In my biased, inexperienced view, most deep learning research is related to either:
a) Deep learning applications (adapting to different domains, solving a particular task)
or
b) Deep learning architectures (ie. 'macro').
My question is: are there people researching unit-level fundamentals? Eg. modifying the current "standard" neuron, like using alternative activation functions, or doing something more complex that just adding inputs (multiplied by their weights) and applying an activation function? (ie. 'micro' architecture)
At least I haven't been able to find new research that involves modifying the classic unit of deep learning models, probably because I'm using the wrong keywords.
