DNNs can product advanced non-linear associations. DNN architectures produce compositional models where the object is expressed as being a layered composition of primitives.[142] The extra layers allow composition of functions from reduced layers, probably modeling elaborate knowledge with fewer models than the usual in the same way performing shallow network. https://deep-learning-ai54296.livebloggs.com/32401946/5-simple-techniques-for-deep-learning-ai