I think the best presentation for week 3 was the one by Camen on using topology in image analysis. It was interesting to see how detecting number of holes and the general boundary of the digit could be used for recognizing the digit. He also evaluated the approach on a realistic dataset in the form of MNIST. It might be interesting to see how such an approach based on topology compares with common approaches/classifiers used in machine learning.
April 29, 2018
Week 3 Presentations
3 Comments
aliman
April 29, 2018 — 20:53
I also really enjoyed Camen’s presentation. The pictures are a nice touch and his explanations were clear and informative.
Philipp
April 29, 2018 — 18:06
I also think Camen’s work was pretty cool, however I think this approach no doubt quickly runs into problems, 9,6 are literally a rotation, 2,5 also can be misidentified with no respect to rotation and and you could misidentify 1,7 easily, for example, but overall it looks like Camen worked hard on research, and gave a nice presentation, it is interesting for people to see ways to apply TDA to a range of tasks.
Also, I was wondering what Camen’s presentation reminded me of and found it: work similar to Camen’s research is mentioned in this 1970 Russian educational movie on topology: https://youtu.be/R5B_BODGqho?t=13m16s (There’s visals so you don’t need to understand the language).
hduan2
April 29, 2018 — 17:41
Camen’s work is kind of cool. He tried a lot of different methods to improve his results. Based on this work, he might use topological properties to distinguish images more accurately. His slides are full of images which made us pay more attention and easily keep up.