A Review of Recent Results on Learning with Finite Memory.

摘要

It is seen that the theory is very general when applied to randomized algorithms and infinite sample sizes. Although deterministic algorithms are not as easily analyzed, their practical importance tempts us to exert additional effort. Similarly, finite sample size problems lack theoretical simplicity but not practical applications. Future work will thus probably continue to pursue these two lines of research. In addition it may be possible to take account of the complexity of the functions f and d. As noted it is surprising that with no constraint on complexity the optimal f and d are rather simple for the infinite sample, randomized algorithm problem. It is therefore possible that even when a suitable such constraint is imposed, there will be little change in the optimal algorithm. (Author)