Active Learning in Recurrent Neural Networks Facilitated by a Hebb-like Learning Rule with Memory