AI Paper Atlas

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

Beyond MLE: Convex Learning for Text Generation

Chenze Shao, Zhengrui Ma, Min Zhang, Yang Feng

NeurIPS 2023 · Large language models › Language, safety and interpretability · approach: Controlled text generation · 0 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
add convex-composition loss (exp, k=1) to TransformerTransformer · WMT14 EN-DE greedyBLEU: 26.48 -> 26.92Improved
add convex-composition loss to Transformer, beam5Transformer · WMT14 DE-ENBLEU: 31.21 -> 31.33Improved
add convex loss: greedy-beam gapTransformer · WMT14 EN-DEBLEU gap greedy vs beam5: 1.09 -> 0.86Improved
add convex loss to CMLMCMLM · WMT14 EN-DEBLEU: 11.22 -> 20.45Improved
add convex loss to Vanilla-NATVanilla-NAT · WMT14 EN-DEBLEU: 10.41 -> 16.74Improved
add convex loss to CTCCTC · WMT14 DE-ENBLEU: 20.53 -> 26.68Improved
increase exponent k in CMLM from 1 to 8 (ablation)CMLM · WMT14 EN-DEOutput NLL: 20.57 -> 4.88Improved
increase k in CMLM from 1 to 5 (fluency) (ablation)CMLM · WMT14 EN-DEExternal PPL: 939.34 -> 213.84Improved
increase k in CMLM from 5 to 8 (fluency) (ablation)CMLM · WMT14 EN-DEExternal PPL: 213.84 -> 218.68No clear effect
raise k for AR model from 1 to 3 (ablation)Transformer + Convex · WMT14 EN-DE beam5BLEU: 27.78 -> 26.76Hurt
use k=0.5 for AR model (ablation)Transformer + Convex · WMT14 EN-DE greedyBLEU: 26.48 -> 26.89Improved
use power-function convex composition instead of exponential (ablation)Vanilla-NAT · WMT14 EN-DEBLEUHurt

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