Yu-Cheng Lin
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Yu-Cheng Lin has a total of 24 co-authors. Here are the top co-authors by number of shared publications:
- Hung-Ying Chu
Dept. of Computer Science & Engineering, Yuan Ze University, Taoyuan, Taiwan
3 shared publications - Shao-Yu Wei
Dept. of Computer Science & Engineering, Yuan Ze University, Taoyuan, Taiwan
3 shared publications - Cheng-Yang Tsai
Department of Computer Science and Engineering_Yuan Ze University_Electronic Design Automation (EDA) Laboratory, Department of Computer Science and Engineering, Yuan Ze University, Taoyuan, Taiwan, Taiwan
2 shared publications - Guan-Wei Chen
Dept. of Computer Science & Engineering, Yuan Ze University, Taoyuan, Taiwan
2 shared publications - I-Hsiang Wang
Department of Computer Science and Engineering_Yuan Ze University_Machine Intelligence & Automation Technology (MIAT) Laboratory, Department of Computer Science and Engineering, Yuan Ze University, Taoyuan, Taiwan, Taiwan
2 shared publications - Jen-Wei Shih
Department of Computer Science and Engineering_Yuan Ze University_Electronic Design Automation (EDA) Laboratory, Department of Computer Science and Engineering, Yuan Ze University, Taoyuan, Taiwan, Taiwan
2 shared publications
Latest Publications
- Open Access
- 17 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 366-371
Evaluating GNN-Based Initialization for Traditional Optimization Algorithms in Routing-Aware Early Floorplanning
- 17 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 366-371
Open AccessEvaluating GNN-Based Initialization for Traditional Optimization Algorithms in Routing-Aware Early Floorplanning
- Open Access
- 17 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 351-356
Parameter-Efficient Fine-Tuning for Functional RTL Debugging with Large Language Models
- 17 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 351-356
Open AccessParameter-Efficient Fine-Tuning for Functional RTL Debugging with Large Language Models
- Open Access
- 17 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 138-143
Segmentation Granularity in Low-Resource Symbolic Music Generation: A Jiangnan Music Case Study
Symbolic music generation for low-resource cultural repertoires remains challenging because limited corpora provide insufficient evidence for both local musical transitions and long-range structural coherence. This paper examines how segmentation granularity affects an LLM-based symbolic music generation pipeline for Jiangnan music. To address data sparsity, a core corpus of 213 scores was expanded to 6,390 sequences via Markov Chains [1] within the YNote [1] and HNote [2] frameworks. By fine-tuning LLaMA 3.1 via the Unsloth framework, we evaluate the trade-offs between rigid metrical partitioning and semantic melodic pausing. Phrase-level segmentation provides stronger local pitch-transition and duration-distribution alignment at small and medium scales, whereas bar-level segmentation shows more stable pitch-distribution and contour behavior across scales. These findings suggest that segmentation granularity should be selected according to the target objective, such as expressive local style modeling or structurally robust symbolic generation.
- 17 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 138-143
Open AccessSegmentation Granularity in Low-Resource Symbolic Music Generation: A Jiangnan Music Case Study
Symbolic music generation for low-resource cultural repertoires remains challenging because limited corpora provide insufficient evidence for both local musical transitions and long-range structural coherence. This paper examines how segmentation granularity affects an LLM-based symbolic music generation pipeline for Jiangnan music. To address data sparsity, a core corpus of 213 scores was expanded to 6,390 sequences via Markov Chains [1] within the YNote [1] and HNote [2] frameworks. By fine-tuning LLaMA 3.1 via the Unsloth framework, we evaluate the trade-offs between rigid metrical partitioning and semantic melodic pausing. Phrase-level segmentation provides stronger local pitch-transition and duration-distribution alignment at small and medium scales, whereas bar-level segmentation shows more stable pitch-distribution and contour behavior across scales. These findings suggest that segmentation granularity should be selected according to the target objective, such as expressive local style modeling or structurally robust symbolic generation.
