How Out-of-Distribution important is
Primary Area: transfer learning, meta learning, and lifelong learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: continual learning, data drift, out-of-distribution, self training
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
TL;DR: MLOps Pipeline and Model for Maintaining Accuracy in a Changing Image Environment
Abstract: Class Incremental Learning (CIL) has gained significant attention in recent years due to its potential to adaptively learn from a non-stationary data distribution. The challenge of CIL primarily revolves around the model's ability to learn new classes without forgetting previously acquired knowledge. Recent research trends has achieved significant milestones, yet the continuity of learning can be further strengthened by integrating the concepts of "self-training", "out-of-distribution", and "data drift". In this paper, we propose a novel approach that integrates "Continual Learning", "Self-Training", "Out-of-Distribution recognition", and "Data Drift" concepts to advance the capabilities of class incremental learning systems. Drawing inspiration from works such as "A Theoretical Study on Solving Continual Learning", and "CSI: Novelty Detection via Contrastive Learning on Distributionally Shifted Instances". We propose a model that satisfies the four concepts mentioned above. Our experimental results demonstrate the efficacy of this method in mitigating catastrophic forgetting and ensuring consistent performance across a diverse range of classes.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 1316
Loading