Reproduce “Real-Time Performance Benchmarking of TinyML Models in Embedded Systems (PICO: Performance of Inference, CPU, and Operations)” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Abhishek Dey, Saurabh Srivastava, Gaurav Singh, Robert G. Pettit
Researchers at George Mason University introduced PICO-TINYML-BENCHMARK, a framework for evaluating real-time TinyML model performance on embedded systems. This work demonstrates that Raspberry Pi 4 offers lower inference latency and resource consumption compared to BeagleBone AI64 for CPU-only execution, while maintaining consistent prediction stability across various models.
Key insights
- The Raspberry Pi 4 consistently delivers faster CPU-only inference speeds and more efficient resource utilization (lower CPU and memory consumption) compared to the BeagleBone AI64 for TinyML workloads.
- Despite performance differences, both embedded platforms demonstrated high and consistent prediction stability, affirming their reliability for real-world TinyML deployments.
- Comprehensive, real-time benchmarking that includes system-level metrics beyond just accuracy is essential for making informed decisions on TinyML model optimization and hardware platform selection.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.