Reproduce “Active Learning on a Budget: Opposite Strategies Suit High and Low Budgets” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Guy Hacohen, Avihu Dekel, Daphna Weinshall
Research from The Hebrew University of Jerusalem demonstrates that optimal active learning strategies depend on the labeling budget, with opposing approaches suitable for low and high data availability. A new method, TypiClust, effectively addresses the 'cold start' problem by selecting typical and diverse examples, achieving robust performance gains over existing active learning techniques and significantly boosting semi-supervised learning accuracy in low-budget scenarios.
Key insights
- Optimal active learning strategies are budget-dependent; querying 'typical' examples is more beneficial in low-budget regimes, while 'atypical' or 'uncertain' examples become more effective for higher budgets.
- Uncertainty sampling is inherently unsuitable for low-budget active learning, not just due to poor uncertainty estimation, but because querying 'hard' examples in early stages hinders foundational learning.
- Self-supervised learning provides robust feature embeddings critical for reliable example selection and overcoming the 'cold start' problem when initial labeled data is scarce or absent.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.