Task Incremental Learning With Static Memory for Audio Classification Without Catastrophic Interference

The deep neural network shows excellent performance on a single task. However, deep neural networks performance degraded when trained continuously on a sequence of new tasks. This phenomenon is known as catastrophic interference. To overcome this problem, the model must be capable of learning new tasks and preserving old tasks. We introduce a single architecture model with static memory to mitigate catastrophic interference. Our proposed method overcome memory usage and model complexity. The proposed method learns the new tasks quickly without forgetting the previously learned tasks. The results show that the proposed method obtains a good tradeoff between previous and current tasks. The performance evaluation shows that the proposed method achieved better accuracy compared to other benchmarks. The proposed method achieved 92% and 95% accuracy on ESC-50 and UrbanSound8K, respectively.