Concept-aware Data Construction Improves In-context Learning of
Language Models

D 2024

Concept-aware Data Construction Improves In-context Learning of Language Models

ŠTEFÁNIK, Michal, Marek KADLČÍK and Petr SOJKA

Basic information

Original name

Concept-aware Data Construction Improves In-context Learning of Language Models

Authors

ŠTEFÁNIK, Michal (703 Slovakia, guarantor, belonging to the institution), Marek KADLČÍK (203 Czech Republic, belonging to the institution) and Petr SOJKA (203 Czech Republic, belonging to the institution)

Edition

Bangkok, Thailand, Findings of the Association for Computational Linguistics ACL 2024, p. 12335-12352, 18 pp. 2024

Publisher

Association for Computational Linguistics

Other information

Language

English

Type of outcome

Stať ve sborníku

Field of Study

10201 Computer sciences, information science, bioinformatics

Country of publisher

United States of America

Confidentiality degree

není předmětem státního či obchodního tajemství

Publication form

electronic version available online

References:

fulltext

Organization unit

Faculty of Informatics

ISBN

979-8-89176-099-8

Keywords (in Czech)

efektivita dat; hluboké jazykové modely; učení s vědomím konceptů; kontextové učení

Keywords in English

data efficiency; LLM; concept-aware training; in-context learning

Abstract

V originále

Many recent language models (LMs) are capable of in-context learning (ICL), manifested in the LMs’ ability to perform a new task solely from natural-language instruction. Previous work curating in-context learners assumes that ICL emerges from a vast over-parametrization or the scale of multi-task training. However, recent theoretical work attributes the ICL ability to concept-dependent training data and creates functional in-context learners even in small-scale, synthetic settings.In this work, we practically explore this newly identified axis of ICL quality. We propose Concept-aware Training (CoAT), a framework for constructing training scenarios that make it beneficial for the LM to learn to utilize the analogical reasoning concepts from demonstrations. We find that by using CoAT, pre-trained transformers can learn to better utilise new latent concepts from demonstrations and that such ability makes ICL more robust to the functional deficiencies of the previous models. Finally, we show that concept-aware in-context learners are much more effective in in-context learning a majority of unseen tasks compared to traditional instruction tuning, and fare comparably also to previous in-context learners trained in large-scale multitask learning requiring magnitudes of more training data.

Links

MUNI/A/1590/2023, interní kód MU

Name: Využití technik umělé inteligence pro zpracování dat, komplexní analýzy a vizualizaci rozsáhlých dat

Investor: Masaryk University, Using artificial intelligence techniques for data processing, complex analysis and visualization of large-scale data

MUNI/A/1608/2023, interní kód MU

Name: Zapojení studentů Fakulty informatiky do mezinárodní vědecké komunity 24

Investor: Masaryk University

Citovat

ŠTEFÁNIK, Michal, Marek KADLČÍK and Petr SOJKA. Concept-aware Data Construction Improves In-context Learning of Language Models. Online. In Lun-Wei Ku, Andre Martins, Vivek Srikumar. Findings of the Association for Computational Linguistics ACL 2024. Bangkok, Thailand: Association for Computational Linguistics, 2024, p. 12335-12352. ISBN 979-8-89176-099-8.

@inproceedings{2425558,
   author = {Štefánik, Michal and Kadlčík, Marek and Sojka, Petr},
   address = {Bangkok, Thailand},
   booktitle = {Findings of the Association for Computational Linguistics ACL 2024},
   editor = {Lun-Wei Ku, Andre Martins, Vivek Srikumar},
   keywords = {data efficiency; LLM; concept-aware training; in-context learning},
   howpublished = {elektronická verze "online"},
   language = {eng},
   location = {Bangkok, Thailand},
   isbn = {979-8-89176-099-8},
   pages = {12335-12352},
   publisher = {Association for Computational Linguistics},
   title = {Concept-aware Data Construction Improves In-context Learning of Language Models},
   url = {https://aclanthology.org/2024.findings-acl.733/},
   year = {2024}
}

TY  - JOUR
ID  - 2425558
AU  - Štefánik, Michal - Kadlčík, Marek - Sojka, Petr
PY  - 2024
TI  - Concept-aware Data Construction Improves In-context Learning of Language Models
PB  - Association for Computational Linguistics
CY  - Bangkok, Thailand
SN  - 9798891760998
KW  - data efficiency
KW  - LLM
KW  - concept-aware training
KW  - in-context learning
UR  - https://aclanthology.org/2024.findings-acl.733/
N2  - Many recent language models (LMs) are capable of in-context learning (ICL), manifested in the LMs’ ability to perform a new task solely from natural-language instruction. Previous work curating in-context learners assumes that ICL emerges from a vast over-parametrization or the scale of multi-task training. However, recent theoretical work attributes the ICL ability to concept-dependent training data and creates functional in-context learners even in small-scale, synthetic settings.In this work, we practically explore this newly identified axis of ICL quality. We propose Concept-aware Training (CoAT), a framework for constructing training scenarios that make it beneficial for the LM to learn to utilize the analogical reasoning concepts from demonstrations. We find that by using CoAT, pre-trained transformers can learn to better utilise new latent concepts from demonstrations and that such ability makes ICL more robust to the functional deficiencies of the previous models. Finally, we show that concept-aware in-context learners are much more effective in in-context learning a majority of unseen tasks compared to traditional instruction tuning, and fare comparably also to previous in-context learners trained in large-scale multitask learning requiring magnitudes of more training data.
ER  -

ŠTEFÁNIK, Michal, Marek KADLČÍK and Petr SOJKA. Concept-aware Data Construction Improves In-context Learning of Language Models. Online. In Lun-Wei Ku, Andre Martins, Vivek Srikumar. \textit{Findings of the Association for Computational Linguistics ACL 2024}. Bangkok, Thailand: Association for Computational Linguistics, 2024, p.~12335-12352. ISBN~979-8-89176-099-8.

Detailed Information on Publication Record