Deep Learning is a type of Artificial Intelligence that has gained popularity in recent years because of how well it can handle real-world complex data like images, sounds, videos and language, just as examples. Because of this, deep learning is a great skill for high school students to learn and use in everything from science fair projects to nonprofits to entrepreneurship. In this blog, we describe what deep learning is and answer common questions you may have about how to learn deep learning and use it.
What is Deep Learning and what are the types of Deep Learning?
Deep learning is a type of Machine Learning which in turn is a type of Artificial Intelligence. Artificial Intelligence (AI) is a broad term that refers to any technology that helps computers do tasks that come easily to the human brain (such as learning, making forecasts, creating strategies, etc.) Machine learning is a subset of AI that focuses on ways that computer programs can learn from data and experiments. Deep Learning, in turn, is a subset of Machine Learning using a technique called Neural Networks which superficially resemble the neurons in the human brain. Deep Learning has been shown to be very effective at complex problems involving lots of data or rich data like images, video, sound, text, etc.
There are many types of Deep Learning, and they differ based on the structure of the neural networks used (Neural Architecture) and how these networks learn. for example, Convolutional Neural Networks (CNNs) are very good at image processing, Long/Short Term Memory Networks (LSTMs) are very good for time series data (such as sequences of temperature from a sensor), and Transformers with Attention (a good example is Bidirectional Encoder Representations or BERTs) is very good at languages and text.
Why is Deep Learning so popular?
Deep learning is great for complex problems that use lots of data. It is also very good at problems that use data that is natural for human interaction (such as pictures, sounds, etc). For this reason, it has been used in everything from detecting diseases from MRIs and X-Rays to creating artwork for kids.
Is Deep Learning difficult?
Not really. The math that drives deep learning is based on advanced calculus and is learned by most students at the undergraduate level in college. However, you do not need to know this math to build and use Deep Learning very effectively. In fact, most people who use Deep Learning in the real world do not know this math :-)!. In our experience, students even in elementary school can build deep learning AIs very easily - to do things like recognize animals, detect whether they are wearing a mask, etc. Students in High School can do projects using Deep Learning for HealthCare, Climate Change, etc. We have had many high school students do successful deep learning projects and win science fairs and other competitions. You can see some examples here.
How do you master Deep Learning?
Start by learning the basic concepts and putting them into practice by building Deep Learning AIs. Then as you get more experience, build bigger projects with real-world data. You can learn how Deep Neural Networks work and how to tune and improve them. Next, you can tweak not just the tuning but also the network design. A good example of what high school students can do is shown in Abhinav’s project on using Transformers and BERT for detecting disasters from tweets.
Which Deep Learning Model is Best?
There is no one answer to this question. Different types of deep learning have shown benefits for different types of problems. Even within a type - such as CNN - there are many variants and designs, and new ones are discovered all the time. If you are solving a particular problem, the best strategy is usually to try several state-of-the-art networks and then tune them.
Is Deep Learning Overhyped?
Some people think so, but probably not. Deep Learning is exciting because many companies have found that it can do amazing things with complex data. So far we are not seeing an end to the capabilities of deep learning - every day there are more examples of industries that have been able to use Deep Learning effectively.
Is Deep Learning the Future? What is Next After Deep Learning?
Hard to say. While deep learning has shown amazing results and seems to continue to do so, there are other types of learning, like Reinforcement Learning, which have also made great strides in the last few years, and new types of AI are constantly being developed. Reinforcement learning is used heavily in self-driving cars, and in any problem where an AI has to develop a strategy to solve a problem. Note - this does not mean one type of learning will replace the other - Deep Learning and Reinforcement Learning have also been combined very effectively. It just means that the future will likely not be just deep learning - it is important to learn other types of AI as well.
Kommentare