Table 3.
Accuracy of different TinyML models and their software tools.
| Model | Software tool | Accuracy | |
|---|---|---|---|
| ProtoNN | EdgeML | 93.58% | |
| CNN+GRU | CMIS-NN | 85.4% | |
| TCN | NEMO/DORY | 93.8% | |
| TCN | TFLite+GAPFlow | 94.0% | |
| TCN | TFlite | 94.0% | |
| TCN | CUBEAI+TFLite | 94.0% | |
| RF | NA | 94.5% | |
| Bonsal | EdgeML | 94.2% | |
| DNN | Gestures dataset | TFLite | 99% |
| DNN | Mnist dataset | TFLite | 99% |
|
SVM, logistic regression, decision tree, random forest |
Gestures dataset | TFLite | 95% |
|
SVM, logistic regression, decision tree, random forest |
Mnist dataset | TFLite | 90.3% |
| CNN-LSTM-DNN | NA | 93.5% | |