Hello Codeforces,↵
↵
It's been a while since my last activity / contest participation on here! I've largely moved on from competitive programming to machine learning, but still tried to keep up with the vibrant community of sport programmers whenever possible.↵
↵
Recently I've decided to orient my focus in ways in which we can combine the representational power of neural networks (great for handling big, noisy, "real-world" data) with the incredible robustness of algorithms (provably correct, trivially generalisable across problem instances, compositionality w.r.t. subroutines and functions). One way forward is teaching neural nets the execution steps of algorithms -- i.e. replicating the intermediate outputs on classical competitive programming algorithms and tasks.↵
↵
I produced the following slide deck (gave as keynote at WWW'20 Workshop on [Deep Learning for Graphs](https://www.aminer.cn/dl4g_www2020)) which outlines the recent explosion of work in the area (including my own). There is credit to Codedorces in one of the slides -- as one of the places that really got me into algorithmic programming / computer science in general -- including a photo of the ACM-ICPC team from Cambridge I coached ([user:dd__,2020-04-22], [user:MeinKraft,2020-04-22], [user:zDule98,2020-04-22]) that won NWERC'17. :)↵
↵
Hope you'd find it interesting!↵
↵
Recording: https://www.youtube.com/watch?v=IPQ6CPoluok↵
Slides: https://petar-v.com/talks/Algo-WWW.pdf↵
↵
Cheers,↵
Petar
↵
It's been a while since my last activity / contest participation on here! I've largely moved on from competitive programming to machine learning, but still tried to keep up with the vibrant community of sport programmers whenever possible.↵
↵
Recently I've decided to orient my focus in ways in which we can combine the representational power of neural networks (great for handling big, noisy, "real-world" data) with the incredible robustness of algorithms (provably correct, trivially generalisable across problem instances, compositionality w.r.t. subroutines and functions). One way forward is teaching neural nets the execution steps of algorithms -- i.e. replicating the intermediate outputs on classical competitive programming algorithms and tasks.↵
↵
I produced the following slide deck (gave as keynote at WWW'20 Workshop on [Deep Learning for Graphs](https://www.aminer.cn/dl4g_www2020)) which outlines the recent explosion of work in the area (including my own). There is credit to Codedorces in one of the slides -- as one of the places that really got me into algorithmic programming / computer science in general -- including a photo of the ACM-ICPC team from Cambridge I coached ([user:dd__,2020-04-22], [user:MeinKraft,2020-04-22], [user:zDule98,2020-04-22]) that won NWERC'17. :)↵
↵
Hope you'd find it interesting!↵
↵
Recording: https://www.youtube.com/watch?v=IPQ6CPoluok↵
Slides: https://petar-v.com/talks/Algo-WWW.pdf↵
↵
Cheers,↵
Petar
