A Survey on Stability of Learning with Limited Labelled Data and its Sensitivity to the Effects of Randomness

Abstract

Learning with limited labelled data, such as prompting, in-context learning, fine-tuning, meta-learning or few-shot learning, aims to effectively train a model using only a small amount of labelled samples. However, these approaches have been observed to be excessively sensitive to the effects of uncontrolled randomness caused by non-determinism in the training process. The randomness negatively affects the stability of the models, leading to large variances in results across training runs. When such sensitivity is disregarded, it can unintentionally, but unfortunately also intentionally, create an imaginary perception of research progress. Recently, this area started to attract research attention and the number of relevant studies is continuously growing. In this survey, we provide a comprehensive overview of 415 papers addressing the effects of randomness on the stability of learning with limited labelled data. We distinguish between four main tasks addressed in the papers (investigate/evaluate; determine; mitigate; benchmark/compare/report randomness effects), providing findings for each one. Furthermore, we identify and discuss seven challenges and open problems together with possible directions to facilitate further research. The ultimate goal of this survey is to emphasise the importance of this growing research area, which so far has not received an appropriate level of attention, and reveal impactful directions for future research.

Accepted to ACM Comput. Surv. 2024

References in corpus (40)

• A Survey of Large Language Models
• Underspecification Presents Challenges for Credibility in Modern Machine Learning
• A Survey on In-context Learning
• Challenges and Applications of Large Language Models
• Can ChatGPT Understand Too? A Comparative Study on ChatGPT and Fine-tuned BERT
• An Information-theoretic Approach to Prompt Engineering Without Ground Truth Labels
• Large Language Model Is Not a Good Few-shot Information Extractor, but a Good Reranker for Hard Samples!
• PromptRobust: Towards Evaluating the Robustness of Large Language Models on Adversarial Prompts
• Towards Training Reproducible Deep Learning Models
• Sources of Irreproducibility in Machine Learning: A Review
• ScatterShot: Interactive In-context Example Curation for Text Transformation
• Randomness In Neural Network Training: Characterizing The Impact of Tooling
• Reproducibility of Machine Learning: Terminology, Recommendations and Open Issues
• Large Language Models Sensitivity to The Order of Options in Multiple-Choice Questions
• Large Language Model-Aware In-Context Learning for Code Generation
• In-Context Learning with Long-Context Models: An In-Depth Exploration
• In-context Learning with Retrieved Demonstrations for Language Models: A Survey
• In-Context Learning with Iterative Demonstration Selection
• Lessons from the Trenches on Reproducible Evaluation of Language Models
• In-context Example Selection with Influences
• Let's Learn Step by Step: Enhancing In-Context Learning Ability with Curriculum Learning
• Fool Your (Vision and) Language Model With Embarrassingly Simple Permutations
• Mind Your Format: Towards Consistent Evaluation of In-Context Learning Improvements
• Efficient multi-prompt evaluation of LLMs
• Which Examples to Annotate for In-Context Learning? Towards Effective and Efficient Selection
• Misconfidence-based Demonstration Selection for LLM In-Context Learning
• MixPro: Simple yet Effective Data Augmentation for Prompt-based Learning
• Revisiting Demonstration Selection Strategies in In-Context Learning
• When Benchmarks are Targets: Revealing the Sensitivity of Large Language Model Leaderboards
• Unveiling Selection Biases: Exploring Order and Token Sensitivity in Large Language Models
• The Impact of Demonstrations on Multilingual In-Context Learning: A Multidimensional Analysis
• StablePT: Towards Stable Prompting for Few-shot Learning via Input Separation
• NoisyICL: A Little Noise in Model Parameters Calibrates In-context Learning
• Batch-ICL: Effective, Efficient, and Order-Agnostic In-Context Learning
• Unveiling the Lexical Sensitivity of LLMs: Combinatorial Optimization for Prompt Enhancement
• On Sensitivity of Learning with Limited Labelled Data to the Effects of Randomness: Impact of Interactions and Systematic Choices
• Quantifying Variance in Evaluation Benchmarks
• Changing Answer Order Can Decrease MMLU Accuracy
• Fighting Randomness with Randomness: Mitigating Optimisation Instability of Fine-Tuning using Delayed Ensemble and Noisy Interpolation
• Designing Informative Metrics for Few-Shot Example Selection

Cited by in corpus (2)

• A Survey on Automatic Credibility Assessment Using Textual Credibility Signals in the Era of Large Language Models
• Improved Sub-Visible Particle Classification in Flow Imaging Microscopy via Generative AI-Based Image Synthesis