Challenge: Existing methods to improve confidence calibration of pre-trained language models are still a mystery.
Approach: They propose a method that leverages model explanations to make models less confident with non-inductive attributions.
Outcome: The proposed method improves confidence calibration in all settings and reduces calibration errors when combined with temperature scaling.

Similar Papers

Can Explanations Be Useful for Calibrating Black Box Models? (2022.acl-long)

Copied to clipboard

Challenge: Existing models are often used as black boxes to adapt to new domains, but there is no single recipe for making them work.
Approach: They propose to use black box models to improve their performance on new domains by leveraging explanations of their behavior.
Outcome: The proposed method improves model generalization performance on two tasks using explanations.
Investigating the Impact of Model Instability on Explanations and Uncertainty (2024.findings-acl)

Copied to clipboard

Challenge: Explainable AI methods are typically evaluated holistically, but small perturbations to inputs can vastly distort explanations.
Approach: They artificially simulate epistemic uncertainty in text input by introducing noise at inference time and measure the effect on the output of pre-trained language models.
Outcome: The proposed model can detect salient tokens when uncertain, but it is not reliable when small perturbations are exposed during training.
Just Ask for Calibration: Strategies for Eliciting Calibrated Confidence Scores from Language Models Fine-Tuned with Human Feedback (2023.emnlp-main)

Copied to clipboard

Challenge: Recent studies have shown that unsupervised pre-training produces large language models whose conditional probabilities are remarkably well-calibrated.
Approach: They propose to use verbalized confidences to extract confidence from large language models with reinforcement learning from human feedback to improve their accuracy.
Outcome: The proposed methods reduce the expected calibration error by 50% for RLHF-LMs such as ChatGPT, GPT-4, and Claude.
Re-Examining Calibration: The Case of Question Answering (2022.findings-emnlp)

Copied to clipboard

Challenge: Existing calibration methods do not provide significant gains in accuracy.
Approach: They propose a new calibration metric that better captures whether the model assigns low confidence to wrong predictions and high confidence to correct predictions.
Outcome: The proposed calibration method better captures whether the model assigns low confidence to wrong predictions and high confidence to correct predictions.
On the Interaction of Belief Bias and Explanations (2021.findings-acl)

Copied to clipboard

Challenge: Existing methods to evaluate explainability fail to account for belief biases affecting human performance . previous studies have shown that neural models can make confident predictions relying on artifacts .
Approach: They propose to account for belief bias in explainability by using models of varying quality and adversarial examples.
Outcome: The proposed methods show that results change when using models of varying quality and adversarial examples.
Explanation-based Finetuning Makes Models More Robust to Spurious Cues (2023.acl-long)

Copied to clipboard

Challenge: Large Language Models (LLMs) learn correlations between labels and features that are irrelevant to the task, leading to poor generalization on out-of-distribution data.
Approach: They propose an explanation-based approach to fine tune large language models to generate a free-text explanation supporting their answer.
Outcome: The proposed model is more robust against spurious cues in terms of accuracy drop across four classification tasks: ComVE (+1.2), CREAK (+9.1), e-SNLI (+5.4), and SBIC (+6.5).
Knowing More About Questions Can Help: Improving Calibration in Question Answering (2021.findings-acl)

Copied to clipboard

Challenge: Existing work on calibration focuses on model confidence, such as the max probability of the predicted class.
Approach: They propose a calibration method which estimates whether model correctly predicts answer for each question.
Outcome: The proposed calibration method achieves 5-10% gains on reading comprehension benchmarks.
On the Calibration of Large Language Models and Alignment (2023.findings-emnlp)

Copied to clipboard

Challenge: Large language models are becoming more popular and are proving to be reliable . however, their reliability is often understudied due to their uncertainty and complex structure .
Approach: They conduct a systematic examination of the calibration of aligned language models throughout the entire construction process including pretraining and alignment training.
Outcome: The results shed light on whether popular large language models are well-calibrated and how the training process influences model calibration.
How Can We Know When Language Models Know? On the Calibration of Language Models for Question Answering (2021.tacl-1)

Copied to clipboard

Challenge: Recent studies have shown that language models capture different types of knowledge regarding facts or commonsense knowledge.
Approach: They examine how language models can be calibrated to make their confidence scores correlate better with the likelihood of correctness.
Outcome: The proposed calibration methods improve confidence scores on QA tasks and improve accuracy.
Calibrating the Confidence of Large Language Models by Eliciting Fidelity (2024.emnlp-main)

Copied to clipboard

Challenge: Large language models with RLHF and RLAIF have good alignment but exhibit overconfidence post-alignment.
Approach: They propose a plug-and-play method to estimate the confidence of large language models.
Outcome: The proposed method has shown good calibration performance on 6 RLHF-LMs on four MCQA datasets.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations