in deep learning you don't have to do feature selection, this is automated by the deep learning algorithm by looking at the patterns of the data. we just feed it in and see the output classification.
Talking about automation, this is a time saver for human being isn't it? The point why computer is really helpful to us is the fact that it could automate task. Now, do we need to understand how the machine do it?
I think this is just a matter of trust or fear from humanity to machine. We don't need to understand anything, but sometimes what it matters is solving the problem.
Like a doctor giving the medicine to a patient, granted he or she will have the understanding of how the medicine works into human body. This understanding however is 'limited', the doctor kind of 'trust' the work of the people who created those medicine that it will do no harm to the body. But the truth of the matter is it does not work always that way. There is a degree of 'blackbox' in the medicine of how it will effect the body, cells, etc. So, in this case the doctor might have an 'overview' understanding of how the medicine will works but at the same-time there is a degree of 'unknown'.
So, this bring me back to the idea of deep learning being able to automate the feature selection process that normally we human do it manually or the need of having a domain expertise.
I think this is why deep learning is so powerful because the computer does what it does best, which is sequential computational processing that is no match with human being. It could see patterns, particularly the bigger the data the better it is for machine to do this ...and they have the superpower in processing big chunks of data in a sequential manner.
So, we have to make a jump to deep learning for this reason and kind of make the classical manual feature selection algorithm kind of obsolete? I don't know the answer to that question, only time will tell but my intuition said 'yes'!
I believe as of now there is still space of classical machine learning especially if the data does not involved of huge amount of data. In this case vision problems specifically would be in the realm of deep learning because of its sheer amount of data.
Talking about automation, this is a time saver for human being isn't it? The point why computer is really helpful to us is the fact that it could automate task. Now, do we need to understand how the machine do it?
I think this is just a matter of trust or fear from humanity to machine. We don't need to understand anything, but sometimes what it matters is solving the problem.
Like a doctor giving the medicine to a patient, granted he or she will have the understanding of how the medicine works into human body. This understanding however is 'limited', the doctor kind of 'trust' the work of the people who created those medicine that it will do no harm to the body. But the truth of the matter is it does not work always that way. There is a degree of 'blackbox' in the medicine of how it will effect the body, cells, etc. So, in this case the doctor might have an 'overview' understanding of how the medicine will works but at the same-time there is a degree of 'unknown'.
So, this bring me back to the idea of deep learning being able to automate the feature selection process that normally we human do it manually or the need of having a domain expertise.
I think this is why deep learning is so powerful because the computer does what it does best, which is sequential computational processing that is no match with human being. It could see patterns, particularly the bigger the data the better it is for machine to do this ...and they have the superpower in processing big chunks of data in a sequential manner.
So, we have to make a jump to deep learning for this reason and kind of make the classical manual feature selection algorithm kind of obsolete? I don't know the answer to that question, only time will tell but my intuition said 'yes'!
I believe as of now there is still space of classical machine learning especially if the data does not involved of huge amount of data. In this case vision problems specifically would be in the realm of deep learning because of its sheer amount of data.
Comments
Post a Comment