Dataset Information

Risk Minimization from Adaptively Collected Data: Guarantees for Supervised and Policy Learning.

ABSTRACT: Empirical risk minimization (ERM) is the workhorse of machine learning, whether for classification and regression or for off-policy policy learning, but its model-agnostic guarantees can fail when we use adaptively collected data, such as the result of running a contextual bandit algorithm. We study a generic importance sampling weighted ERM algorithm for using adaptively collected data to minimize the average of a loss function over a hypothesis class and provide first-of-their-kind generalization guarantees and fast convergence rates. Our results are based on a new maximal inequality that carefully leverages the importance sampling structure to obtain rates with the good dependence on the exploration rate in the data. For regression, we provide fast rates that leverage the strong convexity of squared-error loss. For policy learning, we provide regret guarantees that close an open gap in the existing literature whenever exploration decays to zero, as is the case for bandit-collected data. An empirical investigation validates our theory.

SUBMITTER: Bibaut A

PROVIDER: S-EPMC9799962 | biostudies-literature | 2021 Dec

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Risk Minimization from Adaptively Collected Data: Guarantees for Supervised and Policy Learning.

Bibaut Aurélien A Kallus Nathan N Dimakopoulou Maria M Chambaz Antoine A van der Laan Mark M

Advances in neural information processing systems 20211201

Empirical risk minimization (ERM) is the workhorse of machine learning, whether for classification and regression or for off-policy policy learning, but its model-agnostic guarantees can fail when we use adaptively collected data, such as the result of running a contextual bandit algorithm. We study a generic importance sampling weighted ERM algorithm for using adaptively collected data to minimize the average of a loss function over a hypothesis class and provide first-of-their-kind generalizat ...[more]

PMID: 36590675

Similar Datasets

Project description:Proper expression of the genes plays a vital role in the function of an organism. Recent advancements in DNA microarray technology allow for monitoring the expression level of thousands of genes. One of the important tasks in this context is to understand the underlying mechanisms of gene regulation. Recently, researchers have focused on identifying local DNA elements, or motifs to infer the relation between the expression and the nucleotide sequence of the gene. This study proposes a novel data adaptive representation approach for supervised learning to predict the response associated with the biological sequences. Biological sequences such as DNA and protein are a class of categorical sequences. In machine learning, categorical sequences are generally mapped to a lower dimensional representation for learning tasks to avoid problems with high dimensionality. The proposed method, namely SW-RF (sliding window-random forest), is a feature-based approach requiring two main steps to learn a representation for categorical sequences. In the first step, each sequence is represented by overlapping subsequences of constant length. Then a tree-based learner on this representation is trained to obtain a bag-of-words like representation which is the frequency of subsequences on the terminal nodes of the tree for each sequence. After representation learning, any classifier can be trained on the learned representation. A lasso logistic regression is trained on the learned representation to facilitate the identification of important patterns for the classification task. Our experiments show that proposed approach provides significantly better results in terms of accuracy on both synthetic data and DNA promoter sequence data. Moreover, a common problem for microarray datasets, namely missing values, is handled efficiently by the tree learners in SW-RF. Although the focus of this paper is on biological sequences, SW-RF is flexible in handling any categorical sequence data from different applications.

Dataset Information

Risk Minimization from Adaptively Collected Data: Guarantees for Supervised and Policy Learning.

Publications

Risk Minimization from Adaptively Collected Data: Guarantees for Supervised and Policy Learning.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets