Research1d ago
Meta-Learning Compression For TinyML Object Detection
Adaptive Hierarchical Compression (AHC) enables continual object detection on microcontrollers with under 100KB of memory. The framework uses MAML-based gradient descent to adapt compression ratios across multi-scale features in five steps. This approach targets catastrophic forgetting in memory-constrained environments. It provides a blueprint for deploying evolving vision models on MCU hardware.