Table I. TinyML frameworks. (Cotinued) Compatible Platforms Output Languages Interoperable External Libraries Publicly Available Framework Algorithms weka-porter [49] Decision trees Multiple constrained & nonconstrained platforms C Java JavaScript Weka Yes Particular developer EmbML [50] Decision trees Neural networks SVM Arduino Teensy C++ Scikit-learn Weka No Research group emlearn [51] Decision trees Neural networks Naive Gaussian Bayes Random forest AVR Atmega ESP8266 Linux C Keras Scikit-learn Yes Particular developer uTensor [52] Neural networks mBed boards C++11 TensorFlow Yes Particular developer TinyMLgen [53] Neural networks ARM Cortex-M ESP32 C TensorFlow Lite Yes Particular developer CMix-NN [54] Neural networks ARM Cortex-M C Mobilenet Yes Research group FANN-onMCU [55] Neural networks ARM Cortex-M PULP C FANN Yes Research group Sensors Inertial Weather Health ... GPS I/O Ports TinyML Main Developer Storage MC + Memory Display RAT 1 ... ... n RAT Battery APPS Figure 4. Multi-RAT smart object architecture. 12 IEEE CIRCUITS AND SYSTEMS MAGAZINE THIRD QUARTER 2020