@inproceedings{liu-etal-2024-iad-context,
title = "{IAD}: In-Context Learning Ability Decoupler of Large Language Models in Meta-Training",
author = "Liu, Yuhan and
Chen, Xiuying and
Xing, Gao and
Zhang, Ji and
Yan, Rui",
editor = "Calzolari, Nicoletta and
Kan, Min-Yen and
Hoste, Veronique and
Lenci, Alessandro and
Sakti, Sakriani and
Xue, Nianwen",
booktitle = "Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024)",
month = may,
year = "2024",
address = "Torino, Italia",
publisher = "ELRA and ICCL",
url = "https://aclanthology.org/2024.lrec-main.749",
pages = "8535--8545",
abstract = "Large Language Models (LLMs) exhibit remarkable In-Context Learning (ICL) ability, where the model learns tasks from prompts consisting of input-output examples. However, the pre-training objectives of LLMs often misalign with ICL objectives. They{'}re mainly pre-trained with methods like masked language modeling and next-sentence prediction. On the other hand, ICL leverages example pairs to guide the model in generating task-aware responses such as text classification and question-answering tasks. The basic pre-training task-related capabilities can sometimes overshadow or conflict with task-specific subtleties required in ICL. To address this, we propose an In-context learning Ability Decoupler (IAD). The model aims to separate the ICL ability from the general ability of LLMs in the meta-training phase, where the ICL-related parameters are separately tuned to adapt for ICL tasks. Concretely, we first identify the parameters that are suitable for ICL by transference-driven gradient importance. We then propose a new max-margin loss to emphasize the separation of the general and ICL abilities. The loss is defined as the difference between the output of ICL and the original LLM, aiming to prevent the overconfidence of the LLM. By meta-training these ICL-related parameters with max-margin loss, we enable the model to learn and adapt to new tasks with limited data effectively. Experimental results show that IAD{'}s capability yields state-of-the-art performance on benchmark datasets by utilizing only 30{\%} of the model{'}s parameters. Ablation study and detailed analysis prove the separation of the two abilities.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="liu-etal-2024-iad-context">
<titleInfo>
<title>IAD: In-Context Learning Ability Decoupler of Large Language Models in Meta-Training</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yuhan</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiuying</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Gao</namePart>
<namePart type="family">Xing</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ji</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Rui</namePart>
<namePart type="family">Yan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2024-05</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Nicoletta</namePart>
<namePart type="family">Calzolari</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Min-Yen</namePart>
<namePart type="family">Kan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Veronique</namePart>
<namePart type="family">Hoste</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Alessandro</namePart>
<namePart type="family">Lenci</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sakriani</namePart>
<namePart type="family">Sakti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nianwen</namePart>
<namePart type="family">Xue</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>ELRA and ICCL</publisher>
<place>
<placeTerm type="text">Torino, Italia</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Large Language Models (LLMs) exhibit remarkable In-Context Learning (ICL) ability, where the model learns tasks from prompts consisting of input-output examples. However, the pre-training objectives of LLMs often misalign with ICL objectives. They’re mainly pre-trained with methods like masked language modeling and next-sentence prediction. On the other hand, ICL leverages example pairs to guide the model in generating task-aware responses such as text classification and question-answering tasks. The basic pre-training task-related capabilities can sometimes overshadow or conflict with task-specific subtleties required in ICL. To address this, we propose an In-context learning Ability Decoupler (IAD). The model aims to separate the ICL ability from the general ability of LLMs in the meta-training phase, where the ICL-related parameters are separately tuned to adapt for ICL tasks. Concretely, we first identify the parameters that are suitable for ICL by transference-driven gradient importance. We then propose a new max-margin loss to emphasize the separation of the general and ICL abilities. The loss is defined as the difference between the output of ICL and the original LLM, aiming to prevent the overconfidence of the LLM. By meta-training these ICL-related parameters with max-margin loss, we enable the model to learn and adapt to new tasks with limited data effectively. Experimental results show that IAD’s capability yields state-of-the-art performance on benchmark datasets by utilizing only 30% of the model’s parameters. Ablation study and detailed analysis prove the separation of the two abilities.</abstract>
<identifier type="citekey">liu-etal-2024-iad-context</identifier>
<location>
<url>https://aclanthology.org/2024.lrec-main.749</url>
</location>
<part>
<date>2024-05</date>
<extent unit="page">
<start>8535</start>
<end>8545</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T IAD: In-Context Learning Ability Decoupler of Large Language Models in Meta-Training
%A Liu, Yuhan
%A Chen, Xiuying
%A Xing, Gao
%A Zhang, Ji
%A Yan, Rui
%Y Calzolari, Nicoletta
%Y Kan, Min-Yen
%Y Hoste, Veronique
%Y Lenci, Alessandro
%Y Sakti, Sakriani
%Y Xue, Nianwen
%S Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024)
%D 2024
%8 May
%I ELRA and ICCL
%C Torino, Italia
%F liu-etal-2024-iad-context
%X Large Language Models (LLMs) exhibit remarkable In-Context Learning (ICL) ability, where the model learns tasks from prompts consisting of input-output examples. However, the pre-training objectives of LLMs often misalign with ICL objectives. They’re mainly pre-trained with methods like masked language modeling and next-sentence prediction. On the other hand, ICL leverages example pairs to guide the model in generating task-aware responses such as text classification and question-answering tasks. The basic pre-training task-related capabilities can sometimes overshadow or conflict with task-specific subtleties required in ICL. To address this, we propose an In-context learning Ability Decoupler (IAD). The model aims to separate the ICL ability from the general ability of LLMs in the meta-training phase, where the ICL-related parameters are separately tuned to adapt for ICL tasks. Concretely, we first identify the parameters that are suitable for ICL by transference-driven gradient importance. We then propose a new max-margin loss to emphasize the separation of the general and ICL abilities. The loss is defined as the difference between the output of ICL and the original LLM, aiming to prevent the overconfidence of the LLM. By meta-training these ICL-related parameters with max-margin loss, we enable the model to learn and adapt to new tasks with limited data effectively. Experimental results show that IAD’s capability yields state-of-the-art performance on benchmark datasets by utilizing only 30% of the model’s parameters. Ablation study and detailed analysis prove the separation of the two abilities.
%U https://aclanthology.org/2024.lrec-main.749
%P 8535-8545
Markdown (Informal)
[IAD: In-Context Learning Ability Decoupler of Large Language Models in Meta-Training](https://aclanthology.org/2024.lrec-main.749) (Liu et al., LREC-COLING 2024)
ACL