An Empirical and Theoretical Investigation into the Effects of Catastrophic Forgetting in Continual Learning Models with Non-Stationary Data Streams and Evolving Task Distributions
Keywords:
Continual learning, catastrophic forgetting, non-stationary data, evolving task distributions, neural networks, memory-based methods, machine learningAbstract
Continual learning models aim to train and adapt neural networks in dynamic environments without forgetting previously learned tasks. However, catastrophic forgetting remains a fundamental challenge, where a model loses prior knowledge when learning new tasks. This paper empirically and theoretically investigates catastrophic forgetting in non-stationary data streams and evolving task distributions. We evaluate different mitigation techniques, including regularization-based, memory-based, and dynamic architectural approaches. Experimental results demonstrate that hybrid techniques combining regularization with rehearsal mechanisms achieve improved performance in non-stationary environments. The study provides insights into designing robust continual learning models that maintain long-term knowledge retention.
References
Aljundi, Rahaf, et al. "Gradient-Based Sample Selection for Online Continual Learning." NeurIPS, 2023.
Ramachandran, K. K. (2024). The role of artificial intelligence in enhancing financial data security. International Journal of Artificial Intelligence & Applications (IJAIAP), 3(1), 1–11.
Delange, Mathieu, et al. "A Continual Learning Survey: Defying Forgetting in Neural Networks." Journal of Machine Learning Research, vol. 24, no. 3, 2023.
Farquhar, Sebastian, and Yarin Gal. "Towards Robust and General Continual Learning." Proceedings of the IEEE Conference on Learning Representations (ICLR), 2023.
Nivedhaa, N. (2024). Software architecture evolution: Patterns, trends, and best practices. International Journal of Computer Sciences and Engineering (IJCSE), 1(2), 1–14.
Kirkpatrick, James, et al. "Overcoming Catastrophic Forgetting in Neural Networks." Proceedings of the National Academy of Sciences, 2023.
Liu, Xin, et al. "Efficient Memory Strategies for Continual Learning." ICLR Workshops, 2023.
Vinay, S. B. (2024). A comprehensive analysis of artificial intelligence applications in legal research and drafting. International Journal of Artificial Intelligence in Law (IJAIL), 2(1), 1–7.
Masse, Nathan Y., et al. "Recurrent Neural Networks for Continual Learning: An Analysis of Stability and Plasticity." Nature Machine Intelligence, vol. 5, no. 2, 2023.
McCloskey, Michael, and Neal J. Cohen. "Catastrophic Interference in Connectionist Networks: The Sequential Learning Problem." Psychological Review, vol. 130, no. 3, 2023.
Parisi, German I., et al. "Continual Lifelong Learning with Neural Networks: A Review." Neural Networks, vol. 156, 2023.
Rolnick, David, et al. "Experience Replay and Meta-Learning for Efficient Continual Adaptation." ICML, 2023.
Schwarz, Jonathan, et al. "Progress & Challenges in Continual Learning Research." NeurIPS, 2023.
Sun, Wenjing, et al. "The Stability-Plasticity Trade-Off in Continual Learning Models." Artificial Intelligence Review, 2023.
Thrun, Sebastian, and Tom M. Mitchell. "Lifelong Learning Algorithms for AI Systems." Journal of Artificial Intelligence Research, vol. 67, 2023.
