Research map / Probabilistic and scientific modeling / Variational and generative sampling
Integer Discrete Flows and Lossless Compression
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 an integer discrete flow for lossless compression instead of FLIF | CIFAR10 compression · CIFAR10 test | bits per dimension: 4.37 -> 3.34 | Helped |
| Use an integer discrete flow for lossless compression instead of Bit-Swap | ImageNet32 compression · ImageNet32 test | bits per dimension: 4.50 -> 4.18 | Helped |
| Use an integer discrete flow for lossless compression instead of FLIF | ImageNet64 compression · ImageNet64 test | bits per dimension: 4.55 -> 3.90 | Helped |
| Use an integer discrete flow for lossless compression instead of JP2-WSI | Histology compression · Histology | bits per dimension: 3.04 -> 2.42 | Helped |
| Dequantize, disable rounding, and use a continuous prior (ablation) | IDF · CIFAR10 | bits per dimension (negative log2-likelihood): 3.32 -> 3.31 | Helped |
| Dequantize, disable rounding, and use a continuous prior (ablation) | IDF · ImageNet32 | bits per dimension (negative log2-likelihood): 4.15 -> 4.13 | Helped |
| Dequantize, disable rounding, and use a continuous prior (ablation) | IDF · ImageNet64 | bits per dimension (negative log2-likelihood): 3.90 -> 3.85 | Helped |
| Replace standard coupling with lower triangular coupling (ablation) | IDF · ImageNet32 | bits per dimension (negative log2-likelihood): 4.15 -> 4.15 (standard ±0.009; lower triangular ±0.007) | No clear effect |
| Use a 75%–25% input split for integer discrete coupling | IDF · Image density modeling | performance | Helped |
| Add more flow layers after the point where rounding-induced gradient bias dominates | IDF · Image density modeling | bits per dimension | Hurt |
Closest papers by meaning
- IDF++: Analyzing and Improving Integer Discrete Flows for Lossless Compression (ICLR 2021 · similarity 0.93)
- Fast Lossless Neural Compression with Integer-Only Discrete Flows (ICML 2022 · similarity 0.89)
- Compression with Flows via Local Bits-Back Coding (NeurIPS 2019 · similarity 0.89)
- iFlow: Numerically Invertible Flows for Efficient Lossless Compression via a Uniform Coder (NeurIPS 2021 · similarity 0.88)
- Towards Lossless Implicit Neural Representation via Bit Plane Decomposition (CVPR 2025 · similarity 0.85)
- ReDi: Rectified Discrete Flow (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.