Research map / Learning theory and algorithms / Matrix and tensor methods
LLORMA: Local Low-Rank Matrix Approximation
Joonseok Lee, Seungyeon Kim, Guy Lebanon, Yoram Singer, Samy Bengio
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 LLORMA local low-rank models instead of global SVD (rank 5) | SVD · MovieLens 1M | RMSE: 0.8737 -> 0.8537 | Improved |
| use LLORMA local low-rank models instead of global SVD (rank 5) | SVD · Yelp | RMSE: 1.4775 -> 1.2358 | Improved |
| use LLORMA local low-rank models instead of global SVD (rank 5) | SVD · Bookcrossing | RMSE: 3.3583 -> 2.9482 | Improved |
| increase rank from 1 to 20 for LLORMA (ablation) | LLORMA · MovieLens 1M | RMSE: 0.9135 -> 0.8333 | Improved |
| use Global LLORMA instead of Parallel LLORMA (rank 1) (ablation) | LLORMA · MovieLens 100K | RMSE: 0.9426 -> 0.9072 | Improved |
| gap between Global and Parallel LLORMA as rank increases to 10 (ablation) | LLORMA · MovieLens 100K | RMSE: 0.9426 -> 0.9072 at rank 1; 0.8985 -> 0.8975 at rank 10 | Mixed |
| use Parallel LLORMA instead of Global LLORMA for training time (rank 10) (ablation) | LLORMA · MovieLens 100K | training time: 28:59 -> 11:49 | Improved |
| k-means anchor points instead of random sampled anchors (10 local models, rank 5) (ablation) | Global LLORMA · MovieLens 100K | RMSE: 0.9240 -> 0.9190 | Improved |
| Coverage anchor selection instead of Complete (10 local models, rank 1) (ablation) | Global LLORMA · MovieLens 100K | RMSE: 0.9276 -> 0.9360 | Hurt |
| increase number of anchor points from 10 to 20 (rank 5, Complete) (ablation) | Global LLORMA · MovieLens 100K | RMSE: 0.9240 -> 0.9125 | Improved |
| use LLORMA instead of Bagging of independent models | LLORMA · MovieLens 100K | RMSE: 0.8930 -> 0.8881 | Improved |
| use LLORMA instead of Bagging of independent models | LLORMA · MovieLens 1M | RMSE: 0.8481 -> 0.8423 | Improved |
Later papers that compared with this method
4 later accepted papers report LLORMA as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.
- Non-Compensatory Psychological Models for Recommender Systems (AAAI 2019 · better on 3 of 3 shared results)
- Mixture-Rank Matrix Approximation for Collaborative Filtering (NeurIPS 2017 · better on 6 of 6 shared results)
- ERMMA: Expected Risk Minimization for Matrix Approximation-based Recommender Systems (AAAI 2017 · better on 4 of 4 shared results)
- Low-Rank Matrix Approximation with Stability (ICML 2016 · better on 2 of 2 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Low-Rank Matrix Approximation with Stability (ICML 2016 · similarity 0.88)
- Mixture-Rank Matrix Approximation for Collaborative Filtering (NeurIPS 2017 · similarity 0.87)
- Thy Friend is My Friend: Iterative Collaborative Filtering for Sparse Matrix Estimation (NeurIPS 2017 · similarity 0.86)
- Learning-Based Low-Rank Approximations (NeurIPS 2019 · similarity 0.86)
- ERMMA: Expected Risk Minimization for Matrix Approximation-based Recommender Systems (AAAI 2017 · similarity 0.85)
- Recommendations with Sparse Comparison Data: Provably Fast Convergence for Nonconvex Matrix Factorization (ICML 2025 · similarity 0.85)
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.
