Latent Boost: Leveraging Latent Space Distance Metrics to Augment Classification Performance
Keywords: Latent Space, Distance Metrics, Magnet Loss
TL;DR: Latent Boost enhances classification by optimizing latent representation distances, leading to better clustering, improved classification performance, and faster convergence.
Abstract: The pursuit of boosting classification performance in Machine Learning has primarily focused on refining model architectures and hyperparameters through probabilistic loss optimization. However, such an approach often neglects the profound, untapped potential embedded in internal structural information, which can significantly elevate the training process. In this work, we introduce Latent Boost, a novel approach that incorporates the very definition of classification via latent representation distance metrics to enhance the conventional dataset-oriented classification training. Thus during training, the model is not only optimized for classification metrics of the discrete data points but also adheres to the rule that the collective representation zones of each class should be sharply clustered. By leveraging the rich structural insights of high-dimensional latent representations, Latent Boost not only improves classification metrics like F1-Scores but also brings additional benefits of improved interpretability with higher silhouette scores and steady-fast convergence with fewer training epochs. Latent Boost brings these performance and latent structural benefits with minimum additional cost and no data-specific requirements.
Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation 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.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
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: 10905
Loading