Name
Abstract | ||
|---|---|---|
In this paper we focus on the relationship between concept complexity and generalization error of learned concept descriptions. After introducing the concept of compressibility, we suggest how it could be usefully exploited in order to estimate from the training data the (Kolmogorov) complexity of the concept to be learned. Then, we present an empirical apparatus which allows us to study the relationship between the estimated target concept complexity and the generalization error of different learning algorithms. Results show a linear relationship between the two variates: the generalization error appears to increase as the target concept becomes more complex. While this is expected, quite interesting is the fact that the relationship seems to be (only) linear. Moreover, while the degree of correlation changes for different learners, the "linear" relationship seems not to be affected by the particular learning algorithm. |
Year | Venue | Keywords |
|---|---|---|
2004 | FRONTIERS IN ARTIFICIAL INTELLIGENCE AND APPLICATIONS | generalization error |
Field | DocType | Volume |
Training set,Compressibility,Computer science,Generalization,Correlation,Generalization error,Artificial intelligence,Machine learning | Conference | 110 |
ISSN | Citations | PageRank |
0922-6389 | 0 | 0.34 |
References | Authors | |
4 | 1 | |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Roberto Esposito | 1 | 2 | 3.79 |