Fun coincidence, I read this paper as part of my AI research just recently. It's actually quite neat! Mostly because there is really not much like it in the sen

Fun coincidence, I read this paper as part of my AI research just recently. It's actually quite neat! Mostly because there is really not much like it in the sense that it focuses on exploring a memory mechanism for RL agents - most of the time people just stack frames or slap on an RNN and call it a day. The set of tasks evaluated on is cool too.