AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

HC2L: Hybrid and Cooperative Contrastive Learning for Cross-Lingual Spoken Language Understanding

Bowen Xing, Ivor Wai-Hung Tsang

TPAMI 2024 · Label-efficient and robust learning › Recognition and long-tail · approach: Multilingual language representation pretraining

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 supervised contrastive learning (source, cross-lingual, multilingual) on top of cross-lingual unsupervised CLGL-CLEF (mBERT) · MultiATIS++overall accuracy: 53.56 -> 58.98Improved
add supervised CL to unsupervised CL (UCL+SCL vs only UCL) (ablation)HC2L · MultiATIS++all metricsImproved
use only supervised CL instead of only unsupervised CL (ablation)HC2L · MultiATIS++all metricsImproved
add source language / cross-lingual / multilingual supervised CL individually (ablation)HC2L · MultiATIS++all metricsImproved
add joint-task multi-label supervised CL (ablation)HC2L · MultiATIS++all metricsImproved
add intent SCL and slot SCL single-task mechanisms (ablation)HC2L · MultiATIS++all metricsImproved
swap encoder to XLM-R and add HC2LXLM-R · MultiATIS++ (XLM-R)overall accuracy: 33.31 -> 56.06Improved
replace GL-CLEF with HC2L on XLM-RXLM-R + GL-CLEF · MultiATIS++ (XLM-R)overall accuracy: 52.69 -> 56.06Improved
replace GL-CLEF with HC2L on XLM-RXLM-R + GL-CLEF · MultiATIS++ (XLM-R)intent accuracy: 92.75 -> 94.85Improved
use HC2L on HindiGL-CLEF · MultiATIS++ Hindioverall accuracy: 26.76 -> 34.83Improved
increase supervised CL weights beyond tuned peak or reduce them (ablation)HC2L · MultiATIS++overall accuracyHurt
HC2L with smaller queue (16 vs 32) and extra lossesGL-CLEF · computationGPU memory / training time: 20.7GB -> 18.9GB; 33s -> 35sMixed

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