Core Concepts
Permutation invariance simplifies complex ML problems.
Abstract
Permutation invariance simplifies ML problems by making functions invariant to input order.
Research focuses on testing permutation invariance and reducing dimensions.
Applications include anomaly detection, text retrieval, and classification.
Statistical tests and kernel ridge regression are based on sorting and averaging tricks.
Permutation invariance is relevant in various fields like health sciences and finance.
Exchangeability and permutation invariance are related concepts in probability distributions.
Methods simplify exploitation of permutation invariance in ML architectures.
Permutation invariant functions have smaller complexity and covering numbers.
Kernel ridge regression efficiently recovers permutation invariant functions.
Stats
"Our test statistics take the form T := supt∈[0,1]d √n eFn(t) − Fn(t), where Fn(t) is the empirical CDF at t, eFn(t) = Fn(sort t), and n is the size of the random sample."
"We propose a kernel density estimator (KDE) that averages over a carefully constructed subset of permutations."
"The logarithm of the covering number for the permutation invariant H¨older class with a boundary condition is reduced by a factor of d!."
Quotes
"Permutation invariance simplifies the exploitation of complex problems in machine learning."
"Our methods for testing permutation invariance and kernel ridge regression are based on sorting and averaging tricks."