---
format: "aidr-story-markdown/v1"
id: "10cdad4b06e67e8e782e6615d4f34c30a2231da267bbe6cb97d7ad0b2ebd2f0f"
canonical_url: "https://aidr.today/10cdad4b?lang=en"
title: "Do Vision CIL Methods Transfer to Protein? A Diagnostic Benchmark of Frozen Protein Encoders"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-10-15T04:00:00.000Z"
category: "Research"
topics: ["research"]
source_urls: ["https://arxiv.org/abs/2610.18879"]
summary: "A benchmark of nine class-incremental learning methods across five frozen protein encoders and two enzyme function datasets found that no method transfers reliably across encoders, with instability attributed to weak gradient-based stability signals, heterogeneous embedding geometry, and covariance assumptions varying with dataset scale and encoder."
---

# Do Vision CIL Methods Transfer to Protein? A Diagnostic Benchmark of Frozen Protein Encoders

> [Open the canonical story](<https://aidr.today/10cdad4b?lang=en>)

**Published:** 2026-10-15T04:00:00.000Z
**Category:** Research
**Topics:** research

## Summary

A benchmark of nine class\-incremental learning methods across five frozen protein encoders and two enzyme function datasets found that no method transfers reliably across encoders, with instability attributed to weak gradient\-based stability signals, heterogeneous embedding geometry, and covariance assumptions varying with dataset scale and encoder\.

## Sources

- [Story source](<https://arxiv.org/abs/2610.18879>)