Research map / Image generation / Fast diffusion sampling
Unlocking Cross-Modal Biosignal Synthesis: A Temporally-Aware VAE-Diffusion Model
Chenyang Xu, Dezhen Wang, Hao Wang
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 |
|---|---|---|---|
| remove Enhanced Condition Fusion (cross-attention multi-scale ECG conditioning) (ablation) | Temporally-Aware VAE-Diffusion · EPHNOGRAM test set | Pearson correlation: 0.760 -> 0.810 (adding fusion) | Improved |
| add Temporal Attention blocks (ablation) | Temporally-Aware VAE-Diffusion · EPHNOGRAM test set | Pearson correlation: 0.719 -> 0.810 | Improved |
| add VAE loss during Phase 3 joint fine-tuning (ablation) | Temporally-Aware VAE-Diffusion · EPHNOGRAM test set | Pearson correlation: 0.686 -> 0.810 | Improved |
| use staged three-phase training instead of training jointly from the start (ablation) | Temporally-Aware VAE-Diffusion · EPHNOGRAM test set | Pearson correlation: 0.671 -> 0.810 | Improved |
| diffuse in VAE latent space instead of raw waveform space (ablation) | Temporally-Aware VAE-Diffusion · EPHNOGRAM test set | Pearson correlation: 0.662 -> 0.810 | Improved |
| use cross-attention fusion instead of concatenation (ablation) | Temporally-Aware VAE-Diffusion · EPHNOGRAM test set | S1 location error (ms): 14.5 -> 12.0 | Improved |
| use cross-attention fusion instead of concatenation (ablation) | Temporally-Aware VAE-Diffusion · EPHNOGRAM test set | S1 detection rate (%): 95.13 -> 95.95 | No clear effect |
| inject ECG condition only in the encoder path versus symmetric injection (ablation) | Temporally-Aware VAE-Diffusion · EPHNOGRAM test set | timing and perceptual quality | No clear effect |
| use the proposed VAE-Diffusion instead of a generic two-stage AudioLDM-style latent diffusion | Temporally-Aware VAE-Diffusion · EPHNOGRAM test set | Pearson correlation: 0.740 -> 0.810 | Improved |
| replace U-Net with matched DiT-style transformer denoiser | Temporally-Aware VAE-Diffusion · EPHNOGRAM test set | Pearson correlation: 0.810 -> 0.792 | Hurt |
| evaluate zero-shot on pathological versus normal recordings | Temporally-Aware VAE-Diffusion · PhysioNet/CinC 2016 training-a/MITHSDB | Pearson correlation: 0.77 -> 0.73 | Hurt |
| use DDIM sampling with 50 steps instead of 200-step DDPM (ablation) | Temporally-Aware VAE-Diffusion · inference latency | latency (ms): 215.3 -> approximately 54 | Improved |
Closest papers by meaning
- Region-Disentangled Diffusion Model for High-Fidelity PPG-to-ECG Translation (AAAI 2024 · similarity 0.88)
- Leveraging an ECG Beat Diffusion Model for Morphological Reconstruction from Indirect Signals (NeurIPS 2024 · similarity 0.87)
- ECHOPulse: ECG Controlled Echocardio-gram Video Generation (ICLR 2025 · similarity 0.87)
- Cylindrical Geodesic Flow Matching for Quasiperiodic Physiological Signal Transformation (NeurIPS 2026 · similarity 0.85)
- PhysDiff: Physiology-based Dynamicity Disentangled Diffusion Model for Remote Physiological Measurement (AAAI 2025 · similarity 0.85)
- PhysioWave: A Multi-Scale Wavelet-Transformer for Physiological Signal Representation (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.
