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

查看原文 / View Original 编辑 / Edit

Computer Science > Machine Learning arXiv:2610.18879 (cs) [Submitted on 1 Oct 2026] Title: Do Vision CIL Methods Transfer to Protein? A Diagnostic Benchmark of Frozen Protein Encoders Authors: Ziyuan Gao.

计算机科学 > 机器学习 arXiv:2610.18879 (cs) [提交于 2026 年 10 月 1 日] 标题：视觉类增量学习（CIL）方法能否迁移至蛋白质领域？冻结蛋白质编码器的诊断基准测试作者：Ziyuan Gao。

Abstract: Protein function annotation operates over an expanding label space, making class-incremental learning (CIL) a natural framework for incorporating newly discovered functions without full retraining. Yet existing CIL methods have been developed largely in vision and have not been systematically evaluated over frozen protein representations.

摘要：蛋白质功能注释在不断扩展的标签空间中进行，这使得类增量学习（CIL）成为一种无需完全重新训练即可整合新发现功能的自然框架。然而，现有的 CIL 方法主要是在视觉领域开发的，尚未在冻结的蛋白质表征上进行系统评估。

We conduct a controlled benchmark of nine representative CIL methods across five sequence-, structure-, and sequence-structure-based protein encoders and two enzyme function datasets at different scales, while keeping the encoders frozen and evaluation settings consistent across representations.

我们对九种代表性的 CIL 方法进行了受控基准测试，涵盖了五种基于序列、结构以及序列-结构结合的蛋白质编码器，以及两种不同规模的酶功能数据集，同时保持编码器冻结，并确保不同表征下的评估设置保持一致。

We find that no method transfers reliably across encoders, and that these failures persist across alternative phase constructions. Diagnostic analyses reveal that this instability arises from weak gradient-based stability signals, substantial heterogeneity in embedding geometry, and covariance assumptions whose appropriate capacity changes with both dataset scale and encoder.

我们发现没有任何一种方法能够在不同编码器之间可靠地迁移，且这些失败在不同的阶段构建中持续存在。诊断分析表明，这种不稳定性源于基于梯度的稳定性信号较弱、嵌入几何结构的显著异质性，以及其适用容量随数据集规模和编码器而变化的协方差假设。

These findings suggest that the representation itself should be treated as a central component of continual-learning design rather than a passive feature extractor. More generally, reliable protein CIL will require encoder-aware methods that adapt their statistical assumptions and capacity to the geometry of the representation on which they operate.

这些发现表明，表征本身应被视为持续学习设计的核心组成部分，而非被动的特征提取器。更广泛地说，可靠的蛋白质 CIL 将需要具备编码器感知能力的方法，使其能够根据所操作表征的几何结构来调整统计假设和容量。