Papers by Youngrok Ko
Enhancing Large Language Model Based Sequential Recommender Systems with Pseudo Labels Reconstruction (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Large language models (LLMs) have been utilized in various studies, but their training sequences and text labels can alter their pre-trained weights, reducing their ability to construct and comprehend natural language sentences. |
| Approach: | They propose a reconstruction-based LLM recommendation model that harnesses the feature extraction capability of LLMs while preserving LLM’s sentence generation abilities. |
| Outcome: | The proposed model exploits the key features of both user and item pseudo-labels generated from user reviews while training on sequential data. |
Task-aware Block Pruning with Output Distribution Signals for Large Language Models (2026.findings-eacl)
Copied to clipboard
| Challenge: | Existing methods to estimate block importance rely on representation similarity or computationally expensive sensitivity analyses to estimate task-aware model behavior. |
| Approach: | They propose a novel approach that quantifies block-level uncertainty from the statistics of each block’s early-exited output distribution on a calibration dataset. |
| Outcome: | Experiments show that the proposed approach preserves downstream task performance while reducing inference latency and computational cost. |