Less Data

Research on "less data" methods focuses on developing machine learning models that achieve high performance with significantly reduced training data requirements. Current efforts explore techniques like preference optimization (e.g., Triple Preference Optimization), unsupervised learning within forward-forward algorithms, and neuro-symbolic approaches that leverage domain knowledge or proof systems to improve data efficiency. These advancements are crucial for addressing the high cost and environmental impact of training large models, expanding the applicability of AI to resource-constrained settings, and mitigating bias in datasets where labeled data is scarce or expensive to obtain.