13 Followers
33 Following
11 Posts
Postdoctoral researcher @KIT_Karlsruhe | #SATSolving, #HPC, #SymbolicAI | ๐ŸŽธ๐ŸŽน๐Ÿช•๐ŸŽค๐Ÿ‘จโ€๐Ÿณ๐ŸŽฎ
KIT pagehttps://ae.iti.kit.edu/schreiber.php
Homepagehttps://dominikschreiber.de
Githubhttps://github.com/domschrei
Science Slam | GI-Dissertationspreis 2024

YouTube

In this year's SAT Competition, Mallob @ 1600 threads proved its worth against two other distributed systems, scoring its 5th gold in a row. In the shared-memory track, one of my submissions is (to my knowledge) the first ever parallel submission that produces and checks proof information, which drastically increases confidence in the result. See: http://doi.org/10.4230/LIPIcs.SAT.2024.25 The associated overhead was so small that it also scored some medals!

Details: https://satcompetition.github.io/2024/downloads/satcomp24slides.pdf

Trusted Scalable SAT Solving with On-The-Fly LRAT Checking

I'm very honored to have received the inaugural Fahiem Bacchus Award at #SAT'24 in Pune (India) for my dissertation entitled Scalable #SATSolving and its Application. Many thanks to the award committee, including @arminbiere, for this amazing recognition! It really encourages me to continue this research, further pushing the scalability of #AutomatedReasoning in #HPC and #Cloud systems.

Bald ist September und das bedeutet: Die Award-Season der Informatik beginnt! ๐Ÿ˜Ž

Zusammen mit @ocg und der Schweizer Informatik Gesellschaft verleihen wir den Dissertationspreis an zwei Forschende:
๐Ÿ… Menatallah El-Assady von der ETH Zรผrich
๐Ÿ… https://sueden.social/@dompasch von @KIT_Karlsruhe

Alle Infos: https://gi.de/meldung/erklaerbare-und-symbolische-ki-gi-dissertationspreis

Dominik (@[email protected])

6 Beitrรคge, 32 Folge ich, 11 Follower ยท Postdoctoral researcher @[email protected] | #SATSolving, #HPC, AI planning | ๐ŸŽธ๐ŸŽน๐Ÿช•๐ŸŽค๐Ÿ‘จโ€๐Ÿณ๐ŸŽฎ

sueden.social
Don't call it portfolio #SATsolver! In our latest journal article, we show that getting clause sharing right leads to scalable distributed #SATSolving even if all solver threads are initially identical. This motivates us to drop the prior portfolio-centric intuition for parallel SAT and to rather adopt the term "clause-sharing (SAT) solver" going forward. https://doi.org/10.1613/jair.1.15827 #AutomatedReasoning #KIT #SAT #HPC
MallobSat: Scalable SAT Solving by Clause Sharing | Journal of Artificial Intelligence Research

Today's clause-sharing portfolio #SATSolving cannot produce UNSAT proofs, which makes it unsuitable whenever trusting a result is critical. This changes! In our paper just accepted at #TACAS 2023, #AWS-affiliated researchers (D. Michaelson, M. Heule, M. Whalen, B. Kiesl-Reiter) and I have derived a scalable approach based on LRAT+CaDiCaL+Mallob producing efficiently verifiable proofs, with some interesting distributed algorithms and showing good performance on up to 800 cores. More coming soon!
Mallob in 2020 was using Lingeling(+YalSAT), and so did "Mallob-Ki" in 2022, in contrast to Mallob-Ki(ssat)Ca(dical)Li(ngeling)Glu(cose).
The data is not free from biases โ€“ for example I've been tuning Mallob on the 2020 benchmarks for a while. Still, some very nice improvement on what I'd consider to be a broad set of benchmarks.
Without running any new experiments, here's the progress of sequential, parallel, and cloud #SATSolving from 2020 to 2022 based on #SATCompetition data. I used the intersection of the 2020 benchmarks and the 2022 anniversary benchmarks (โ†’ 354 instances). Showing the 2020 winners and the 2022 anniversary winners. (higher = better; legend is sorted according to the curves' positioning)
#Introduction #AcademicMastodon
Hi there! I'm a #CompSci doctoral researcher at @KIT_Karlsruhe, Algorithm Engineering group, with a focus on distributed algorithms, #HPC, and automated reasoning. Author of Mallob (#Cpp, LGPL) โ€“ a decentralized job processing platform and the current state of the art in large-scale #SATSolving. Might post some insights or results on here from time to time.
BABIES OR LITERATURE BUT NOT BOTH: Baby shoes
https://xkcd.com/2702/
What If 2 Gift Guide

xkcd