> Our work so far has primarily been on general feedforward NNs. Our investigations have not yet involved much on specialized networks such as convolutionalNNs (CNNs) for image classification, recurrent NNs (RNNs) for text processing,and so on. Though we intend to adapt our ideas to these frameworks,we viewthem as separate, orthogonal issues.
This is a interesting insight. Future work is needed though. The vast majority of impressive results using NNs have been achieved by specialized networks, with special mention for resnets in vision [and Go] and transformers for NLP.
This is a interesting insight. Future work is needed though. The vast majority of impressive results using NNs have been achieved by specialized networks, with special mention for resnets in vision [and Go] and transformers for NLP.