Active Learning
Although our Text Recognition Model is trained with base-line dataset, which contains around 80 millions of auto-generated images, it is still not possible to recognize all possible Text Images out there in the world. So, if there is a situation when a text is not recognizable, what should we do? Do we have to retrain the model from scratch again?
Well, just because we come across something we don't know, it does not necessarily mean that we have to relearn everything from scratch, does it? We just need to learn a little more about what we don't know yet. For the next batch of our learning, we just need to include what we don't know yet.
And of course, it is "Active Learning" - the model will predict what it does not know, and corrected wrong predictions will be included in the next batch of training until it learns. With Active Learning, we can make our model continuously learn; however, it is a human-in-the-loop method and human's intervention and annotations are highly important.
|
အကြိတ်အဖူးအနာ | အကြိတ်.အဖု.အနာ |
|
ကလိပ္ပါလ် | ဇာလိပ္ဖိုလ် |
|
ဩလျှေလျှေက်ေပေါ် | ခြောက် ပျောက်၏ |
