Applying Deep Learning to Classify Massive Amounts of Text Using Convolutional Neural Systems

Preview this chapter:

Applying Deep Learning to Classify Massive Amounts of Text Using Convolutional Neural Systems, Page 1 of 1

< Previous page | Next page > /docserver/preview/fulltext/9789815305395/chapter-18-1.gif

Supervised learning based on deep learning is often used for mass-scale picture categorization. However, it takes a lot of computing effort and energy to retrain these vast networks to accept new, unknown data. When retraining, it is possible that training samples used before would not be accessible. We present a scalable, gradually expanding CNN that can learn new jobs while reusing some of the base networks and an efficient training mechanism. Our approach takes cues from transfer learning methods, but unlike other approaches, it retains knowledge of previously mastered tasks. Convolutional layers from the early section of the base network are reused in the updated network, and a few more convolutional kernels are added to the later layers to facilitate learning a new set of classes. On the task of categorising texts, we tested the suggested method. Our method achieves comparable classification accuracy to the standard incremental learning method in which networks are updated solely with new training samples, without any network sharing), while also being more resourcefriendly and taking less time and space to train.