Research map / Architectures and efficiency / RNNs and normalization
Energy-Efficient Random Variate Generation via Compressed Lookup Tables
Johann Ukrow, Anna Kazachkova, Nicolas Alder, Sven Köhler, Rainer Schlosser, Ralf Herbrich
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 cLUT instead of JAX sampling | discrete sampler · Python, n in [10^6,10^7) | sampling time (s): 0.7982 -> 0.1016 | Improved |
| use cLUT instead of NumPy sampling | discrete sampler · Python, n in [10^1,10^5) | sampling time (s): 0.6680 -> 0.0374 | Improved |
| use cLUT instead of Alias method | C sampler · C, n in [10^7,10^8) | energy per sample (nJ): 887.653 -> 450.155 | Improved |
| use cLUT instead of Alias method | C sampler · C, n in [10^3,10^4) | time per sample (ns): 26.803 -> 15.475 | Improved |
| use cLUT instead of NumPy discrete sampler | TrueSkill · TrueSkill | sampling time and energy: 1.65 -> 0.46 | Improved |
| use cLUT instead of NumPy continuous mixture sampler | TrueSkill · TrueSkill | sampling time: 0.88 -> 0.46 | Improved |
| use cLUT instead of JAX default sampler | diffusion model · Diffusion model | energy: 4702.91 -> 2962.17 | Improved |
| use cLUT instead of JAX | discrete sampler preprocessing · Python, n in [10^6,10^7) | preprocessing time (s): 0.6528 -> 0.3925 | Improved |
| use cLUT instead of NumPy | discrete sampler preprocessing · Python, n in [10^6,10^7) | preprocessing time (s): 0.0308 -> 0.3925 | Hurt |
| use cLUT instead of Alias method | C sampler · peak memory | memory | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- PyTorch — PyTorch: An Imperative Style, High-Performance Deep Learning Library (NeurIPS 2019) · better on 2 of 4 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
- Energy-Based Modelling for Discrete and Mixed Data via Heat Equations on Structured Spaces (NeurIPS 2024 · similarity 0.82)
- Faster Sampling via Stochastic Gradient Proximal Sampler (ICML 2024 · similarity 0.82)
- Batched Energy-Entropy acquisition for Bayesian Optimization (NeurIPS 2024 · similarity 0.82)
- Joint Learning of Energy-based Models and their Partition Function (ICML 2025 · similarity 0.82)
- Differentially Private High-dimensional Variable Selection via Integer Programming (NeurIPS 2025 · similarity 0.82)
- Scalable Discrete Diffusion Samplers: Combinatorial Optimization and Statistical Physics (ICLR 2025 · similarity 0.82)
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.
