Title :
The importance of convexity in learning with squared loss
Author :
Sun Lee, Wee ; Bartlett, Peter L. ; Williamson, Robert C.
Author_Institution :
Sch. of Electr. Eng., Univ. of New South Wales, Canberra, ACT, Australia
fDate :
9/1/1998 12:00:00 AM
Abstract :
We show that if the closure of a function class F under the metric induced by some probability distribution is not convex, then the sample complexity for agnostically learning F with squared loss (using only hypotheses in F) is Ω(ln(1/δ)/ε2) where 1-δ is the probability of success and ε is the required accuracy. In comparison, if the class F is convex and has finite pseudodimension, then the sample complexity is O(1/ε(ln(1/ε)+ln(1/b)). If a nonconvex class F has finite pseudodimension, then the sample complexity for agnostically learning the closure of the convex hull of F, is O(1/ε(1/ε(ln(1/ε)+ln(1/δ)). Hence, for agnostic learning, learning the convex hull provides better approximation capabilities with little sample complexity penalty
Keywords :
computational complexity; learning (artificial intelligence); probability; accuracy; agnostic learning; approximation; convex hull; convexity; finite pseudodimension; function closure; hypotheses; nonconvex class; probability distribution; sample complexity penalty; squared loss; success probability; Artificial neural networks; Australia Council; Computational modeling; Computer networks; Probability distribution; Sun; Systems engineering and theory;
Journal_Title :
Information Theory, IEEE Transactions on