Correcting Class Imbalance in Prior-Data Fitted Networks for Tabular Classification

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: McDowell, Samuel, Stromberg, Nathan, Sankar, Lalitha
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911703474634752
author McDowell, Samuel
Stromberg, Nathan
Sankar, Lalitha
author_facet McDowell, Samuel
Stromberg, Nathan
Sankar, Lalitha
contents Prior-data fitted networks (PFNs) have achieved exceptional performance on tabular classification tasks. However, like other classifiers, their performance can suffer under the effect of class imbalance, resulting in poor performance for rare classes. Several techniques exist which attempt to mitigate the deleterious effect of class imbalance on classification performance, but the in-context learning (ICL) dynamic of PFNs means that loss-based strategies are impossible, and other techniques are unproven. We have adapted several classical techniques addressing class imbalance and analyzed their performance on PFN classification. We observe that thresholding performs exceptionally well because of the calibration characteristics of PFNs, and downsampling performs comparably because of PFNs exceptional limited-data performance, with the additional benefit of reduced computation cost for inference.
format Preprint
id arxiv_https___arxiv_org_abs_2605_21742
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Correcting Class Imbalance in Prior-Data Fitted Networks for Tabular Classification
McDowell, Samuel
Stromberg, Nathan
Sankar, Lalitha
Machine Learning
Information Theory
Prior-data fitted networks (PFNs) have achieved exceptional performance on tabular classification tasks. However, like other classifiers, their performance can suffer under the effect of class imbalance, resulting in poor performance for rare classes. Several techniques exist which attempt to mitigate the deleterious effect of class imbalance on classification performance, but the in-context learning (ICL) dynamic of PFNs means that loss-based strategies are impossible, and other techniques are unproven. We have adapted several classical techniques addressing class imbalance and analyzed their performance on PFN classification. We observe that thresholding performs exceptionally well because of the calibration characteristics of PFNs, and downsampling performs comparably because of PFNs exceptional limited-data performance, with the additional benefit of reduced computation cost for inference.
title Correcting Class Imbalance in Prior-Data Fitted Networks for Tabular Classification
topic Machine Learning
Information Theory
url https://arxiv.org/abs/2605.21742