Introduction to Artificial Intelligence
I think current progress in AI is the most significant innovation of our times. To a large extent, it suggests that reasoning, as is experienced by humans, is possible outside machines. This effort has roots in the 60s, but has only recently become successful at an anthropologically significant scale.
We will study paradigms of Artificial Intelligence, and importantly, what makes neural networks click. There’s so many topics I am interested in here. I will provide some background and use this space as a scratch pad to understand what is going on in some current developments.
There are two subsections to this topic:
‘Understanding AI’: general introduction to the field. Potentially for beginners with a mathematical interest, who have already looked at other great online resources.
‘Topics in AI’: technical discussions on specific themes that I find intereting and want to learn about.
Here and there I will talk in terms of categories. This is because different mathematical templates are often fit on neural networks: conditional probability distributions, dynamical systems, vector spaces, and so on. Category theory is a mathematical language that allows us to find a common underlying structure.
In particular, it is sufficiently abstract to provide a template for us to talk about objects and morphisms, then replace them appropriately with neural networks and loss functions. I think all of this can be intuitively understood by replacing the notion of morphism with that of function, but if you want to read more about category theory, I am sharing resources in the main intro page for the topic.