M³-CIL: A Multidimensional Evaluation Protocol for Class-Incremental Learning
Resumo
No aprendizado incremental de classes, o esquecimento catastrófico é um desafio central. Atualmente, métodos que mitigam este problema são avaliados sobretudo pela acurácia final, ignorando restrições operacionais. Este trabalho propõe o M³-CIL (Multidimensional Metric Matrix), protocolo que investiga a viabilidade prática de tais abordagens em três eixos de análise: retenção de conhecimento, sensibilidade hiperparamétrica e eficiência operacional. Após analisar sete métodos clássicos em cenários de complexidade crescente, conclui-se que estratégias de Replay (especialmente o ER-ACE) oferecem o melhor equilíbrio entre estabilidade-plasticidade, enquanto técnicas de Regularização mostram-se ineficazes em contextos complexos.Referências
Chavan, V., Koch, P., Schlüter, M., and Briese, C. (2023). Towards realistic evaluation of industrial continual learning scenarios with an emphasis on energy consumption and computational footprint. In 2023 IEEE/CVF International Conference on Computer Vision (ICCV), pages 11472–11484.
Goodfellow, I. J., Mirza, M., Xiao, D., Courville, A., and Bengio, Y. (2015). An empirical investigation of catastrophic forgetting in gradient-based neural networks.
Kirkpatrick, J., Pascanu, R., Rabinowitz, N., Veness, J., Desjardins, G., Rusu, A. A., Milan, K., Quan, J., Ramalho, T., Grabska-Barwinska, A., Hassabis, D., Clopath, C., Kumaran, D., and Hadsell, R. (2017). Overcoming catastrophic forgetting in neural networks. Proceedings of the National Academy of Sciences, 114(13):3521–3526.
Kumar, S., Marklund, H., Rao, A., Zhu, Y., Jeon, H. J., Yueyang, L., and Van Roy, B. (2025). Continual learning as computationally constrained reinforcement learning. Foundations and Trends® in Machine Learning, 18(4):913–1053.
Li, Z. and Hoiem, D. (2017). Learning without forgetting.
Lopez-Paz, D. and Ranzato, M. (2017). Gradient episodic memory for continual learning. In Proceedings of the 31st International Conference on Neural Information Processing Systems, pages 6470–6479.
Mermillod, M., Bugaiska, A., and BONIN, P. (2013). The stability-plasticity dilemma: investigating the continuum from catastrophic forgetting to age-limited learning effects. Frontiers in Psychology, Volume 4 - 2013.
Prabhu, A., Torr, P. H. S., and Dokania, P. K. (2020). Gdumb: A simple approach that questions our progress in continual learning. In Computer Vision – ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part II, page 524–540, Berlin, Heidelberg. Springer-Verlag.
Rebuffi, S.-A., Kolesnikov, A., Sperl, G., and Lampert, C. H. (2017). icarl: Incremental classifier and representation learning.
van de Ven, G. M. and Tolias, A. S. (2019). Three scenarios for continual learning.
Zhou, D.-W., Wang, Q.-W., Qi, Z.-H., Ye, H.-J., Zhan, D.-C., and Liu, Z. (2024). Class-incremental learning: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence, 46(12):9851–9873.
Goodfellow, I. J., Mirza, M., Xiao, D., Courville, A., and Bengio, Y. (2015). An empirical investigation of catastrophic forgetting in gradient-based neural networks.
Kirkpatrick, J., Pascanu, R., Rabinowitz, N., Veness, J., Desjardins, G., Rusu, A. A., Milan, K., Quan, J., Ramalho, T., Grabska-Barwinska, A., Hassabis, D., Clopath, C., Kumaran, D., and Hadsell, R. (2017). Overcoming catastrophic forgetting in neural networks. Proceedings of the National Academy of Sciences, 114(13):3521–3526.
Kumar, S., Marklund, H., Rao, A., Zhu, Y., Jeon, H. J., Yueyang, L., and Van Roy, B. (2025). Continual learning as computationally constrained reinforcement learning. Foundations and Trends® in Machine Learning, 18(4):913–1053.
Li, Z. and Hoiem, D. (2017). Learning without forgetting.
Lopez-Paz, D. and Ranzato, M. (2017). Gradient episodic memory for continual learning. In Proceedings of the 31st International Conference on Neural Information Processing Systems, pages 6470–6479.
Mermillod, M., Bugaiska, A., and BONIN, P. (2013). The stability-plasticity dilemma: investigating the continuum from catastrophic forgetting to age-limited learning effects. Frontiers in Psychology, Volume 4 - 2013.
Prabhu, A., Torr, P. H. S., and Dokania, P. K. (2020). Gdumb: A simple approach that questions our progress in continual learning. In Computer Vision – ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part II, page 524–540, Berlin, Heidelberg. Springer-Verlag.
Rebuffi, S.-A., Kolesnikov, A., Sperl, G., and Lampert, C. H. (2017). icarl: Incremental classifier and representation learning.
van de Ven, G. M. and Tolias, A. S. (2019). Three scenarios for continual learning.
Zhou, D.-W., Wang, Q.-W., Qi, Z.-H., Ye, H.-J., Zhan, D.-C., and Liu, Z. (2024). Class-incremental learning: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence, 46(12):9851–9873.
Publicado
19/10/2026
Como Citar
LEITE, Liedson D. O.; SARMENTO, Gabriel F.; OLIVEIRA, Thyago T. M. de; CAVALCANTE, Rodolfo C.; MINKU, Leandro L.; OLIVEIRA, Gustavo H. F. de M..
M³-CIL: A Multidimensional Evaluation Protocol for Class-Incremental Learning. In: ENCONTRO NACIONAL DE INTELIGÊNCIA ARTIFICIAL E COMPUTACIONAL (ENIAC), 23. , 2026, Cuiabá/MT.
Anais [...].
Porto Alegre: Sociedade Brasileira de Computação,
2026
.
p. 2421-2432.
ISSN 2763-9061.
DOI: https://doi.org/10.5753/eniac.2026.28133.
