ADA Lab @ UCSD

Project Triptych

Overview

Triptych is an end-to-end model selection management system (MSMS) that aims to simplify and accelerate the process of sourcing data/features and selecting ML models. Our guiding principles are to exploit the semantics of the data and the ML task to the extent possible to reduce work for the data scientist and reduce runtimes and costs. We apply these principles to remove or mitigate different bottlenecks in this end-to-end process, eventually unifying these components to yield an integrated ‘‘operating system’’ for ML analytics tasks. Please refer to the ACM SIGMOD Record paper below for more details of this vision.

Active Component Projects

Cerebro Efficient and reproducible model selection on deep learning systems.

Morpheus Integrating linear algebra and relational algebra to simplify feature engineering for ML.

SortingHat ML schema inference and automatic data preparation.

Publications

• Some Damaging Delusions of Deep Learning Practice (and How to Avoid Them)
  Arun Kumar, Supun Nakandala, and Yuhao Zhang
  KDD 2021 Deep Learning Day | Extended Abstract PDF | Talk slides | Talk video

• Towards an Optimized GROUP BY Abstraction for Large-Scale Machine Learning
  Side Li and Arun Kumar
  VLDB 2021 | Paper PDF | TechReport | Talk video | Code Release

• Distributed Deep Learning on Data Systems: A Comparative Analysis of Approaches
  Yuhao Zhang, Frank McQuillan, Nandish Jayaram, Nikhil Kak, Ekta Khanna, Orhan Kislal, Domino Valdano, and Arun Kumar
  VLDB 2021 | Paper PDF | TechReport | Talk video | Code release

• Intermittent Human-in-the-Loop Model Selection using Cerebro: A Demonstration
  Liangde Li, Supun Nakandala, and Arun Kumar
  VLDB 2021 Demo | Paper PDF | TechReport | Video

• Towards A Polyglot Framework for Factorized ML
  David Justo, Shaoqing Yi, Lukas Stadler, Nadia Polikarpova, and Arun Kumar
  VLDB 2021 (Industrial Track) | Paper PDF | TechReport | Talk video | Code coming soon

• Towards Benchmarking Feature Type Inference for AutoML Platforms
  Vraj Shah, Jonathan Lacanlale, Premanand Kumar, Kevin Yang, and Arun Kumar
  ACM SIGMOD 2021 | Paper PDF | TechReport | Talk Videos: Short Talk Long Talk | Data, Code, and Pre-trained Models on GitHub | Python library

• The CNN Hip Accelerometer Posture (CHAP) Method for Classifying Sitting Patterns from Hip Accelerometers: A Validation Study
  Mikael Anne Greenwood-Hickman, Supun Nakandala, Marta M. Jankowska, Fatima Tuz-Zahra, John Bellettiere, Jordan Carlson, Paul R. Hibbing, Jingjing Zou, Andrea Z. LaCroix, Arun Kumar, and Loki Natarajan
  Medicine and Science in Sports and Exercise Journal, 2021 | Paper PDF coming soon | Code

• Application of Convolutional Neural Network Algorithms for Advancing Sedentary and Activity Bout Classification
  Supun Nakandala, Marta Jankowska, Fatima Tuz-Zahra, John Bellettiere, Jordan Carlson, Andrea LaCroix, Sheri Hartman, Dori Rosenberg, Jingjing Zou, Arun Kumar, and Loki Natarajan
  Journal for the Measurement of Physical Behaviour, 2021 | Paper PDF and BibTeX | Code

• Cerebro: A Layered Data Platform for Scalable Deep Learning
  Arun Kumar, Supun Nakandala, Yuhao Zhang, Side Li, Advitya Gemawat, and Kabir Nagrecha
  CIDR 2021 (Vision paper) | Paper PDF and BibTeX | Talk video

• Cerebro: A Data System for Optimized Deep Learning Model Selection
  Supun Nakandala, Yuhao Zhang, and Arun Kumar
  VLDB 2020 | Paper PDF and BibTeX | Errata | TechReport | Talk videos: Youtube Bilibili | Blog post | SAIS Talk video | Source code and documentation

