# Pathologies of Pre-trained Language Models in Few-shot Fine-tuning

arXiv: https://arxiv.org/abs/2204.08039 · HTML: https://mlsanity.com/paper/2204.08039/

## Key facts

- Cited by 3 papers in the map (data through 14 Oct 2026).
- Builds on 25 of its 33 references that are in the map.
- First posted on arXiv 17 Apr 2022.

## Builds on (most cited first, 25 of 25)

- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [RoBERTa: A Robustly Optimized BERT Pretraining Approach](https://mlsanity.com/paper/1907.11692/) (2019)
- [Axiomatic Attribution for Deep Networks](https://mlsanity.com/paper/1703.01365/) (2017)
- [A Broad-Coverage Challenge Corpus for Sentence Understanding through Inference](https://mlsanity.com/paper/1704.05426/) (2017)
- [Character-level Convolutional Networks for Text Classification](https://mlsanity.com/paper/1509.01626/) (2015)
- [Language Models as Knowledge Bases?](https://mlsanity.com/paper/1909.01066/) (2019)
- [A large annotated corpus for learning natural language inference](https://mlsanity.com/paper/1508.05326/) (2015)
- [Calibrate Before Use: Improving Few-Shot Performance of Language Models](https://mlsanity.com/paper/2102.09690/) (2021)
- [Making Pre-trained Language Models Better Few-shot Learners](https://mlsanity.com/paper/2012.15723/) (2020)
- [How Can We Know What Language Models Know?](https://mlsanity.com/paper/1911.12543/) (2019)
- [Exploiting Cloze Questions for Few Shot Text Classification and Natural Language Inference](https://mlsanity.com/paper/2001.07676/) (2020)
- [It's Not Just Size That Matters: Small Language Models Are Also Few-Shot Learners](https://mlsanity.com/paper/2009.07118/) (2020)
- [Towards Faithfully Interpretable NLP Systems: How should we define and evaluate faithfulness?](https://mlsanity.com/paper/2004.03685/) (2020)
- [Explainable Prediction of Medical Codes from Clinical Text](https://mlsanity.com/paper/1802.05695/) (2018)
- [Towards Debiasing Fact Verification Models](https://mlsanity.com/paper/1908.05267/) (2019)
- [Self-training Improves Pre-training for Natural Language Understanding](https://mlsanity.com/paper/2010.02194/) (2020)
- [Towards Interpreting and Mitigating Shortcut Learning Behavior of NLU Models](https://mlsanity.com/paper/2103.06922/) (2021)
- [A Continuously Growing Dataset of Sentential Paraphrases](https://mlsanity.com/paper/1708.00391/) (2017)
- [Generating Hierarchical Explanations on Text Classification via Feature Interaction Detection](https://mlsanity.com/paper/2004.02015/) (2020)
- [Learning Variational Word Masks to Improve the Interpretability of Neural Text Classifiers](https://mlsanity.com/paper/2010.00667/) (2020)
- [Analyzing Compositionality-Sensitivity of NLI Models](https://mlsanity.com/paper/1811.07033/) (2018)
- [Avoiding Inference Heuristics in Few-shot Prompt-based Finetuning](https://mlsanity.com/paper/2109.04144/) (2021)
- [Revisiting Self-Training for Few-Shot Learning of Language Model](https://mlsanity.com/paper/2110.01256/) (2021)
- [Explaining Neural Network Predictions on Sentence Pairs via Learning Word-Group Masks](https://mlsanity.com/paper/2104.04488/) (2021)

## Built on by (most cited first, 3 of 3)

- [Towards Equipping Transformer with the Ability of Systematic Compositionality](https://mlsanity.com/paper/2312.07280/) (2023)
- [Feeding LLM Annotations to BERT Classifiers at Your Own Risk](https://mlsanity.com/paper/2504.15432/) (2025)
- [Entropy Minimization without Model Collapse: Mitigating Prediction Bias in Medical Imaging](https://mlsanity.com/paper/2606.02339/) (2026)

## Abstract (arXiv)

Although adapting pre-trained language models with few examples has shown promising performance on text classification, there is a lack of understanding of where the performance gain comes from. In this work, we propose to answer this question by interpreting the adaptation behavior using post-hoc explanations from model predictions. By modeling feature statistics of explanations, we discover that (1) without fine-tuning, pre-trained models (e.g. BERT and RoBERTa) show strong prediction bias across labels; (2) although few-shot fine-tuning can mitigate the prediction bias and demonstrate promising prediction performance, our analysis shows models gain performance improvement by capturing non-task-related features (e.g. stop words) or shallow data patterns (e.g. lexical overlaps). These observations alert that pursuing model performance with fewer examples may incur pathological prediction behavior, which requires further sanity check on model predictions and careful design in model evaluations in few-shot fine-tuning.