3 ms·
It seems like a structural problem. Deep learning generally performs better than anything else that is well known but it also has well known limitations and in
by ilaksh 7y ago
It seems like a structural problem. Deep learning generally performs better than anything else that is well known but it also has well known limitations and inefficiencies.
People should question all of the assumptions, from the idea of using NNs and the particular type of NN and all of the core parts of the belief system. Because these certain aspects are fixed on faith more than anything else.
If you want efficiency of training, adaptability, online, generality, true understanding, those assumptions might need to go. Which would not mean you could learn from DL systems, just that core structures would not be fixed.