In a hopeless attempt to clear some of the confusion about patterns & concepts, here is an excerpt from my recently edited part 4:
A pattern is a set of matching inputs, the same concept as fuzzy cluster in terms of unsupervised learning. In my model, match is quantified by comparison as a measure of compression, so a pattern is a compressed representation of multiple inputs. Technically, every compared input forms a pattern, but only those with an above-average compression count, - they are forwarded to higher levels for extended search. Compression is adjusted for overlap in aggregated match & miss representation: partial redundancy to previously forwarded cross-compared inputs. This adjustment increases selectivity/ sparseness of representation on a higher level. A more "exclusive" definition of a pattern is the recurrent match itself: a subset of each input shared across a set thereof. This is actually a higher-derivation pattern: an above-average match of a match. Just like above-average match selects an input for a higher-level search, above-average match of a match selects a common subset to a higher integration vs. differentiation level within a pattern itself. That subset also has a priority for extended search. The most basic hierarchical sub-differentiation within a pattern is by match of a binary sign for relative match, forming continuous segments of above | below average match across input queue. Selective elevation increases both predictive value & potential syntactic complexity of patterns: the number of different variables within it. That's because comparison of each input variable adds two new variable types: relative match (m) & miss (d) relative to same-type variable of a template pattern. Both are signed, as well as aggregated across multiple comparisons within the length of a constant sign: L(m) & L(d). Relative match determines comparison vs. aggregation for individual differences, forming a queue of ds within positive L(m). New types of derivatives are also formed by comparison across different types of S-T or derived coordinates. ..... The patterns I described here are not qualitatively different from our semantic concepts, which are either generalized empirical patterns (objects & processes), or are strictly relational. There is no other way to define a "concept". Given sufficient computational resources & discoverable mathematical shortcuts, search over incrementally complex syntax will discover patterns / concepts on & beyond the level of natural language. http://www.cognitivealgorithm.info/2012/01/cognitive-algorithm.html ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-c97d2393 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-2484a968 Powered by Listbox: http://www.listbox.com
