Research map / Probabilistic and scientific modeling / Variational and generative sampling
Importance Sampling for Minibatches
Dominik Csiba, Peter Richtárik
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 tau-importance sampling instead of tau-nice sampling on dense extreme-norm artificial data, tau=32 | dfSDCA · Artificial data, extreme norms (80% nonzeros) | empirical speedup: 154 : 28 (theoretical : empirical) | Improved |
| use tau-importance sampling on sparse extreme-norm artificial data, tau=32 | dfSDCA · Artificial data, extreme norms (10% nonzeros) | empirical speedup: 32 : 6.1 (theoretical : empirical) | Improved |
| use tau-importance sampling on chisq1 dense artificial data, tau=32 | dfSDCA · Artificial data, chisq1 (80% nonzeros) | empirical speedup: 10 : 3.9 (theoretical : empirical) | Improved |
| use tau-importance sampling on uniform-norm sparse artificial data, tau=1 | dfSDCA · Artificial data, uniform norms (10% nonzeros) | empirical speedup: 1.2 : 1.0 (theoretical : empirical) | No clear effect |
| use tau-importance sampling on uniform-norm dense artificial data, tau=32 | dfSDCA · Artificial data, uniform norms (80% nonzeros) | empirical speedup: 1.8 : 1.3 (theoretical : empirical) | Improved |
| use tau-importance sampling on aloi, tau=1 | dfSDCA · aloi | empirical speedup: 13 : 7.8 (theoretical : empirical) | Improved |
| use tau-importance sampling on aloi, tau=32 | dfSDCA · aloi | empirical speedup: 8.8 : 6.7 (theoretical : empirical) | Improved |
| use tau-importance sampling on w8a, tau=1 | dfSDCA · w8a | empirical speedup: 2.8 : 2.0 (theoretical : empirical) | Improved |
| use tau-importance sampling on ijcnn1, tau=32 | dfSDCA · ijcnn1 | empirical speedup: 2.3 : 1.8 (theoretical : empirical) | Improved |
| increase minibatch size tau from 1 to 32 for tau-imp over tau-nice on url (ablation) | dfSDCA · url | empirical speedup: 2.3 -> 1.7 | Hurt |
| increase minibatch size tau from 1 to 32 for tau-imp over tau-nice on dense chisq1 data (ablation) | dfSDCA · Artificial data, chisq1 (80% nonzeros) | empirical speedup: 1.3 -> 3.9 | Improved |
| use denser data (80% instead of 10% nonzeros) for extreme-norm data at tau=32 (ablation) | dfSDCA · Artificial data, extreme norms | empirical speedup: 6.1 -> 28 | Improved |
Closest papers by meaning
- Training Deep Models Faster with Robust, Approximate Importance Sampling (NeurIPS 2018 · similarity 0.87)
- Nearly Optimal Sample Complexity for Learning with Label Proportions (ICML 2025 · similarity 0.85)
- Boosting Test Performance with Importance Sampling--a Subpopulation Perspective (AAAI 2025 · similarity 0.85)
- Accelerating Neural Field Training via Soft Mining (CVPR 2024 · similarity 0.85)
- LEAPS: A discrete neural sampler via locally equivariant networks (ICML 2025 · similarity 0.84)
- Breaking the Frozen Subspace: Importance Sampling for Low-Rank Optimization in LLM Pretraining (NeurIPS 2025 · similarity 0.84)
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.
