Research map / Image generation / Autoregressive and latent generation
SIGMA-PPG: Statistical-prior Informed Generative Masking Architecture for PPG Foundation Model
Zongheng Guo, Tao Chen, Yang Jiao, Yi Pan, Xiao Hu, Manuela Ferrario
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 |
|---|---|---|---|
| add semantic consistency loss L_Con to the tokenizer (ablation) | SIGMA-PPG (30M) · Systolic BP, linear probing | MAE: 15.62 -> 13.07 | Improved |
| add semantic consistency loss L_Con to the tokenizer (ablation) | SIGMA-PPG (30M) · Heart rate, linear probing | MAE: 9.21 -> 5.83 | Improved |
| add semantic consistency loss L_Con to the tokenizer (ablation) | SIGMA-PPG · full fine-tuning, all downstream tasks | MAE and AUC | Improved |
| replace random masking with prior-guided adversarial masking (ablation) | SIGMA-PPG · regression and classification downstream tasks | MAE and AUC | Improved |
| use unconstrained adversarial Teacher (no statistical prior) instead of prior-guided (ablation) | SIGMA-PPG · pre-training masking | representation quality | Hurt |
| use static prior-based masking instead of adaptive Teacher (ablation) | SIGMA-PPG · pre-training masking | representation quality | Hurt |
| use standard random masking (ablation) | SIGMA-PPG · pre-training masking | representation quality | Hurt |
| use generative masked pretraining vs best contrastive baseline | SIGMA-PPG (30M) · SpO2, BIDMC, full fine-tuning | MAE: 1.265 -> 0.1457 | Improved |
| scale SIGMA-PPG from 5M to 30M parameters (ablation) | SIGMA-PPG · full fine-tuning average | AUC: 0.8596 -> 0.9082 | Improved |
| use SIGMA-PPG instead of Pulse-PPG under linear probing on classification | SIGMA-PPG (30M) · linear probing classification average | AUC: 0.8820 -> 0.8510 | Hurt |
| use SIGMA-PPG instead of Pulse-PPG under linear probing on regression | SIGMA-PPG (30M) · linear probing regression average | MAE: 6.781 -> 6.636 | Improved |
| use SIGMA-PPG (30M) instead of Pulse-PPG for compute | SIGMA-PPG (30M) · Computational cost | FLOPs: 27.11 -> 2.56 GMac | Improved |
Closest papers by meaning
- PaPaGei: Open Foundation Models for Optical Physiological Signals (ICLR 2025 · similarity 0.88)
- Constraint Latent Space Matters: An Anti-anomalous Waveform Transformation Solution from Photoplethysmography to Arterial Blood Pressure (AAAI 2024 · similarity 0.85)
- PhysDiff: Physiology-based Dynamicity Disentangled Diffusion Model for Remote Physiological Measurement (AAAI 2025 · similarity 0.84)
- Remote Photoplethysmography in Real-World and Extreme Lighting Scenarios (CVPR 2025 · similarity 0.84)
- RhythmMamba: Fast, Lightweight, and Accurate Remote Physiological Measurement (AAAI 2025 · similarity 0.83)
- FactorizePhys: Matrix Factorization for Multidimensional Attention in Remote Physiological Sensing (NeurIPS 2024 · similarity 0.83)
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.
