Retraining affecting previous knowledge

[Parvez-Khan-1]

Hello Franck, I am successfully able to re-train the Neuro-NER model but one weird thing is happening.When i retrained a previous model , then on whatever data i retrained it , it giving correct result but it giving wrong result on its previous knowledge(Whatever it learned before re-training)

Example : We have xyz model which detect fruits correctly , then i re-trained it on some cars so after re-training it detecting cars correctly but not fruits.

Please help i am stuck at this point.

[Franck-Dernoncourt]

That's pretty much expected. You could try training on both cars and fruits to mitigate this issue. Or use one model for fruits, and one model for cars.

[Parvez-Khan-1]

thanks for replying , but consider the following scenario : If i train model on cars only , and its detecting some of the cars incorrectly then i re-trained it again by correcting those wrongly detected cars.so now after re-training it affecting some of the previous correct detection.

[InternetMedical]

@Parvez-Khan-1 Hello, Khan. Nice to hear that you manage to re-train the Neuro-NER model on new dataset. You said 'When i retrained a previous model'. Do you mean that you fine-tuning the pre-trained model on a new dataset? Hope for your answer.