You, me or it: Nobody is Perfect!
“There is nothing known as “Perfect“. It’s only those imperfections which we choose not to see!!” These words, by the great Albert Einstein aka ‘the Genius‘ himself, sums up today’s article. Deep learning is doing pretty good nowadays & I’m not an antagonist here, but truth to be told, there lies some so called “imperfections” in our beloved deep learning as well. And I’m going to highlight some of those today.
- When we think about AI, the first thing which comes to our mind is “data”. Now, the blame on deep learning here is the “amount of data” it needs, it’s not always possible to get most out of a deep neural net model by supplying a limited quota of data. There must be a good amount of data to fulfill the insatiable needs of neural nets, as mapping of real-life non-linear problems in form of weights is not a straight-forward task.
- Next thing that might hinder a deep learning project is the computing power it needs. And it is also one of the reasons behind the delay in deep learning’s arrival in mainstream.
- And even though it is inspired from the working of a human brain, but it still lacks the reasoning ability which a human possesses.
- Another complaint about deep learning is that, it only learns the complex relationship between the inputs & outputs, but the inherent representation is not well understood. One famous example is on this link: https://gizmodo.com/british-cops-want-to-use-ai-to-spot-porn-but-it-keeps-m-1821384511/amp
- And as the deep learning models are all black-boxes, we can’t understand much except the input & output layers in a model. The hidden layers are all clusters of weight values which can’t be perceived by humans easily.
By: Amit Bishnoi