Controlling State-Space Explosion in Chunk Learning

Varying combinations of perceptual cues may be relevant for learning and action-selection. However, storing all possible cue combinations in memory is computationally implausible in sufficiently complex environments due to a state-space explosion. Some psychological models suggest that cue combinations, i.e. chunks, should be generated at a conservative rate (EPAM/CHREST; e.g. Feigenbaum & Simon, 1984). Other models suggest that chunk retrieval is based on statistical regularities in the environment (i.e. recency and frequency; Anderson & Schooler, 1991). We present a computational model of chunk generation based on these two principles, and demonstrate how combining these principles alleviates state-space explosion, producing great savings in memory while maintaining a high level of performance.