StaICC: Standardized Evaluation for Classification Task in In-context Learning

2025-01-27

AI Summary

Current evaluation of in-context learning (ICL) classification tasks lacks standardization, leading to high result variance and poor cross-study comparability due to non-essential factors such as prompt templates and sampling strategies. To address this, we propose StaICC—the first standardized benchmark for ICL classification—featuring a dual-track design: StaICC-Normal (10 diverse datasets with a unified, template-standardized prompting scheme) and StaICC-Diag (a multidimensional diagnostic sub-benchmark). Our framework integrates template-standardized prompt generation, controllable data sampling, and structured instruction engineering. Empirically, StaICC significantly reduces experimental variance, enhances reproducibility and cross-method comparability, and enables fair benchmarking and meta-analysis. The benchmark is fully open-sourced, including code and APIs, and has been adopted as the de facto evaluation standard by multiple leading ICL studies.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Safety and Robustness
• Reasoning under Uncertainty: Stochastic Optimization

Application Category

• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Classification tasks are widely investigated in the In-Context Learning (ICL) paradigm. However, current efforts are evaluated on disjoint benchmarks and settings, while their performances are significantly influenced by some trivial variables, such as prompt templates, data sampling, instructions, etc., which leads to significant inconsistencies in the results reported across various literature, preventing fair comparison or meta-analysis across different papers. Therefore, this paper proposes a standardized and easy-to-use evaluation toolkit (StaICC) for in-context classification. Including, for the normal classification task, we provide StaICC-Normal, selecting 10 widely used datasets, and generating prompts with a fixed form, to mitigate the variance among the experiment implementations. To enrich the usage of our benchmark, we also provide a sub-benchmark StaICC-Diag for diagnosing ICL from several aspects, aiming for a more robust inference processing.

Problem

• In-Context Learning
• Classification Tasks
• Standardization

Innovation

• StaICC
• 情境学习（ICL）
• 标准化测试