OASIS: Open-world Adaptive Self-supervised and Imbalanced-aware System

2025/08/20 by Kim, Miru, Joe, Mugon, Kwon, Minhae
#FOS: Computer and information sciences #I.2 #Machine Learning (cs.LG)

paper · doi:10.48550/arxiv.2508.16656

Abstract

The expansion of machine learning into dynamic environments presents challenges in handling open-world problems where label shift, covariate shift, and unknown classes emerge. Post-training methods have been explored to address these challenges, adapting models to newly emerging data. However, these methods struggle when the initial pre-training is performed on class-imbalanced datasets, limiting generalization to minority classes. To address this, we propose a method that effectively handles open-world problems even when pre-training is conducted on imbalanced data. Our contrastive-based pre-training approach enhances classification performance, particularly for underrepresented classes. Our post-training mechanism generates reliable pseudo-labels, improving model robustness against open-world problems. We also introduce selective activation criteria to optimize the post-training process, reducing unnecessary computation. Extensive experiments demonstrate that our method significantly outperforms state-of-the-art adaptation techniques in both accuracy and efficiency across diverse open-world scenarios.

Citations

• Enhancing Zero-Shot Vision Models by Label-Free Prompt Distribution Learning and Bias Correcting
• Learning Fair Invariant Representations under Covariate and Correlation Shifts Simultaneously
• Tolerant Algorithms for Learning with Arbitrary Covariate Shift
• Bridging Multicalibration and Out-of-distribution Generalization Beyond Covariate Shift
• On the Benefits of Public Representations for Private Transfer Learning under Distribution Shift
• Spectral Prompt Tuning:Unveiling Unseen Classes for Zero-Shot Semantic Segmentation
• Class Distribution Shifts in Zero-Shot Learning: Learning Robust Representations
• L²R: Lifelong Learning for First-stage Retrieval with Backward-Compatible Representations
• SAILOR: Structural Augmentation Based Tail Node Representation Learning
• Domain Adaptation under Open Set Label Shift
• Adapting to Online Label Shift with Provable Guarantees
• MineDojo: Building Open-Ended Embodied Agents with Internet-Scale Knowledge
• Out-of-Distribution Detection with Deep Nearest Neighbors
• Online Adaptation to Label Distribution Shift
• Towards Open World Object Detection
• Counterfactual Zero-Shot and Open-Set Visual Recognition
• Disentangling Label Distribution for Long-tailed Visual Recognition
• Synthesizing the Unseen for Zero-shot Object Detection
• Robust Fairness under Covariate Shift
• Improving robustness against common corruptions by covariate shift adaptation
• Benchmarking Neural Network Robustness to Common Corruptions and Perturbations
• Unsupervised Domain Adaptation by Backpropagation
• ImageNet Large Scale Visual Recognition Challenge
• ImageNet Large Scale Visual Recognition Challenge
• T2I-CompBench++: An Enhanced and Comprehensive Benchmark for Compositional Text-to-image Generation

Related

• CReST: A Class-Rebalancing Self-Training Framework for Imbalanced Semi-Supervised Learning
• ART: Adaptive Resampling-based Training for Imbalanced Classification
• Rethinking Open-World Semi-Supervised Learning: Distribution Mismatch and Inductive Inference
• Self-Supervised Features Improve Open-World Learning
• OwMatch: Conditional Self-Labeling with Consistency for Open-World Semi-Supervised Learning