The training and validation accuracies are as follows: Epoch 1/5, Training Accuracy: 0.9442, Validation Accuracy: 0.7626, Time: 27.09 seconds Epoch 2/5, Training Accuracy: 0.9631, Validation Accuracy: 0.7518, Time: 28.14 seconds Epoch 3/5, Training Accuracy: 0.9757, Validation Accuracy: 0.7914, Time: 27.54 seconds Epoch 4/5, Training Accuracy: 0.9730, Validation Accuracy: 0.7698, Time: 27.30 seconds Epoch 5/5, Training Accuracy: 0.9865, Validation Accuracy: 0.7482, Time: 27.74 seconds It is a text based dataset of news headlines in English and Hindi. Size is 1680+ records. Model used is multilingual BERT with Adam optimizer.
Is the model overfitting? If so, how can we improve the model?
We were expecting a steep rise in the training accuracy and the validation accuracy curve to be close to it, not sure if this graph is optimal or not.