Yuka Ikarashi

I will be defending on April 24th (co-located with Amanda!). Please lmk if you're interested in joining, and we hope to see you there!

mass migration of proofs to appendix

@cHHillee Oh no I learned about the movie after finishing the book. I bought it a few years ago but it was rotting in my shelf

@c20 In preparation for the movie?

Just finished reading Project Hail Mary. Good good good.

Wow thanks for the shout out!

Dave Patterson’s op-ed is a great overview on how taxpayer-funded research built the backbone of modern tech: RISC, RAID, cloud, ... the list goes on. Encourage your representatives to fully fund the NSF and CISE! thehill.com/opinion/techno…

@Roy_MK8DX 見ましたこれまじでアツいです

.@c20 👀

I'm giving a talk at GPU mode tomorrow. Feel free to join the livestream: youtube.com/live/J58AdFTHp…

Can there be an invasion of Iran? Hardly. Two maps explain why, and also why Iran is the way it is today, whether its regime will fall, what other superpowers will do, and in general why Iran is the way it is today 1. Iran is a mountain & desert fortress

@disconcision Welp almost everyone will graduate by the year 2028 🤷‍♀️

@c20 untrue, some of us arent graduating😭

Story of all the 5th year PhD students

Please retweet the post and help announce the workshop!

Here is the workshop schedule. We have exciting speakers presenting their papers published at ASPLOS, OOPSLA, ICFP, PLDI, and so on. plr.csail.mit.edu

We are hosting the MIT Programming Languages Review on April 25th in person here at MIT! The PLR is a student-run workshop that aims to highlight the best papers from the past year that we believe will have a significant impact on shaping the future direction of PL research.

@GPU_MODE I cannot DM you for some reason (maybe we have to mutually follow?) I followed you so can you try DMing me? If that doesn't work, my email is "yuka at csail.mit.edu"

@c20 Please DM me so we can coordinate a date!

Exo 2 news is out!

@GPU_MODE We're actively working on the Exo GPU backend and I won't be able to talk about GPUs, so the talk will be around vector machines and (non-gpu) tensor accelerators. I'll be interested in giving a talk if that's okay with you.

@c20 Would you be interested in giving a talk on GPU MODE? @GPUMODE/videos" target="_blank" rel="nofollow noopener">youtube.com/@GPUMODE/videos

@tikhonjelvis Thanks for your comments/interests!

@c20 sweet, looking forward to seeing that in the future :)

@c20 up on HN again: news.ycombinator.com/item?id=433626… it's catching people's imaginations :) the question in the thread about why normal compilers can't automatically do the optimizations somebody would express in Exo seems pretty interesting

@tikhonjelvis abstraction in the scheduling library. And yes I think this approach is applicable beyond dense linear algebra, and we're working on it!

@tikhonjelvis Exo is "low-level" compared to other user schedulable languages like Halide, and that's its core strength! Exo offers fine-grained control for peak performance, and provide automation via the implementation of libraries. We showed in the paper that Exo could recover Halide-level

@tikhonjelvis also is not general (by definition) so it's not a silver bullet

@tikhonjelvis For dense linear algebra-like kernels, optimizations are usually structural and are agnostic to input data and runtime information. Knowing parameters like tensor shapes certainly makes the optimization easier and more efficient, but specialization increases the binary size and