Research map / Optimization / Optimizers
LITE: A Lightweight Lazy Sampler for Efficient SGD
Amir Daghestani, Mikael Johansson
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes LITE, a lightweight lazy importance sampler for SGD that aims to make gradient-norm-proportional sampling practical in deep learning under proxy cost, stale scores, and fair update budgets.
Problem
Importance sampling for SGD is theoretically variance-minimizing, but in deep learning it often fails to beat uniform sampling once the cost of computing proxy scores, stale scores, and fair update budgets are accounted for.
Approach
A lightweight, lazily updated sampler that samples examples in proportion to gradient-norm-based scores while keeping proxy computation cheap and handling staleness of scores; abstract provides no further technical detail.
Surprising findings
- Importance sampling often fails to outperform uniform sampling in modern deep learning once proxy cost, stale scores, and fair update budgets are accounted for.
Experimental setup
Baselines: uniform sampling
Builds on: importance sampling for SGD
Closest papers by meaning
- Not All Samples Are Created Equal: Deep Learning with Importance Sampling (ICML 2018 · similarity 0.90)
- LEAPS: A discrete neural sampler via locally equivariant networks (ICML 2025 · similarity 0.83)
- Estimating the Probabilities of Rare Outputs in Language Models (ICLR 2025 · similarity 0.83)
- Optimizing importance weighting in the presence of sub-population shifts (ICLR 2025 · similarity 0.83)
- Breaking the Frozen Subspace: Importance Sampling for Low-Rank Optimization in LLM Pretraining (NeurIPS 2025 · similarity 0.83)
- Swift Sampler: Efficient Learning of Sampler by 10 Parameters (NeurIPS 2024 · similarity 0.83)
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.
