Dan and I weren't too familiar with Elm when we first started working on Redux. We had read the Elm Architecture document, but for whatever reason (perhaps our unfamiliarity with its ML syntax) neither of us really understood it. It was only later, after coming up with "reducers" and reducer composition independently, did we realize how similar Redux is to Elm.
The biggest difference is that the Elm Architecture is fractal. An Elm model's structure mirrors the structure of the UI, via composition. In other words, most Elm update functions (reducers, in Redux parlance) have a corresponding view.
Redux also uses reducer composition, but its composition does not mirror the composition of React components. In fact, we typically recommend that Redux reducers do not correspond directly to components. Those types of reducers are usually best implemented using local state.