While neural networks are in many ways inspired by the way the human brain functions and learns, in particular human neurons, as machine learning technologies develop they are increasingly moving away from neuron functions and into a field of their own. Neural networks such as Artificial Neural Networks (ANNs) were originally inspired by the way neurons function, mimicking the way neurons receive a connection, make a decision or decide a function, and make an outgoing connection to the next neuron, or in the case of an ANN, the next layer of the network. Neural networks have also been developed to mimic the plasticity of the human brain, programmed to develop the ability to modify and strengthen certain models based on data received just as the human brain strengthens certain connections and recognizes patterns between data.
However, despite the similarity in structure, human neurons remain much more complex than neural networks and have the ability to adapt and complete different tasks, as well as perform nonlinear calculations and change the speed and transmissions of signals based on different factors. Neural networks remain linear systems, where information is passed through networks and these routes can’t be bypassed to perform these nonlinear, adaptive functions. Machine learning so far is dependent on supervised learning, where pairs of data are fed to a program to teach it to recognize patterns and match data, while most of human learning involves unsupervised and reinforcement learning, or learning through experience. This level of learning is as of yet inaccessible to machine learning programs, which still rely on human checks and supervision to ensure that the programs are matching the correct data and producing useful results.
While the human brain is the ultimate inspiration for machine learning programs, new programs like GAN and RNN are building off of different principles and not necessarily modeled after the human brain. The future for machine learning will likely diverge from neuron-type modeling, but the goal – to create programs that can learn, think, and produce conclusions like human neurons – remains the same.
Sources:
https://medium.com/swlh/do-neural-networks-really-work-like-neurons-667859dbfb4f