Research map / Learning theory and algorithms / Matrix and tensor methods
Fast Exact Matrix Completion: A Unified Optimization Framework for Matrix Completion
Dimitris Bertsimas, Michael Lingzhi Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use fastImpute instead of IMC on Netflix with side information | fastImpute · Netflix full 471,268x14,538 (side information) | MAPE: 21.0% -> 20.7% | Improved |
| use fastImpute instead of SIALS on Netflix without side information | fastImpute · Netflix full 471,268x14,538 | MAPE: 20.1% -> 13.8% | Improved |
| use fastImpute instead of SIALS on full Netflix without side information | fastImpute · Netflix full 471,268x14,538 | runtime: 38256s -> 2098s | Improved |
| use fastImpute instead of IMC on synthetic data with 95% missing | fastImpute · Synthetic n=1e4, m=1e3, p=100, k=5 | MAPE: 2.5% -> 0.1% | Improved |
| use stochastic fastImpute instead of full-gradient projImpute (ablation) | fastImpute · Synthetic n=1e6, m=1e3, p=100, k=5 | runtime: 6035s -> 390s | Improved |
| use fastImpute instead of SIALS, SISVD, MFSGD and GROUSE on large realistic combinations | fastImpute · Synthetic, no side information | runtime and MAPE: -90% T and -71% MAPE vs SIALS; -99% and -95% vs SISVD; -96% and -45% vs MFSGD; -81% and -15% vs GROUSE | Improved |
| increase the fraction of missing entries from 20% to 95% | IMC · Synthetic with side information | MAPE: 0.01% -> 2.5% | Hurt |
| increase the fraction of missing entries from 20% to 95% | SIALS · Synthetic without side information | MAPE: 0.6% -> 12.7% | Hurt |
| increase the fraction of missing entries from 20% to 95% | fastImpute · Synthetic with side information | MAPE: 0.7% -> 0.1% | Improved |
| increase n (matrix size) toward 1e6 | MFSGD · Synthetic without side information | MAPE: 2.1% -> 7.0% | Hurt |
| increase n from 1e3 to 1e6 | fastImpute · Synthetic without side information | MAPE: 3.5% -> 2.1% | Improved |
| increase true rank k from 5 to 30 | fastImpute · Synthetic without side information | runtime: 4.2s -> 49.5s | Mixed |
Closest papers by meaning
- Provable Efficient Online Matrix Completion via Non-convex Stochastic Gradient Descent (NeurIPS 2016 · similarity 0.92)
- A New Theory for Matrix Completion (NeurIPS 2017 · similarity 0.90)
- Matrix Completion has No Spurious Local Minimum (NeurIPS 2016 · similarity 0.90)
- Symmetric Matrix Completion with ReLU Sampling (ICML 2024 · similarity 0.89)
- Generalization Analysis of Deep Non-linear Matrix Completion (ICML 2024 · similarity 0.88)
- Semi-Random Matrix Completion via Flow-Based Adaptive Reweighting (NeurIPS 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
