ABSTRACT
Connectionist models have been around for half a century. Their ability to learn from examples makes them prime candidates as models of human memory. Here, we will review how connectionist models learn and forget. Forgetting can mean many things, from short-term forgetting at the scale of seconds or shorter to very long-term forgetting over several decades. It can be viewed as a side effect of diffuse noise and decay at the level of synapses or as an active process, perhaps to safeguard important memories or to extract high-level abstractions from our daily experiences. Connectionist models can accommodate all of these mechanisms, some of which lie at the heart of the principles by which they work.
