Great results on training data but poor results on new data is overfitting. It usually comes from too little data, too many epochs or an oversized model.
Remedies: more and more varied data, fewer epochs, and stopping when validation loss starts rising (early stopping).
Example
Stop training when validation loss starts going up — that is where overfitting begins.
বাংলায়: ওভারফিটিং — মডেল উদাহরণগুলো মুখস্থ করে ফেলে, নতুন প্রশ্নে খারাপ করে।