So how do we get from today's neural networks to true AI?

[sujonkumar6300]

Software
In this context, software refers to neural network architectures trained with an optimization algorithm to solve a specific task.

Neural networks are currently the de-facto tool for learning to solve tasks that involve supervised learning to categorize a large data set. But this is not artificial intelligence. True artificial intelligence requires acting in the real world, learning austria consumer email list unsupervised from new experiences and combining this with knowledge previously acquired in different circumstances, with the goal of solving the current challenge.


Neural network architectures. It is true that neural networks have the advantage of being able to learn the ideal parameters for an algorithm automatically from data, which gives them an advantage over manual configuration strategies. However, what is not mentioned is that the architecture of that neural network, which is the basis of training to solve a specific task, cannot be generated automatically from data but must be created by hand; this is one of the major limitations of this field. There is already some research to solve this problem [1], but we need more. The problem with learning neural network architectures from data is that it currently takes too much time to experiment with different architectures on a large data set. We need to overcome this limitation.