Efficient Long-Tail Learning in Latent Space by sampling Synthetic Data
This addresses the problem of biased models in imbalanced datasets for machine learning practitioners, offering a computationally efficient solution that is incremental but effective.
The paper tackles long-tail classification in imbalanced datasets by generating synthetic data in the latent space of Vision Foundation Models and training a linear classifier with mixed real and synthetic data, achieving new state-of-the-art on CIFAR-100-LT and strong performance on Places-LT while reducing trainable parameters to just those in the linear model.
Imbalanced classification datasets pose significant challenges in machine learning, often leading to biased models that perform poorly on underrepresented classes. With the rise of foundation models, recent research has focused on the full, partial, and parameter-efficient fine-tuning of these models to deal with long-tail classification. Despite the impressive performance of these works on the benchmark datasets, they still fail to close the gap with the networks trained using the balanced datasets and still require substantial computational resources, even for relatively smaller datasets. Underscoring the importance of computational efficiency and simplicity, in this work we propose a novel framework that leverages the rich semantic latent space of Vision Foundation Models to generate synthetic data and train a simple linear classifier using a mixture of real and synthetic data for long-tail classification. The computational efficiency gain arises from the number of trainable parameters that are reduced to just the number of parameters in the linear model. Our method sets a new state-of-the-art for the CIFAR-100-LT benchmark and demonstrates strong performance on the Places-LT benchmark, highlighting the effectiveness and adaptability of our simple and effective approach.