# Distribution-Free One-Pass Learning

## TL;DR

This paper introduces DFOP, a distribution-free online learning method capable of updating models with single-pass data scans, handling distribution changes without prior knowledge, and providing theoretical guarantees of convergence.

## Contribution

The paper presents DFOP, a novel distribution-free one-pass learning algorithm that adapts to changing data distributions in an online setting without needing prior distribution information.

## Key findings

- DFOP effectively handles distribution changes during data accumulation.
- Theoretical guarantees show error decreases until convergence.
- Experimental results validate DFOP's performance in regression and classification.

## Abstract

In many large-scale machine learning applications, data are accumulated with time, and thus, an appropriate model should be able to update in an online paradigm. Moreover, as the whole data volume is unknown when constructing the model, it is desired to scan each data item only once with a storage independent with the data volume. It is also noteworthy that the distribution underlying may change during the data accumulation procedure. To handle such tasks, in this paper we propose DFOP, a distribution-free one-pass learning approach. This approach works well when distribution change occurs during data accumulation, without requiring prior knowledge about the change. Every data item can be discarded once it has been scanned. Besides, theoretical guarantee shows that the estimate error, under a mild assumption, decreases until convergence with high probability. The performance of DFOP for both regression and classification are validated in experiments.