@CyberneticForests funny, but technically more accurate would be the people laying on the tracks as input (utilitarian problem) and the levers as output (utilitarian solution), the moving tram is an external constant, so could be put above the whole thing ;)

@CyberneticForests ... then again, with an inference model, the levers would be input and people on the tracks output (i.e. these are the levers people have pulled in examples, and these people have survived as a result); the machine is then run inverted and suggests, for a desired outcome, one possible configuration in which levers could be pulled.

@CyberneticForests and yet a third interpretation, completely in tune with the version here, is that the input resembles instances of unavoidable doom that awaits their victims, and all hidden layer nodes are levers (which is quite correct). this machine would likely be trained to find a configuration of levers that optimizes for the maximum of survivors, given a particular configuration of trams on rails. the network would then begin to map the train tracks themselves.

@lritter @CyberneticForests
@stablehorde_generator
draw for me a representation of a deep neural network but all the input resembles instances of unavoidable doom caused by unstoppable trolleys that awaits their victims, and all hidden layer nodes are levers. the output is optimized numbers of survivors.