A strong finish with this week of talks. Having no background in topological data analysis, I learned quite a lot this week about the scope of possible applications of TDA, especially when considering areas like the arts. I think I was most interested to see the talks by Bryan and Joe on how TDA could possibly be used to distinguish authors and composers. Bryan’s talk raised a lot of questions for me about whether there is a link between math and aesthetics, some sort of underlying structure to music that can be picked out as musics vs noise. My favourite for the week is going to have to go to Joe’s though, because his talk introduced me to a couple of new tools that I’d like to have a play around with, specifically the Word2vec. Neural networks are fascinating in their own right, but I’m very much going to spend some time playing with the vector space of words and look into how they defined it.
April 29, 2018
FP Week 3
7 Comments
bjiang8
April 30, 2018 — 00:52
Joe’s work is extremely cool. How to get the meaning of the word is always a big problem for AI. If we can describe it with manifold or topology, that would be a awesome progress.
gljohns3
April 29, 2018 — 22:27
I think that idea is beautiful in physics; the representations of the rotations groups manifest themselves as particles. Tight! But the romantic side of me does not want something so human as music to be expressible in a closed form. It should be like Toaism, the harder you look for the Tao, the further you are from it. Then again, my rational side is pretty sure that thought is just emotion trying to justify itself. By the way, I’m ranting to avoid studying for the stat mec qualifier. I think Roger Penrose has said something along the lines of physics is not all in math is not all in consciousness is not all in physics, and vice versa. Kind neat thought.
gljohns3
April 29, 2018 — 22:28
This should’ve been a reply to Nick’s comment…
gljohns3
April 29, 2018 — 21:25
I really enjoyed the all the TDA too. I was surprised that word2vec actually worked so well under addition. As the music and math, I’m not sure if I’d be excited or disappointed if TDA worked really well for bringing out underlying features. Wouldn’t that possibly take some of the humanity out of it. Or is that just maybe some kind of purist notion?
I guess what I’m saying is that if someone played something beautiful and then some dude was like “Oh his phrases simply did so and so,” I probably wouldn’t like that person. You can’t quantize my feelings man!
Nick
April 29, 2018 — 22:15
I don’t think of it necessarily as “quantizing feelings” so much as it being an interesting and possibly beautiful relationship between our minds, perceptions and math. It’s more of a romantic notion in my head of how math is so deeply ingrained in us that we’re programmed to see and appreciate it everywhere.
Bryan
April 29, 2018 — 17:13
Joe’s talk was really interesting. I had thought his presentation was the best. The programs he brought up are interesting to use
jhmuelle
April 29, 2018 — 17:06
By the way I wrote my code in python and used a library called gensim for word2vec. If you want to see my code let me know and I’ll send it your way. Gensim has a number of other word embeddings other than word2vec, they’re not as fancy but they’re still pretty cool.