What is deep learning and what do researchers who do research in deep learning actually do ?
In: Engineering
Deep learning is a type of machine learning that uses specific algorithms called “neural networks”. Most AI models these days use some kind of deep learning.
At a fundamental level, it isn’t terribly different from other machine learning methods. The goal is to generate a computer model that lets you tell it some information about a lot of things (the “training set”). For example, you might input a bunch of pictures and tell the computer which ones are real people’s faces and which ones are AI generated.
Then the model uses some fancy algorithms and math to find patterns in those things that you have told it. This is where the real “magic” happens. Deep learning is a specific way that the computer learns patterns using multiple steps that all happen sequentially and automatically (called “hidden layers” in this context). Compared to a non-deep learning model, a deep learning model is almost always more complex and therefore requires more computing resources to make and may even require more training data to be particularly worthwhile, but the result usually makes more accurate predictions because it can “learn” more subtle or complex patterns. Another trade off of deep learning is that, without some special additions (like “attention layers”), it can be pretty much impossible to actually figure out what specific patterns the model learned. You just know it “learned” *something*, and you then “validate” or “test” it on new data (that you still already know the correct answer to) to see how useful or accurate that “something” is. Repeat until you’re satisfied.
Finally, once you’ve “trained” your model, you then give it a new picture and ask it “Is this real or AI?” The deep learning model then breaks down this image into parts and looks for the patterns it learned from the training data to give you an answer, which might be a “Yes/No” or something more granular like “75% likely to be AI generated”.
TL;DR a specific type of machine learning/AI model that lets you detect more complex patterns to make better predictions than a simpler model.
Latest Answers