Mihail Stoian

Publications

2025

• DPconv: Super-Polynomially Faster Join Ordering [code]
  Mihail Stoian, Andreas Kipf
  SIGMOD 2025
  
  [Slides @Microsoft GSL]

  TL;DR

  
  

  Forget about $O(3^n)$-time dynamic programming evaluations.

• Virtual: Compressing Data Lake Files
  Mihail Stoian, Alexander van Renen, Jan Kobiolka, Ping-Lin Kuo, Andreas Zimmerer, Josif Grabocka, Andreas Kipf
  EDBT Demo 2025

  TL;DR

  
  

  Virtual learns sparse regressors to level up Parquet file sizes, while having bounded column scan overhead in the number of reference columns.

• Optimizing Linearized Join Enumeration by Adapting to the Query Structure
  Altan Birler, Mihail Stoian, Thomas Neumann
  BTW 2025

  TL;DR

  
  

  Just forget about LinDP - this is way faster.

2024

• Lightweight Correlation-Aware Table Compression [code] [recording]
  Mihail Stoian, Alexander van Renen, Jan Kobiolka, Ping-Lin Kuo, Josif Grabocka, Andreas Kipf
  Table Representation Learning @ NeurIPS 2024
  
  [Slides @TRL]

  TL;DR

  
  

  Virtual learns sparse regressors to level up Parquet file sizes, while having bounded column scan overhead in the number of reference columns.

• Unified Mechanism-Specific Amplification by Subsampling and Group Privacy Amplification [code]
  Jan Schuchardt, Mihail Stoian*, Arthur Kosmala*, Stephan Günnemann
  NeurIPS 2024

  TL;DR

  
  

  The differential privacy framework for deriving mechanism-specific guarantees.

• DataLoom: Simplifying Data Loading with LLMs [code]
  Alexander van Renen, Mihail Stoian, Andreas Kipf
  VLDB Demo 2024

• Approximate Min-Sum Subset Convolution
  Mihail Stoian
  WAOA @ ALGO 2024
  
  [Slides @WAOA]

  TL;DR

  
  

  First proposal for approximate min-sum subset convolution. This results in out-of-the-box exp-time $(1 + \varepsilon)$-approximations for prize-collecting Steiner tree, min-cost $k$-coloring, protein networks, and more applications in computational biology.

• Corra: Correlation-Aware Column Compression
  Hanwen Liu, Mihail Stoian, Alexander van Renen, Andreas Kipf
  CloudDB @ VLDB 2024

  TL;DR

  
  

  Are you still using FOR-, Delta-, RLE-encodings? Correlation-aware column encodings can compress your data even better!

• On the Optimal Linear Contraction Order of Tree Tensor Networks, and Beyond [netzwerk]
  Mihail Stoian, Richard Milbradt, Christian B. Mendl
  SIAM Journal on Scientific Computing
  
  → Check out our package netzwerk with plug-in for opt_einsum and cotengra.

  TL;DR

  
  

  Polynomial-time contraction ordering algorithm for tree tensor networks for the total contraction cost. Extension of the well-known IKKBZ algorithm in databases.

2023

• Fast Joint Shapley Values [code] [recording]
  Mihail Stoian
  SRC @ SIGMOD 2023

• Faster FFT-based Wildcard Pattern Matching [code] [recording]
  Mihail Stoian
  SRC @ SIGMOD 2023

2022

• Concurrent Link-Cut Trees [code] [recording]
  Mihail Stoian | Advised by Jana Giceva and Philipp Fent
  SRC @ SIGMOD 2022

2021

• Towards Practical Learned Indexing [code] [recording]
  Mihail Stoian, Andreas Kipf, Ryan Marcus, and Tim Kraska
  AIDB @ VLDB 2021

• Benchmarking Learned Indexes [blog] [code] [leaderboard]
  Ryan Marcus, Andreas Kipf, Alexander van Renen, Mihail Stoian, Sanchit Misra, Alfons Kemper, Thomas Neumann, and Tim Kraska
  VLDB 2021

2020

• RadixSpline: A Single-Pass Learned Index [code] [talk]
  Andreas Kipf^*, Ryan Marcus^*, Alexander van Renen^*, Mihail Stoian, Alfons Kemper, Tim Kraska, and Thomas Neumann
  aiDM @ SIGMOD 2020

2019

• SOSD: A Benchmark for Learned Indexes [code]
  Andreas Kipf^*, Ryan Marcus^*, Alexander van Renen^*, Mihail Stoian, Alfons Kemper, Thomas Neumann, and Tim Kraska
  ML For Systems @ NeurIPS 2019

Preprints

• Did Fourier Really Meet Möbius? Fast Subset Convolution via FFT
  Mihail Stoian

  TL;DR

  
  

  There is no need for Zeta/Möbius transforms in fast subset convolution. FFT suffices. Even in the same running time.

• TSP Escapes the $O(2^n n^2)$ Curse [proof-of-concept] [blog]
  Mihail Stoian

  TL;DR++

  
  

  This xkcd needs a remake.

  Original: https://xkcd.com/399/

Invited Talks

• DPconv: Super-Polynomially Faster Join Ordering
  January, 2025
  @Microsoft GSL

• Virtual: Compressing World's Parquet Files
  January, 2025
  @TUMuchData

• What do databases and tensor networks have in common?
  August, 2023
  @Universität Jena @Joachim Giesen's group. Check out their amazing tools: Matrix Calculus, used in RelationalAI's AutoDiff, and more!

Contact

mihail.stoian@utn.de