Pruned Adaptation Modules for Pre-Trained Model based Class-Incremental Learning

📄 Abstract

The continual learning literature has rapidly shifted from traditional class incremental learning (CIL) techniques to foundation model (FM)-based CIL methods without a clear understanding of how these newer approaches compare to strong, lightweight convolutional baselines. This abrupt transition has created a substantial methodological gap, making it difficult to assess whether recent FM-based CIL progress reflects genuine advances or merely the absence of rigorous baselines. To address this gap, we introduce Pruned Adaptation Modules (PAM), a simple yet effective method that freezes the vast majority of the pre-trained ResNet while enabling scalable continual adaptation through sparse task-specific layers. PAM yields up to a ~5×reduction in trainable parameters and a ~6×reduction in total parameters, significantly reducing the cost of continual updates. Across diverse benchmarks, PAM consistently mitigates catastrophic forgetting and outperforms state-of-the-art FM-based CIL approaches. Our findings position PAM as a strong and transparent baseline that helps bridge the gap between traditional and FM-based CIL, guiding future research for a more accurate assessment of true progress in continual adaptation.

---

🚀 How to Run the Code

All experiments can be run using the main.py script. Below is a summary of the key configuration option. Please make sure to change them within the main.py file to run in the setup you wish.

📌 Main Configurations in main.py

Argument	Description	Example Values
dataset_name	Dataset to be used for class-incremental learning	'cifar100', 'imagenetr', 'cars', 'cub'
increment	Number of new classes introduced per task	5,10, 20, etc.
sparsity	Percentage of weights to prune in the PAM module (e.g., 0.96 means 96% sparsity)	0.95, 0.96, 0.97
model	Backbone model variant (pre-trained ResNet architecture)	'resnet18', 'resnet50', 'resnet101', 'resnet152'

📂 Dataset Requirements

• For ImageNet-R, Stanford Cars, and CUB-200, please upload the datasets into the data/ directory.

🧪 Environment Setup

All experiments were conducted using the environment defined in pam.yml.

To ensure reproducibility, please create and activate the conda environment using the provided pam.yml file:

conda env create -f sam.yml
conda activate sam