Currently, OutlierDetectionNetworks.jl builds directly upon Flux which is a low-level interface and provides little integration with MLJ. We should evaluate using MLJFlux as a higher-level backend.
Additionally, there are very interesting ideas in FastAI, on how to provide a higher-level interface to NN-based algorithms.
Currently,
OutlierDetectionNetworks.jl
builds directly uponFlux
which is a low-level interface and provides little integration with MLJ. We should evaluate using MLJFlux as a higher-level backend.Additionally, there are very interesting ideas in FastAI, on how to provide a higher-level interface to NN-based algorithms.