• Enabling and Optimizing Non-linear Feature Interactions in Factorized Linear Algebra
  Side Li, Lingjiao Chen, and Arun Kumar
  ACM SIGMOD 2019 | Paper PDF and BibTeX | Code and Data on Github

• Tuple-oriented Compression for Large-scale Mini-batch Stochastic Gradient Descent
  Fengan Li, Lingjiao Chen, Yijing Zeng, Arun Kumar, Jeffrey Naughton, Jignesh Patel, and Xi Wu
  ACM SIGMOD 2019 | Paper PDF | TechReport | Code on GitHub

• Model-based Pricing for Machine Learning in a Data Marketplace
  Lingjiao Chen, Paraschos Koutris, and Arun Kumar
  ACM SIGMOD 2019 | Paper PDF | TechReport | Code and Data coming soon

• Cerebro: Efficient and Reproducible Model Selection on Deep Learning Systems
  Supun Nakandala, Yuhao Zhang, and Arun Kumar
  ACM SIGMOD 2019 DEEM Workshop | Paper PDF and BibTeX | TechReport | Blog post

• The ML Data Prep Zoo: Towards Semi-Automatic Data Preparation for ML
  Vraj Shah and Arun Kumar
  ACM SIGMOD 2019 DEEM Workshop | Paper PDF and BibTeX | TechReport | Blog post | Data Prep Zoo Repository on GitHub

• Demonstration of Nimbus: Model-based Pricing for Machine Learning in a Data Marketplace
  Lingjiao Chen, Hongyi Wang, Leshang Chen, Paraschos Koutris, and Arun Kumar
  ACM SIGMOD 2019 Demo | Paper PDF | Video coming soon

• A Comparative Evaluation of Systems for Scalable Linear Algebra-based Analytics
  Anthony Thomas and Arun Kumar
  VLDB 2018/2019 | Paper PDF | TechReport | Code and Data

• Are Key-Foreign Key Joins Safe to Avoid when Learning High-Capacity Classifiers?
  Vraj Shah, Arun Kumar, and Xiaojin Zhu.
  VLDB 2018 | Paper PDF and BibTeX| TechReport | Code and Data

• Towards Linear Algebra over Normalized Data
  Lingjiao Chen, Arun Kumar, Jeffrey Naughton, and Jignesh Patel
  VLDB 2017 | Paper PDF | TechReport | Code and Data

• Model-based Pricing: Do Not Pay for More than What You Learn!
  Lingjiao Chen, Paraschos Koutris, and Arun Kumar
  ACM SIGMOD 2017 DEEM Workshop | Paper PDF

• Cerebro: A System to Manage Deep Learning for Relational Data Analytics
  Arun Kumar
  CIDR 2017 Abstract | Paper PDF

• To Join or Not to Join? Thinking Twice about Joins before Feature Selection
  Arun Kumar, Jeffrey Naughton, Jignesh M. Patel, and Xiaojin Zhu
  ACM SIGMOD 2016 | Paper PDF and BibTeX | TechReport | Code and Data

• Model Selection Management Systems: The Next Frontier of Advanced Analytics
  Arun Kumar, Robert McCann, Jeffrey Naughton, and Jignesh M. Patel
  ACM SIGMOD Record Dec 2015 Vision Track | Paper PDF

Technical Reports

• An Empirical Study on the (Non-)Importance of Cleaning Categorical Duplicates before ML
  Vraj Shah, Thomas Parashos, and Arun Kumar
  Under submission | TechReport

• Nautilus: An Optimized System for Deep Transfer Learning over Evolving Training Datasets
  Supun Nakandala and Arun Kumar
  Under submission | TechReport

• SystemX: A Scalable and Optimized Data System for Large Multi-Model Deep Learning
  Kabir Nagrecha and Arun Kumar
  Under submission | TechReport

• Improving Feature Type Inference Accuracy of TFDV with SortingHat
  Vraj Shah, Kevin Yang, and Arun Kumar
  TechReport

Past Projects

Hamlet Exploiting database schema information to simplify data sourcing.

Nimbus Enabling the first ML-aware cloud-based commodity market for the new black gold: training data.

SLAB The first comprehensive benchmark comparison of scalable linear algebra systems.