MLbase: A Distributed Machine Learning System
Talkat Parallel and Distributed Algorithms for Inference and Optimization (Simons Institute, Oct. 21–24, 2013) · 2013-10-21..10-24·
summary
A declarative system called MLBase aims to let users request a classifier and get models back without coding algorithms, using MQL for queries and an MLI runtime layer that runs on Spark or Tupleware. It returns a streaming set of models with a Top five-minute cutoff and uses binders to plug algorithms in. Early release includes MLI and several binders in Spark, experiments show MLI performance near GraphLab/MATLAB and vectorized inner loops give large cache-driven speedups, while optimizer trades time versus quality.
AI-written summary · about
chapters (25)
- 0:00Intro
- 0:50Bias
- 2:58Why MLbase
- 4:15Meet the team
- 4:54What is the goal
- 6:55Code examples
- 8:01Hints
- 8:25How does it work
- 9:20Initial design
- 10:46Optimization
- 11:01MLeye abstraction
- 11:49MLI abstraction
- 12:07Bynars
- 12:48Spark
- 13:01ML vs MATLAB
- 13:43Optimization of ML
- 18:12Robustness
- 19:11Optimizations
- 20:02Inbalance of Labels
- 20:59Data Quality
- 21:35Optimization Potential
- 22:39System Performance
- 23:34Call Loops
- 24:42Direction
- 25:23Summary
related talks
06mML for Systems and Systems for ML
12mOpenMLDB: An Open-Source Real-Time Feature Platform
07mAmeet Talwalkar, Assistant Professor of Computer Science, UCLA @ MLconf SF Part 1
16mAmeet Talwalkar, Assistant Professor of Computer Science, UCLA @ MLconf SF Part 2
39mMachine Learning with Scala on Spark
24mMLflow: An open platform to simplify...
46mML in Java, YES it's possible!
01h 04mIntroduction to ML.NET