AI in Astronomy: A Dublin High School Student's Journey
Many high schoolers who are interested in AI or coding may wonder how they can get better at AI or start learning about AI. Knowing how to use AI can be a valuable skill for many fields, simply because it can be applicable to so many different situations. One such application is AI in astronomy, which I'm going to use as an example as I guide you through how to learn and use AI.
Learning a Programming Language
The first step to understanding AI is to educate yourself in a programming language. A programming language is essentially the language of the AI; it's how you build and edit your model, look at data, and automate the machine learning process. Understanding even just the basics of AI will be helpful in the long run, and having a good foundation of knowledge about how to use a programming language will be really useful. An easy programming language to start off with is Python. There are various guides and programs online on how to learn languages.
Learning about AI Models and Data Sets
There are many different types of models in AI that all do many different things; some are designed to group things together or draw boundaries, others are created to give the computer ""vision"", and there are some that are used to translate languages. A strong understanding of what a problem is and how you want to solve it will help you choose the right model for the job.
Data is also extremely important to the machine learning process, and also requires a strong understanding of what needs to be done. Data is used in order to train the model; some data is fed into the model that it can learn and recognize patterns from. There are many different ways of obtaining data, either by yourself or downloading it from datasets online (which is recommended and much easier than gathering your own data).
Preparing and Improving your Data Set
Once you have your data, it needs to be prepared well in order to properly be used to train your model. It's highly recommended to keep your data short and simple, getting rid of any excess data that won't help towards your problem. For example, if I wanted to use AI to automatically classify stars based on the amount of light they give off, I wouldn't need to use the tilt of the star in my data since it's completely unrelated to the problem. This also helps make the model more accurate, as it might use tilt as a parameter to classify stars based on the light they give off even though you want the classification to be based purely on light levels.
Another way to improve data is with a balanced dataset. You want to have a balanced dataset because if there are too few correct examples then the model can classify everything as incorrect and still be accurate. Say you have the model which is supposed to classify stars based on their light, but your training data only has a few examples of bright stars and a lot of examples of dim stars. Your model could learn to just classify all of the data as dim stars and still be accurate—even the bright stars. This is because the percentage of incorrect classifications is so small since only a small amount of bright stars are being classified incorrectly and a large amount of dim stars are being classified correctly. To combat an unbalanced dataset one can generate fake data for the model to train on, and there are actually various different methods of doing so. One of these involves generating data points similar to the type of data of which more is needed. Other methods of improving datasets can involve processes such as data normalization, and much more. Learning about how to prep datasets is extremely important for any AI model and application, and you can improve your skills at processing data through practice with working in AI.
Finding a Problem
In astronomy, one example of people using AI to their advantage is with gravitational lens. Astronomers use gravitational lens, the bending of light caused by massive objects, to locate these bodies. However, AI significantly speeds up this process as it can work at a much faster rate than normal humans. In this case, computer vision is used—AI is used to ""look"" at images and process them to determine whether or not there is a gravitational lens present. These astronomers wouldn't use Natural Language Processing (NLP)—AI used to understand and translate languages or text—to accept image data.
Before jumping into AI, you should focus on a problem you believe that AI can help solve. Look at all of the different types of models available and what they can do, then think about how that can be applied to what you are interested in—like using AI in astronomy. Sometimes, certain subjects will need different AI models in order to solve the problem at hand, but that's the great thing about AI. There are so many different types of AI which means it can be applied to a wide variety of different fields.
Practicing AI and Using Datasets
Once you have laid all of the groundwork for learning AI—knowing how to use a programming language, looking at all the different types of AI, and knowing what problem you want to solve with which AI model—you can begin working with AI. Much of creating a well-functioning and accurate AI model revolves around trial and error: you will likely have to change the parameters of your model multiple times in order to find the best settings. There are also many different types of ways you can edit your model found online; you can add different components to your model and play around with it. Ultimately, the more you work with AI, the more comfortable you will become at setting it up, using it, and applying it.
Trying to learn AI may seem intimidating for some, but all you really need to start is experience in a programming language, a general understanding of the types of AI models, and knowing what issue you want to solve or optimize. Throughout the process, you will learn more about how to create AI models as you explore different optimization methods online, like how to improve datasets or the models themselves. I hope you are now more inspired to pursue AI based projects, because AI is becoming increasingly more prevalent throughout our daily lives and careers.
Website Sources:
https://bigdata-madesimple.com/want-get-started-artificial-intelligence-7-easy-steps/
https://www.theverge.com/2017/11/15/16654352/ai-astronomy-space-exploration-data
