Research map / Probabilistic and scientific modeling / Variational and generative sampling
OSOA: One-Shot Online Adaptation of Deep Generative Models for Lossless Compression
Chen Zhang, Shifeng Zhang, Fabio Maria Carlucci, Zhenguo Li
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 |
|---|---|---|---|
| adapt the pretrained model online once per batch | HiLLoC CIFAR10 · SET32 | theoretical bpd: 3.830 -> 3.505 | Improved |
| adapt the pretrained model online once per batch | HiLLoC CIFAR10 · SET64 | theoretical bpd: 3.561 -> 3.090 | Improved |
| adapt the pretrained model online once per batch | HiLLoC CIFAR10 · SET128 | theoretical bpd: 3.268 -> 2.682 | Improved |
| replace OSOA with FineTune v2 and save the adapted model | HiLLoC CIFAR10 · SET32 | total storage bpd: OSOA 3.505; FineTune v2 3.322 + 3.258 | Hurt |
| add multiple optimizer updates per batch (ablation) | HiLLoC CIFAR10 OSOA · SET32 | theoretical bpd: 3.505 -> 3.109 | Improved |
| add 100 optimizer updates per batch (ablation) | IDF++ CIFAR10 OSOA · SET32 | theoretical bpd: 3.225 -> 3.105 | Hurt |
| add 500 optimizer updates per batch (ablation) | HiLLoC CIFAR10 OSOA · SET32 | theoretical bpd: 3.505 -> 3.109 | Improved |
| early-stop adaptation at selected batches instead of adapting through all batches (ablation) | HiLLoC CIFAR10 OSOA · SET128 | theoretical bpd: 2.820; 2.774; 2.717; 2.690 | Mixed |
Closest papers by meaning
- Practical lossless compression with latent variables using bits back coding (ICLR 2019 · similarity 0.86)
- Evaluating Lossy Compression Rates of Deep Generative Models (ICML 2020 · similarity 0.85)
- Compression with Flows via Local Bits-Back Coding (NeurIPS 2019 · similarity 0.85)
- Real-Time Adaptive Image Compression (ICML 2017 · similarity 0.85)
- Once-for-All: Controllable Generative Image Compression with Dynamic Granularity Adaptation (ICLR 2025 · similarity 0.84)
- CALLIC: Content Adaptive Learning for Lossless Image Compression (AAAI 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.
