Course Work Threads

When I first got familiar with AI, I was told there were basically two paths towards it -- from bottom to the top and from top to the bottom. Ideally I wish to go both ways, and I think I'm sort of a bit close now.

CS 450, ECE 408, and ECE 374 are the courses that I took in 24 Fall. To me, all these courses are about how to compromise when solving problems with machines as well as their limits, which is pathetic but indeed useful. One can also probably tell that I will turn to the magical math rather than the magical nature for a while, which is wretchedly true.

Fortunately we do have some insights from the gap (which can be wide) between nature and math. PSYC 489 was offered in the psychology department, which introduced me to a more cognitively related perspective about machine learning and neural network, as well as interesting new concepts such as "analog" and "associative neural network". The former one is kind of like "representation" that we are talking about in ML domain nowadays, and the latter one makes me curious about why it has been in such silence. Overall, this course was about offering hands-on experiences of building neural networks from scratch and observing them as "scientific" approaches rather than engineering ones according to prof. Hummel.

These are the threads that I am currently working on, for which I have been longing since high school and are supposed to be stuck with me for a considerably long period .

This thread is exactly what our major is requiring us for, that is, to understand modern computers as basic systems down to the physical level. The best idea I get from this thread is likely to be building abstract levels, or in other words, different layers of interfaces, to form a really complicated control system. This is somehow similar to the role of feature extraction in machine learning, I suppose. By the way, my intuition tells me that building discrete logic systems is somehow related to the manner of neural networks, which I now know is not only an intuition.

These can be a bit too far. Overall, the coolest thing about this thread is still that it enables us to turn objective physic laws into highly abstract systems that function magically, which is exactly what "engineering" means, I guess. (Though the painful experiences of debugging are probably telling me that I wouldn't be a happy engineer.)

This is the most surprising, impacting, while curative line for me. My mind kept turning into completely different ones along this thread. Specifically, up through this thread, I gradually abandoned my belief for the existence of findable natural fundamental truth. Instead, I started to inspect everything as illusion in our cognition (which helps me a lot in making good scores out of exams since I don't pester with why math exists when reviewing for finals anymore).

Hume is still my favorite philosopher. Bacon, Kant, and Wittgenstein are also awesome.

This is the thread from which the previous thread saved me. All these courses were initally assigned as engineering supplements only, but everything changed as they were given by prof. Honold, who was an excellent pure mathematician from Germany. Frankly speaking, I was rather delighted for the most of the time, though it became indeed tremendously challenging to finish the weekly assignments (yet I got the best explanation for the determinant I've ever seen). Almost all of them were freaking analysis problems in pure math manners of various domains, and my pride and belief made me insist on working by myself, paying consecutive 24 hours each week. I was believing myself chasing the most canonical way to describe the world even more than when I was learning physics. (Though later glimpse on others' work made me realize I was not that sensitive to skills, and it would be a pipe dream to avoid skills when doing math.)

And yeah, homework and midterms did turn out great, but I was missing two vital facts then. First was that the final assessment, which took the largest weight, was still supposed to be on application point of view. Secondly and more importantly, it was not enough to solve problems with conclusions in a limited time through just listening to the awfully rigorous proof constructions on lectures (which filled up the lecture time, with rare people in presence). It took me long enough to realize these, after my suffering from self-doubts for several months while crying over my final scores. Overall, it's still a great tour. I believe it reasonable that challenges which prof. Honold has left me will keep in company with me for at least the following decade.

This is a path which I had been looking forward to since my 12. However, miserably, when I got in high school, I found myself limited in learning advanced physics topics. Hence, in my second year in high school, I angrily turned to explore the reasons why I was limited, and then got familiar with some concepts in cognitive science, neuroscience, and deep learning. That's just how I started my adventure in artificial intelligence. Since then, I set aside my wish to research on physics myself and turned to expect machine to do that for me. (Actually I'm expecting machine to do a lot more than physics, since I have been daring to guess that doing existing science can be a bit far from knowing everything.)

What's fortunate was that understanding those physics topics didn't seem that challenging as soon as I was in college, which proved my hypothesis that the reason was no more than my limited working memory due to being just a little kid.