Release – November 2021
Automaton AI has added a new feature in ADVIT where you can create and train models within the ADVIT platform itself. You can add the model’s name, model version, select appropriate model architecture, select the Deep Learning framework of your choice, set hyperparameters & parameters, and finally, select the dataset and categories you want to train your model on. Now, instead of only being able to auto annotate your dataset and export it to feed to your model, you can also configure and train the DNN models in ADVIT with just a few clicks
User has the access to create different versions of a specific model, that can be used to train on different categories. It will help data scientists to keep track of different versions of the specific model and can also differentiate the trained data. Various model versions can also be maintained to check the accuracy of a particular dataset. Users can set different frameworks, parameters & hyperparameter values to differentiate the model accuracy over several datasets.
Real-Time Training Statistics:
This feature in ADVIT allows data scientists to monitor the real-time status of the model training. A few of the parameters which are monitored are learning rate, regression loss, classification loss, learning rate Decay, mAP (if applicable), etc. Based on the statistics, a data scientist can take a run time call whether the training is to be continued or stopped and make necessary changes in parameters and hyperparameters in the next model training version.
Button Click Model Evaluation:
Once the model is trained, data scientists can evaluate the model by using the weights stored for multiple epochs or with the latest/best weights. The results for evaluation are then downloaded automatically for ease of visualization. This can help the data scientist for a quick glimpse of how the model is performing on unseen datasets.