Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
SNAP-UQ: Self-supervised Next-Activation Prediction for Single-Pass Uncertainty in TinyML
Ismail Lamaakal, Chaymae Yahyati, Khalid El Makkaoui, Ibrahim Ouahbi, Yassine Maleh
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 |
|---|---|---|---|
| use SNAP-UQ instead of ensemble for latency | EE-ens · Big-MCU SpeechCmd | latency (ms): 85 -> 52 | Improved |
| use SNAP-UQ instead of deep ensemble for flash | DEEP · Big-MCU CIFAR-10 | flash (KB): 680 -> 292 | Improved |
| use SNAP-UQ instead of baseline for flash | BASE · Big-MCU CIFAR-10 | flash (KB): 280 -> 292 | Hurt |
| use SNAP-UQ instead of ensembles on small MCU | EE-ens/DEEP · Small-MCU CIFAR-10 | deployability | Improved |
| use SNAP-UQ on MNIST-C stream | BASE · MNIST-C CID stream | AUPRC: 0.54 -> 0.66 | Improved |
| use SNAP-UQ on MNIST-C stream | BASE · MNIST-C CID stream | detection delay (frames): 42 -> 24 | Improved |
| use SNAP-UQ on SpeechCmd-C stream | EE-ens · SpeechCmd-C CID stream | AUPRC: 0.59 -> 0.65 | Improved |
| use SNAP-UQ for failure detection | BASE · MNIST ID correct vs wrong | AUROC: 0.75 -> 0.90 | Improved |
| use SNAP-UQ for OOD detection | BASE · MNIST ID vs OOD | AUROC: 0.07 -> 0.86 | Improved |
| use SNAP-UQ instead of G-ODIN for OOD | G-ODIN · CIFAR-10 ID vs OOD | AUROC: 0.95 -> 0.94 | Hurt |
| use SNAP-UQ instead of temperature scaling | Temp. scaled · MNIST ID calibration | NLL: 0.242 -> 0.202 | Improved |
| use SNAP-UQ+ instead of DEEP | DEEP · CIFAR-10 ID calibration | NLL: 0.365 -> 0.363 | No clear effect |
Closest papers by meaning
- QUTE: Quantifying Uncertainty in TinyML models with Early-exit-assisted ensembles for model-monitoring (ICML 2025 · similarity 0.90)
- AtomNet: Designing Tiny Models from Operators Under Extreme MCU Constraints (AAAI 2025 · similarity 0.85)
- TinyTTA: Efficient Test-time Adaptation via Early-exit Ensembles on Edge Devices (NeurIPS 2024 · similarity 0.84)
- TinyTrain: Resource-Aware Task-Adaptive Sparse Training of DNNs at the Data-Scarce Edge (ICML 2024 · similarity 0.83)
- Robust Uncertainty Quantification Using Conformalised Monte Carlo Prediction (AAAI 2024 · similarity 0.82)
- Uncertainty Quantification with the Empirical Neural Tangent Kernel (NeurIPS 2025 · similarity 0.82)
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.
