Skip to main content

Overview

COSP is a technique that improves few-shot learning by selecting high-quality examples based on consistency and confidence of model responses. It identifies examples the model can process reliably. The process involves:
  1. Example Generation: Generate multiple responses per example, collect confidence scores
  2. Example Selection: Select examples with low entropy and high repetitiveness

Example

Benefits

  • Improved Consistency: Select examples with low entropy and high repetitiveness
  • Automated Selection: No manual example curation needed
  • Quality Metrics: Quantifiable measure of example quality

References

  1. Original COSP Paper (https://arxiv.org/abs/2305.14121)
  2. Self-Consistency Improves Chain of Thought Reasoning (https://arxiv.org/abs/2203.11171)