
Alvin Cheung
@alvinkcheung
Followers
2K
Following
1K
Media
53
Statuses
875
Data management and programming systems; Associate professor @Berkeley_EECS, member of @ucbrise, SLICE Lab, and @UCBIDS
Joined February 2015
And we just won the outstanding paper award at the MLArchSys workshop this year! Congrats Charles, @sahilb17 and @Sophia_Shao_! 🎉🎉
In this work, we show how to use off-the-shelf LLMs to generate code for accelerators. This is interesting as accelerators are often "low resources," i.e., there isn't much code written using such accelerators to train custom models. Check out our paper for details!.
2
2
37
In this work, we show how to use off-the-shelf LLMs to generate code for accelerators. This is interesting as accelerators are often "low resources," i.e., there isn't much code written using such accelerators to train custom models. Check out our paper for details!.
💥𝗖𝗮𝗻 𝗟𝗟𝗠𝘀 𝗼𝗽𝘁𝗶𝗺𝗶𝘇𝗲 𝗹𝗼𝘄-𝗹𝗲𝘃𝗲𝗹 𝗮𝗰𝗰𝗲𝗹𝗲𝗿𝗮𝘁𝗼𝗿 𝗰𝗼𝗱𝗲?.Yes! (With some help.) In our paper Autocomp: LLM-Driven Code Optimization for Tensor Accelerators, we show how LLMs can optimize super low-resource accelerator code. 🧵 1/4.
1
3
8
RT @adityagp: We've been working on extracting information in templatized PDFs for the last couple of years, leveraging the best of LLMs an….
0
88
0
RT @ShadajL: New Blog Post! I argue that we’re *still* lacking a programming model that’s native to distributed sys….
0
75
0
happening now at NeurIPS!.
We’re excited to present our work on Verified Code Transpilation using LLMs at #NeurIPS2024 today! 🤖🔍 Come visit our poster in the East Exhibit Hall A-C #2904 from 11AM-2PM.
0
0
6
Many streaming systems out there lack formal semantics. Check out Flo, our semantics framework for reasoning about streaming systems. We've used it to reason about DBSP, Flink, and our very own .
Thrilled to share that our paper “Flo: A Semantic Foundation for Progressive Stream Processing” (with @mbpmilano, @alvinkcheung, and @joe_hellerstein) will appear at POPL 2025! Check out the preprint at and read on for more!
1
1
11
RT @Berkeley_EECS: EECS is hiring! Open faculty positions are now available. We welcome applicants from all areas focusing on originality a….
0
13
0
Humbly honored to receive the Dahl-Nygaard Junior Prize @ECOOPconf this year. Thank you AITO, and all the great students, mentors, and colleagues I have the fortune to learn from. Excited that our work is recognized by both the PL and DB communities!
7
3
38
RT @ShriramKMurthi: We're still getting the OOPSLA 2025 site up. But for those who're wondering: OOPSLA 2025 will have two rounds of submis….
0
36
0
Also, ICYMI @ShadajL presented his work on a Rust-based streaming dataflow language for distributed workflows. Check out and the hydro project:
0
2
10
At #pldi today @gabomatute will be presenting his work on syntactical search on code. We support both wildcards and even partially parsable queries! Check it out:
0
0
4
RT @JAldrichPL: We're working to make liquid types more usable! If you've worked with LiquidHaskell or are interested in doing so, please….
0
2
0
RT @pinartozun: ICDE 2024 offers attendance support to students (graduate or undergraduate), postdoctoral fellows/researchers, or junior fa….
0
5
0
ICYMI like last year there are **two** deadlines for ECOOP in 2024. It's an awesome conference. Submit your best work to @ECOOPconf !.
I hope you’ll find our new reviewing model attractive as an author, & I encourage you to submit your work so it can be part of the great program we’ll be publishing in ECOOP 2024! Again, ECOOP papers are due on January 17 (round 1) and April 17 (round 2).
1
1
6
RT @TeresaWatanabe: I spoke with community college grads who stand to earn more than some peers at elite @Stanford @UCLA @UCBerkeley. Major….
0
15
0
Incredibly humbled to receive the #VLDB2023 early career research contributions award this year. This is only possible thanks to all my students, collaborators, mentors, and teachers from @uwcse @Berkeley_EECS and beyond. Thank you for your support and inspiration over the years!.
8
6
114
Can functional dependencies only be discovered using ML? Turns out many of them are hidden in app code! Lily will be presenting Coco this afternoon at #VLDB2023, where we analyze code to extract FDs and use a "search and verify" algo to rewrite queries.
0
1
20