Workshop on Distributions, Relational Algebra, Graphs, Semi-Rings, Tensors, and All ThatDRAGSTERS 2023
Workshop on Distributions, Relational Algebra, Graphs, Semi-Rings, Tensors, and All That (aka. DRAGSTERS)
Over the last decade, researchers working in compilers, programming languages, databases, machine learning, supercomputing, and related fields have productively explored both analogies and direct isomorphisms between objects like tensors, graphs, probability distributions and relations. These connections have consequences up and down computing stacks: ranging from common APIs and language concepts, to shared algorithmic and complexity results, to common compilation strategies, and to hardware design considerations.
This workshop aims to bring together a broad set of researchers in these areas to better understand the similarities and differences. The program will contain a mixture of invited talks and short talks. Submission and talk materials will not be published. This workshop seeks talks on topics ranging from (but not limited to):
- Dense & sparse linear/tensor algebra compilers
- Graph processing systems
- Systems for probabilistic graphical models
- Semi-ring/graph BLAS API design & implementation
- Mixed linear/relational-algebra languages
- Communication lower bounds for tensor contractions, etc.
- Asymptotically optimal join query and tensor contraction algorithms
- Integrating query processing with downstream computations (e.g. ML pipelines)
- Hardware for computing on sparse and irregular data
- Data-structure synthesis
- Datalog connections for all of the above
- Domain-specific languages for image processing, simulation, machine learning, data analytics, and other application domains making use of these abstractions
Preference will be given to talks about new work, but talks need not necessarily communicate novel research. Historical or scholarly talks on the above subjects will be judged based on their ability to connect disparate topics.
Sat 17 JunDisplayed time zone: Eastern Time (US & Canada) change
09:00 - 11:00 | |||
09:00 20mTalk | Matrix Decompositions over Database Joins DRAGSTERS Dan Olteanu University of Zurich, Nils Vortmeier Ruhr University Bochum, Dorde Zivanovic University of Oxford | ||
09:20 20mTalk | NASOQ: Numerically Accurate Sparsity-Oriented QP Solver DRAGSTERS | ||
09:40 20mTalk | UniSparse: An Intermediate Language and Compiler for General Sparse Format Customization DRAGSTERS Jie Liu Cornell University, Zhongyuan Zhao , Zijian Ding Peking University, Benjamin Brock Parallel Computing Lab (PCL), Intel, Hongbo Rong Intel Labs, Zhiru Zhang Cornell University, USA | ||
10:00 20mTalk | Unification as a means of completing partial data structures DRAGSTERS Joachim Kristensen University of Oslo, Robin Kaarsgaard University of Southern Denmark, Michael Kirkedal Thomsen University of Oslo & University of Copenhagen | ||
10:20 20mTalk | Formalizing DRAGSTERS DRAGSTERS Scott Kovach Stanford University | ||
10:40 20mTalk | Scaling Decision--Theoretic Probabilistic Programming Through Factorization DRAGSTERS |
11:20 - 12:30 | |||
11:20 30mTalk | Keynote (Fredrik Kjolstad): Portable Compilation of Sparse Computation DRAGSTERS Fredrik Kjolstad Stanford University | ||
11:50 20mTalk | F-IVM: Analytics over Relational Databases under Updates DRAGSTERS Ahmet Kara University of Zurich, Milos Nikolic University of Edinburgh, Dan Olteanu University of Zurich, Haozhe Zhang University of Zurich | ||
12:10 20mTalk | TeAAL: A Declarative Framework for Modeling Sparse Tensor Accelerators DRAGSTERS Nandeeka Nayak University of Illinois at Urbana-Champaign, Toluwanimi O. Odemuyiwa University of California, Davis, Shubham Ugare University of Illinois at Urbana-Champaign, Christopher W. Fletcher University of Illinois--Urbana Champaign, Michael Pellauer Nvidia, Joel S Emer MIT/NVIDIA |
Accepted Talks
Call for Talks
We welcome talk submissions about and connecting the previously listed topics. All submissions and reviews will take place on HotCRP.
Talk Abstracts
We invite submission for talks broadly, including talks covering published and in-progress work, and including talks drawing historical connections. Submissions should be in the form of a 1 to 2-page extended abstract that describes the central insights and ideas to be discussed in the proposed talk.
Talks are intended to foster discussion. The program will include time for open-ended discussion inspired by the talks as well as Q&A.