From Embeddings to Equations: Genetic-Programming Surrogates for Interpretable Transformer Classification

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Khorshidi, Mohammad Sadegh, Yazdanjue, Navid, Gharoun, Hassan, Nikoo, Mohammad Reza, Chen, Fang, Gandomi, Amir H.
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916970645946368
author Khorshidi, Mohammad Sadegh
Yazdanjue, Navid
Gharoun, Hassan
Nikoo, Mohammad Reza
Chen, Fang
Gandomi, Amir H.
author_facet Khorshidi, Mohammad Sadegh
Yazdanjue, Navid
Gharoun, Hassan
Nikoo, Mohammad Reza
Chen, Fang
Gandomi, Amir H.
contents We study symbolic surrogate modeling of frozen Transformer embeddings to obtain compact, auditable classifiers with calibrated probabilities. For five benchmarks (SST2G, 20NG, MNIST, CIFAR10, MSC17), embeddings from ModernBERT, DINOv2, and SigLIP are partitioned on the training set into disjoint, information-preserving views via semantic-preserving feature partitioning (SPFP). A cooperative multi-population genetic program (MEGP) then learns additive, closed-form logit programs over these views. Across 30 runs per dataset we report F1, AUC, log-loss, Brier, expected calibration error (ECE), and symbolic complexity; a canonical model is chosen by a one-standard-error rule on validation F1 with a parsimony tie-break. Temperature scaling fitted on validation yields substantial ECE reductions on test. The resulting surrogates achieve strong discrimination (up to F1 around 0.99 on MNIST, CIFAR10, MSC17; around 0.95 on SST2G), while 20NG remains most challenging. We provide reliability diagrams, dimension usage and overlap statistics, contribution-based importances, and global effect profiles (PDP and ALE), demonstrating faithful, cross-modal explanations grounded in explicit programs.
format Preprint
id arxiv_https___arxiv_org_abs_2509_21341
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle From Embeddings to Equations: Genetic-Programming Surrogates for Interpretable Transformer Classification
Khorshidi, Mohammad Sadegh
Yazdanjue, Navid
Gharoun, Hassan
Nikoo, Mohammad Reza
Chen, Fang
Gandomi, Amir H.
Neural and Evolutionary Computing
Artificial Intelligence
Machine Learning
68T07
We study symbolic surrogate modeling of frozen Transformer embeddings to obtain compact, auditable classifiers with calibrated probabilities. For five benchmarks (SST2G, 20NG, MNIST, CIFAR10, MSC17), embeddings from ModernBERT, DINOv2, and SigLIP are partitioned on the training set into disjoint, information-preserving views via semantic-preserving feature partitioning (SPFP). A cooperative multi-population genetic program (MEGP) then learns additive, closed-form logit programs over these views. Across 30 runs per dataset we report F1, AUC, log-loss, Brier, expected calibration error (ECE), and symbolic complexity; a canonical model is chosen by a one-standard-error rule on validation F1 with a parsimony tie-break. Temperature scaling fitted on validation yields substantial ECE reductions on test. The resulting surrogates achieve strong discrimination (up to F1 around 0.99 on MNIST, CIFAR10, MSC17; around 0.95 on SST2G), while 20NG remains most challenging. We provide reliability diagrams, dimension usage and overlap statistics, contribution-based importances, and global effect profiles (PDP and ALE), demonstrating faithful, cross-modal explanations grounded in explicit programs.
title From Embeddings to Equations: Genetic-Programming Surrogates for Interpretable Transformer Classification
topic Neural and Evolutionary Computing
Artificial Intelligence
Machine Learning
68T07
url https://arxiv.org/abs/2509.21341