Comment by logicchains

To clarify, is the sLSTM strictly necessary (to achieve better accuracy than those other architectures), or is the mLSTM good enough? The [1/0] model in the paper seemed to do quite well.