Neural Network Inference when only 1 input changes

Edward Hamilton asked 2 weeks ago

I’m interested in using a neural network in the objective function of local search based combinatorial optimization (e.g. scheduling). During optimization, often only 1 (or a few) decision variables (inputs to NN) change and we want to know the resultant objective function. So my question is, are there NN inference engines that have memory and if only 1 input changes would incrementally infer the output?

Your Answer

13 + 16 =