Genetics-Based Learning and Statistical Generalization

Wah, Benjamin W.; Ieumwananonthachai, Arthur; Yu, Tong

doi:10.1142/9789812819918_0011

Cited by 1 publication

(2 citation statements)

References 7 publications

(8 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Probabilities of mean have been used to evaluate generalizability in various genetics-based learning and generalization experiments [39], [15], [34], [24], [41], [40], [16], [38]. These include the learning of load balancing strategies in distributed systems and multicomputers, the tuning of parameters in VLSI cell placement and routing, the tuning of fitness functions in genetics-based VLSI circuit testing, the automated design of feedforward neural networks, the design of heuristics in branch-and-bound search, range estimation in stereo vision, and the learning of parameters for blind equalization in signal processing.…”

Section: ) No Hypothesis Is Better Than H 0 In All Subdomainsmentioning

confidence: 99%

“…Hence, when hypothesis are ordered using their probabilities of win and performance is normalized by any method in which the baseline can be changed, anomalies in performance ordering may still happen. As illustrated in [41], this phenomenon happens because not only the mean but the variance of the baseline are important in determining the ordering of the hypotheses. The variance of the performance values places another degree of freedom in the performance ordering, which can change the ordering when the variance changes.…”

mentioning

confidence: 99%

See 1 more Smart Citation

Generalization and generalizability measures

Wah

1999

IEEE Trans. Knowl. Data Eng.

View full text Add to dashboard Cite

In this paper, we define the generalization problem, summarize various approaches in generalization, identify the credit assignment problem, and present the problem and some solutions in measuring generalizability. We discuss anomalies in the ordering of hypotheses in a subdomain when performance is normalized and averaged, and show conditions under which anomalies can be eliminated. To generalize performance across subdomains, we present a measure called probability of win that measures the probability whether one hypothesis is better than another. Finally, we discuss some limitations in using probabilities of win and illustrate their application in finding new parameter values for TimberWolf, a package for VLSI cell placement and routing. Index Terms-Anomalies in generalization, credit assignment problem generalization, machine learning, subdomains, probability of win, VLSI cell placement and routing.

show abstract