Research map / Bandits and online learning / Online convex optimization and games
Robust Frequent Directions with Application in Online Learning
Luo Luo, Cheng Chen, Zhihua Zhang, Wu-Jun Li, Tong Zhang
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 RFD instead of FD in the sketch | sketched online Newton step · rcv1 test set | accuracy (%): 95.6529 -> 96.1139 | Improved |
| use RFD instead of FD in the sketch | sketched online Newton step · farm-ads test set | accuracy (%): 89.9437 -> 90.3459 | Improved |
| use RFD instead of FD in the sketch | sketched online Newton step · real-sim test set | accuracy (%): 96.7361 -> 96.4037 | Hurt |
| use RFD instead of FD in matrix approximation | covariance sketching · relative spectral norm error | relative spectral error | Improved |
| use RFD instead of PFD in matrix approximation | covariance sketching · relative spectral norm error | relative spectral error | Improved |
| set alpha0=0 for RFD (no tuning) (ablation) | RFD-SON · rcv1 test set | accuracy (%): 96.1139 -> 96.4762 | Improved |
| set alpha0=0 for RFD (no tuning) (ablation) | RFD-SON · farm-ads test set | accuracy (%): 90.3459 -> 88.8978 | Hurt |
| vary alpha0 hyperparameter (ablation) | ADAGRAD, FD-SON, PFD-SON, RP-SON, Oja-SON · six LIBSVM datasets | test accuracy | Hurt |
| vary alpha0 hyperparameter (ablation) | RFD-SON · a9a, gisette, sido0, farm-ads | test accuracy | No clear effect |
| use sketched RFD0-SON instead of full Hessian | online Newton step · gisette | average iteration cost: 2.6296e-01 -> 1.4328e-03 | Improved |
Closest papers by meaning
- Efficient Second Order Online Learning by Sketching (NeurIPS 2016 · similarity 0.87)
- Efficient Adaptive Online Learning via Frequent Directions (TPAMI 2021 · similarity 0.86)
- Incremental Randomized Sketching for Online Kernel Learning (ICML 2019 · similarity 0.86)
- Dimension-Free Adaptive Subgradient Methods with Frequent Directions (ICML 2025 · similarity 0.84)
- Limited Memory Online Gradient Descent for Kernelized Pairwise Learning with Dynamic Averaging (AAAI 2024 · similarity 0.82)
- Can Gaussian Sketching Converge Faster on a Preconditioned Landscape? (ICML 2024 · similarity 0.81)
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.
