Epstein Secrets

have come to the following conclusions: • for relatively straightforward procedure learning problems, hilldimbing with random restart and a strong Occam bias is an effective method • for more difficult problems that elude hinclimbing, probabilistic evolutionary program learn- Mg is an effective metho

r the latter are learned via MOSES or via some other means. The textual realization of Combo resembles LISP with less syntactic sugar. Internally a Combo program is represented as a program tree. • Composer: In the PLN design, a rule is denoted a composer if it needs premises for generating its conse

ve its fingers to pick up a cup, coordinating its movements with its visual perceptions. It's certainly possible to represent this sort of thing in Combo, but Combo may be an awkward tool. However, if one represents low-level procedures like this using another method, e.g. learned cell assemblies in a

the question of programs in what programming language, and in this context we have made a fairly traditional choice, using a special language called Combo that is essentially a minor variant of LISP, and supplying Combo with a set of customized primitives intended to reduce the length of the typical p

is learned faster (9s67 instead of 21s47 for co:Q. Conversely the results for lamb show that when action sequence building-block is enabled, if the Occam's razor is too weak it can dramatically slow down the search. That is because in this circumstance the search is misled by longer candidates that fi

tricks, fetch_ball and triple_kick to be described below. in order to calibrate the size penalty function: 1. fetch_ball, which corresponds to the Combo program and_seg (goto (ball) grab (ball) goto (owner) drop) 2. triple_kick, if the stick is near the ball then kick 3 times with the left leg an

n x, two possibilities exist: EFTA00624385 33.6 Hierarchical Program Learning 239 • There is already a model for PredicateNodes at position x in Combo trees in the given population, in which case a population of PredicateNodes potentially living at that position is drawn from the known model, and

re learning" - it's not actually a procedure learning process, but rather a process that utilizes the fruits of procedure learning. The well-known "Occam's razor" heuristic says that all else being equal, simpler is better. This notion is embodied mathematically in the Solomonoff-Levin "universal prio

uesses produced by pure procedure learning methods), it may still be valuable to have this procedure in compact and rapidly executable form such as Combo provides. To proceed with the technical description of predicate schematization in CogPrime. we first need the notion of an "executable predicate".

tand: all else equal, a predicate embodied in a small Combo tree is better than a predicate embodied in a big one. There is some work hidden here in Combo tree reduction; ideally, one would like to find the smallest representation of a given Combo tree, but this is a computationally formidable problem