This paper explores how TinyML enables deep learning model deployment on edge devices by optimizing accuracy, inference time, and model size. Using quantization and pruning, models like VGG16, MobileNet, and custom architectures were compressed and evaluated on datasets from fashion, radiology, and dermatology. TensorFlow Lite optimization showed significant reductions in model size while maintaining acceptable accuracy and inference speed. Results highlight the trade-offs between performance metrics and underscore TinyML’s value in enabling real-time, efficient, low-power AI applications across diverse domains such as healthcare, IoT, and smart devices.

Join for free to read