Persistent activity, continuous attractor networks, and capacity limits
Working memory (WM) is maintained by persistent neural firing in prefrontal cortex (Goldman-Rakic, 1995). The continuous attractor network model (Compte et al., 2000; Wang, 2001) explains this: recurrent excitation creates localized "bumps" of activity in a ring-like network, where each position represents a remembered stimulus. Lateral inhibition stabilizes each bump and prevents the network from becoming globally active. The capacity limit (~4 items, Cowan 2001; ~3 items for precision, Luck & Vogel 1997) arises because bumps compete for inhibitory resources — too many items cause mutual interference and bump collapse, explaining why WM capacity is severely limited.