AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

FACE: Evaluating Natural Language Generation with Fourier Analysis of Cross-Entropy

Zuhao Yang, Yingfang Yuan, Yang Xu, SHUO ZHAN, Huajun Bai, Kefan Chen

NeurIPS 2023 · Large language models › Language, safety and interpretability · approach: Statistical frameworks for LLM evaluation · 3 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Use contrastive decoding instead of nucleus samplingGPT2 conditional generation · Conditional generationFACE-SO: 0.42 -> 0.44Improved
Use contrastive decoding instead of greedy decodingGPT2 conditional generation · Conditional generationFACE-CORR: 0.56 -> 0.75Improved
Use contrastive decoding instead of nucleus samplingGPT2 conditional generation · Conditional generationPerplexity: 13.13 -> 14.39Mixed
Add temperature to top-k samplingGPT2 conditional generation · Conditional generationFACE-SO: 0.40 -> 0.34Hurt
Use FACE-SO instead of MAUVE for human-like correlationFACE-SO · Human judgmentsSpearman correlation: 0.952 -> 0.881Hurt
Restrict to human-vs-model pairs: FACE-SO instead of MAUVEFACE-SO · Human-like judgmentsSpearman correlation: 0.214 -> 0.357Improved
Use human-human folds vs human-model foldFACE-SO · Sanity testFACE-SO: 0.44 -> 0.45Improved
Use human-model fold instead of human-human foldFACE-SPEAR · Sanity testFACE-SPEAR: 0.05 -> 0.17Improved
Increase GPT2 size from sm to xlGPT2 · Wiki textMAUVE: 0.677 -> 0.186Hurt
Use nucleus sampling instead of ancestral samplingGPT2-xl · MAUVE original dataFACE-SO: 0.472 -> 0.481Improved
Change estimator model m_est among GPT2-sm to xl (ablation)FACE · webtext and GPT2 outputsspectra shapeNo clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy