Research map / Label-efficient and robust learning / Recognition and long-tail
Joint Enhancement and Classification using Coupled Diffusion Models of Signals and Logits
Gilad Nurko, Roi Benita, Yehoshua Dissen, Tomohiro Nakatani, Marc Delcroix, Shoko Araki, Joseph Keshet
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 |
|---|---|---|---|
| couple logit diffusion with signal diffusion (Parallel) instead of enhance-then-classify | Parallel · CIFAR-100 30% GN | accuracy: 60.4 -> 68.5 | Improved |
| couple logit diffusion with signal diffusion (Nested) | Nested · CIFAR-10 30% GN | accuracy: 60.4 -> 81.2 | Improved |
| couple logit diffusion with signal diffusion (Parallel) | Parallel · IN32-100 30% GN | accuracy: 50.6 -> 65.5 | Improved |
| couple diffusion on MNIST instead of URIE | Nested · MNIST 30% GN | accuracy: 99.3 -> 99.1 | Hurt |
| couple logit diffusion with signal diffusion (Nested) for ASR | Nested · EARS-WHAM | WER (%): 10.04 -> 9.75 | Improved |
| couple logit diffusion with signal diffusion (Nested) for ASR | Nested · EARS-Reverb | WER (%): 4.48 -> 3.83 | Improved |
| use Dissen enhancer trained with task loss instead of audio-only enhancement | Dissen · EARS-Reverb | WER (%): 4.48 -> 6.46 | Hurt |
| condition on clean estimate instead of noisy sample in Parallel (ablation) | Parallel · CIFAR-100 30% GN | accuracy: 67.0 -> 68.5 | Improved |
| condition on clean estimate instead of noisy sample in Parallel (ablation) | Parallel · IN32-100 Gaussian Blur | accuracy: 61.8 -> 63.3 | Improved |
| use DDPM stochastic sampling instead of DDIM in Parallel (ablation) | Parallel · CIFAR-100 Gaussian Blur | accuracy: 64.8 -> 69.7 | Improved |
| use DDPM stochastic sampling instead of DDIM in Parallel (ablation) | Parallel · CIFAR-100 15% GN | accuracy: 70.7 -> 73.5 | Improved |
| scale to 224x224 ImageNet 10-class with Parallel | Parallel · ImageNet 224 30% GN | accuracy: 86.5 -> 92.4 | Improved |
Closest papers by meaning
- Noise Matters: Optimizing Matching Noise for Diffusion Classifiers (NeurIPS 2025 · similarity 0.85)
- RestoreGrad: Signal Restoration Using Conditional Denoising Diffusion Models with Jointly Learned Prior (ICML 2025 · similarity 0.85)
- An Efficient Framework for Enhancing Discriminative Models via Diffusion Techniques (AAAI 2025 · similarity 0.85)
- Complex-Cycle-Consistent Diffusion Model for Monaural Speech Enhancement (AAAI 2025 · similarity 0.84)
- Your Diffusion Model is Secretly a Noise Classifier and Benefits from Contrastive Training (NeurIPS 2024 · similarity 0.84)
- ADDP: Learning General Representations for Image Recognition and Generation with Alternating Denoising Diffusion Process (ICLR 2024 · 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.
