Optimally Combining Classifiers Using Unlabeled Data

03/05/2015
by   Akshay Balsubramani, et al.
0

We develop a worst-case analysis of aggregation of classifier ensembles for binary classification. The task of predicting to minimize error is formulated as a game played over a given set of unlabeled data (a transductive setting), where prior label information is encoded as constraints on the game. The minimax solution of this game identifies cases where a weighted combination of the classifiers can perform significantly better than any single classifier.

READ FULL TEXT

page 1

page 2

page 3

page 4

research
05/19/2017

Estimating Accuracy from Unlabeled Data: A Probabilistic Logic Approach

We propose an efficient method to estimate the accuracy of classifiers u...
research
10/02/2012

Nonparametric Unsupervised Classification

Unsupervised classification methods learn a discriminative classifier fr...
research
02/25/2016

Learning to Abstain from Binary Prediction

A binary classifier capable of abstaining from making a label prediction...
research
06/02/2023

Streaming algorithms for evaluating noisy judges on unlabeled data – binary classification

The evaluation of noisy binary classifiers on unlabeled data is treated ...
research
08/26/2022

Confusion Matrices and Accuracy Statistics for Binary Classifiers Using Unlabeled Data: The Diagnostic Test Approach

Medical researchers have solved the problem of estimating the sensitivit...
research
12/12/2012

Continuation Methods for Mixing Heterogenous Sources

A number of modern learning tasks involve estimation from heterogeneous ...
research
06/15/2023

Non-Asymptotic Performance of Social Machine Learning Under Limited Data

This paper studies the probability of error associated with the social m...

Please sign up or login with your details

Forgot password? Click here to reset