The Solution for the sequential task continual learning track of the 2nd Greater Bay Area International Algorithm Competition

TL;DR

This paper introduces a data-free, parameter-isolation continual learning algorithm that effectively addresses catastrophic forgetting in sequential tasks, winning second place in a competitive international algorithm challenge.

Contribution

The method uniquely employs parameter isolation, freezing batch normalization, and adaptive strategies without expanding the network or using external data, for both task and domain incremental learning.

Findings

Achieved second place in the competition.

Effectively mitigated catastrophic forgetting.

Reduced storage and inference time through mask compression.

Abstract

This paper presents a data-free, parameter-isolation-based continual learning algorithm we developed for the sequential task continual learning track of the 2nd Greater Bay Area International Algorithm Competition. The method learns an independent parameter subspace for each task within the network's convolutional and linear layers and freezes the batch normalization layers after the first task. Specifically, for domain incremental setting where all domains share a classification head, we freeze the shared classification head after first task is completed, effectively solving the issue of catastrophic forgetting. Additionally, facing the challenge of domain incremental settings without providing a task identity, we designed an inference task identity strategy, selecting an appropriate mask matrix for each sample. Furthermore, we introduced a gradient supplementation strategy to enhance the importance of unselected parameters for the current task, facilitating learning for new tasks. We also implemented an adaptive importance scoring strategy that dynamically adjusts the amount of parameters to optimize single-task performance while reducing parameter usage. Moreover, considering the limitations of storage space and inference time, we designed a mask matrix compression strategy to save storage space and improve the speed of encryption and decryption of the mask matrix. Our approach does not require expanding the core network or using external auxiliary networks or data, and performs well under both task incremental and domain incremental settings. This solution ultimately won a second-place prize in the competition.