patterns that are a little bit more abstract so you can actually perform tasks with deep learning algorithms that you would not be able to with traditional machine learning algorithms this could be for example natural language processing you might not always know what kind of features to extract or generate from your text to be able to do sentiment analysis that is analyzing if a certain text if a given text has a positive or negative connotation whereas deep learning handles this very well just by looking at the examples which were labeled another example is actually the example that i've given earlier when you're trying to distinguish between cats and dogs it might be a little bit hard to create features just by looking at the animals photos whereas with deep learning you don't have to worry about that you can just feed your data into the deep learning algorithm and then it will extract everything for you so i think this makes deep learning algorithms better when it comes to abstract tasks which would be really hard for a human to even format and there are of course other cool things that you can do with deep learning for example turning speech to text and that's exactly what assembly ai does so if you're currently interested in having a speech-to-text api integrated in any of the projects that you're working on you can give assembly ai a try you can follow the link in the description to get your own free api token to start working on it whenever you like if you like this video please give it a like and maybe even consider subscribing we will be here every week bringing you information on deep learning machine learning and tutorials on the latest technologies that are out there we will be very happy to see you with us if you have any ideas or videos that we can make definitely go ahead and write a comment and let us know that would be awesome so thanks for watching and have a nice rest of your day