Can Input Attributions Explain Inductive Reasoning in In-Context Learning?

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Ye, Mengyu, Kuribayashi, Tatsuki, Kobayashi, Goro, Suzuki, Jun
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911046217760768
author Ye, Mengyu
Kuribayashi, Tatsuki
Kobayashi, Goro
Suzuki, Jun
author_facet Ye, Mengyu
Kuribayashi, Tatsuki
Kobayashi, Goro
Suzuki, Jun
contents Interpreting the internal process of neural models has long been a challenge. This challenge remains relevant in the era of large language models (LLMs) and in-context learning (ICL); for example, ICL poses a new issue of interpreting which example in the few-shot examples contributed to identifying/solving the task. To this end, in this paper, we design synthetic diagnostic tasks of inductive reasoning, inspired by the generalization tests typically adopted in psycholinguistics. Here, most in-context examples are ambiguous w.r.t. their underlying rule, and one critical example disambiguates it. The question is whether conventional input attribution (IA) methods can track such a reasoning process, i.e., identify the influential example, in ICL. Our experiments provide several practical findings; for example, a certain simple IA method works the best, and the larger the model, the generally harder it is to interpret the ICL with gradient-based IA methods.
format Preprint
id arxiv_https___arxiv_org_abs_2412_15628
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Can Input Attributions Explain Inductive Reasoning in In-Context Learning?
Ye, Mengyu
Kuribayashi, Tatsuki
Kobayashi, Goro
Suzuki, Jun
Computation and Language
Interpreting the internal process of neural models has long been a challenge. This challenge remains relevant in the era of large language models (LLMs) and in-context learning (ICL); for example, ICL poses a new issue of interpreting which example in the few-shot examples contributed to identifying/solving the task. To this end, in this paper, we design synthetic diagnostic tasks of inductive reasoning, inspired by the generalization tests typically adopted in psycholinguistics. Here, most in-context examples are ambiguous w.r.t. their underlying rule, and one critical example disambiguates it. The question is whether conventional input attribution (IA) methods can track such a reasoning process, i.e., identify the influential example, in ICL. Our experiments provide several practical findings; for example, a certain simple IA method works the best, and the larger the model, the generally harder it is to interpret the ICL with gradient-based IA methods.
title Can Input Attributions Explain Inductive Reasoning in In-Context Learning?
topic Computation and Language
url https://arxiv.org/abs/2412.15628