The difference between online learning and offline learning

Online Learning And Offline Learning

The first understanding

Online learning

Online learning , Usually input one piece of data at a time （ Instead of a batch）, After training, update the weight directly . Process the data sequentially one by one , But the weight will be updated directly after each data training , But I don't know whether it's right or wrong , If a weight update is wrong , After that, the weight update may always be wrong , The resulting model may gradually go in the wrong direction , Residual error appears . Online learning first processes data in sequence , They produce a model , And put this model into practice , There is no need to provide a complete training data set at the beginning . As more real-time data enters the model , The model will be constantly updated during operation

Offline learning

Offline learning , Similar to batch learning , Suppose the whole sample has m Data , Offline training will train m The number of integral multiples of , Then bring in the next , Run the whole sample , At this time, the error rate may not satisfy you , Do the above operation again for the whole sample , Until the error is very small . Offline learning is a batch Update the weight after training , Therefore, all data must be in every training operation （batch） It's all available in , In this way, the network will not be brought to the extreme because of the update error of individual data .

The second understanding

In online learning , On the contrary , Online algorithms process data sequentially . They produce a model , And put this model into practice , There is no need to provide a complete training data set at the beginning . As more real-time data arrives , The model will be constantly updated during operation

In offline learning , All training data must be available during model training . Only after training , Models can be used . In short , Train first , Reuse model , Don't use the model until you finish training .

Simply speaking

Online learning ： One sample is trained , Update weights directly
Offline learning ： All samples are trained , Update the weight