Donghyun Choi

DongHyun Choi, Lucas Spangher, Chris Hidey et al.

Transformer-based Large Language Models, which suffer from high computational costs, advance so quickly that techniques proposed to streamline earlier iterations are not guaranteed to benefit more modern models. Building upon the Funnel Transformer proposed by Dai and Le (2020), which progressively compresses intermediate representations, we investigate the impact of funneling in contemporary Gemma2 Transformer architectures. We systematically evaluate various funnel configurations and recovery methods, comparing: (1) standard pretraining to funnel-aware pretraining strategies, (2) the impact of funnel-aware fine-tuning, and (3) the type of sequence recovery operation. Our results demonstrate that funneling creates information bottlenecks that propagate through deeper network layers, particularly in larger models (e.g., Gemma 7B), leading to at times unmanageable performance lost. However, carefully selecting the funneling layer and employing effective recovery strategies, can substantially mitigate performance losses, achieving up to a 44\% reduction in latency. Our findings highlight key trade-offs between computational efficiency and model accuracy, providing practical guidance for deploying funnel-based approaches in large-scale natural language applications.

DongHyun Choi, Myeong Cheol Shin, EungGyun Kim et al.

Out-of-domain (OOD) input detection is vital in a task-oriented dialogue system since the acceptance of unsupported inputs could lead to an incorrect response of the system. This paper proposes OutFlip, a method to generate out-of-domain samples using only in-domain training dataset automatically. A white-box natural language attack method HotFlip is revised to generate out-of-domain samples instead of adversarial examples. Our evaluation results showed that integrating OutFlip-generated out-of-domain samples into the training dataset could significantly improve an intent classification model's out-of-domain detection performance.

DongHyun Choi, IlNam Park, Myeong Cheol Shin et al.

This paper attempts to analyze the Korean sentence classification system for a chatbot. Sentence classification is the task of classifying an input sentence based on predefined categories. However, spelling or space error contained in the input sentence causes problems in morphological analysis and tokenization. This paper proposes a novel approach of Integrated Eojeol (Korean syntactic word separated by space) Embedding to reduce the effect that poorly analyzed morphemes may make on sentence classification. It also proposes two noise insertion methods that further improve classification performance. Our evaluation results indicate that the proposed system classifies erroneous sentences more accurately than the baseline system by 17%p.0