- Open Access
- 17 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 124-130
Toward Fair Comparison of Symbolic Music Generators: A Unified YNote Protocol for Jiangnan-Style Music with GNNs and LLMs
Fair comparisons between symbolic music generators based on large language models (LLMs) and graph neural networks (GNNs) are often hindered by discrepancies in representation choices and task formulations. To address this issue, we propose a unified evaluation protocol for Jiangnan-style symbolic music generation by adopting YNote as a shared event representation and defining a bar-level boundary-completion task. Each 4/4 bar is quantized into discrete temporal units, with two boundary anchors specified: the first non-rest event and the final event of the bar. Given identical inputs and constraints, models are required to generate intermediate events that exactly fill the temporal gap while remaining consistent with both boundary conditions. Under this unified setting, we conduct a systematic comparison between a GraphSAGE-based GNN and an LLaMA-3.1-based LLM. The evaluation includes distributional consistency for pitch and duration, local transition modeling using bigram KL divergence and bigram coverage, diversity measured by Self-BLEU, and validity and runtime efficiency. Experimental results indicate that the GNN produces more stable and less repetitive outputs, whereas the LLM achieves higher bigram coverage but exhibits greater variability as the generation scale increases.
- 17 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 124-130
Open AccessToward Fair Comparison of Symbolic Music Generators: A Unified YNote Protocol for Jiangnan-Style Music with GNNs and LLMs
Fair comparisons between symbolic music generators based on large language models (LLMs) and graph neural networks (GNNs) are often hindered by discrepancies in representation choices and task formulations. To address this issue, we propose a unified evaluation protocol for Jiangnan-style symbolic music generation by adopting YNote as a shared event representation and defining a bar-level boundary-completion task. Each 4/4 bar is quantized into discrete temporal units, with two boundary anchors specified: the first non-rest event and the final event of the bar. Given identical inputs and constraints, models are required to generate intermediate events that exactly fill the temporal gap while remaining consistent with both boundary conditions. Under this unified setting, we conduct a systematic comparison between a GraphSAGE-based GNN and an LLaMA-3.1-based LLM. The evaluation includes distributional consistency for pitch and duration, local transition modeling using bigram KL divergence and bigram coverage, diversity measured by Self-BLEU, and validity and runtime efficiency. Experimental results indicate that the GNN produces more stable and less repetitive outputs, whereas the LLM achieves higher bigram coverage but exhibits greater variability as the generation scale increases.
- Open Access
- 4 February 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-8
An Automated and Scalable CNN-Based Framework for APB Error Diagnosis in SoC Designs
- 4 February 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-8
Open AccessAn Automated and Scalable CNN-Based Framework for APB Error Diagnosis in SoC Designs
- Open Access
- 20 December 2025
- Published by Association for Computing Machinery (ACM)
- p. 448-454
A Multi-stage Error Diagnosis for APB Transaction
- 20 December 2025
- Published by Association for Computing Machinery (ACM)
- p. 448-454
Open AccessA Multi-stage Error Diagnosis for APB Transaction
- Open Access
- 20 December 2025
- Published by Association for Computing Machinery (ACM)
- p. 407-415
Large Language Models for 3D IC Space Planning
- 20 December 2025
- Published by Association for Computing Machinery (ACM)
- p. 407-415
Open AccessLarge Language Models for 3D IC Space Planning
- Open Access
- 14 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 91642-91654
Automatic Timbre Transformation Using Enhanced Diffusion Model
We present a novel timbre transfer model that uses an enhanced diffusion architecture to convert music from various instruments into Erhu timbre. The Erhu, a traditional Chinese instrument, is difficult to simulate due to its rich vibrato and smooth note transitions. Existing Musical Instrument Digital Interface systems struggle to capture its nuanced dynamics. Our model integrates a Pitch Encoder, a Loudness Encoder, and a Diffusion Decoder. The encoders extract pitch features and dynamic loudness variations, guiding the decoder in generating realistic Erhu timbre. By extracting general musical features, the system generalizes to unseen input types without retraining. Evaluations based on pitch accuracy, cosine similarity, and Fréchet Audio Distance show that our model achieves 96% pitch accuracy and high fidelity in Erhu timbre reproduction. This study demonstrates the potential of diffusion-based timbre transfer models in music generation and provides new directions for future work on both music generation and timbre transfer.
- 14 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 91642-91654
Open AccessAutomatic Timbre Transformation Using Enhanced Diffusion Model
We present a novel timbre transfer model that uses an enhanced diffusion architecture to convert music from various instruments into Erhu timbre. The Erhu, a traditional Chinese instrument, is difficult to simulate due to its rich vibrato and smooth note transitions. Existing Musical Instrument Digital Interface systems struggle to capture its nuanced dynamics. Our model integrates a Pitch Encoder, a Loudness Encoder, and a Diffusion Decoder. The encoders extract pitch features and dynamic loudness variations, guiding the decoder in generating realistic Erhu timbre. By extracting general musical features, the system generalizes to unseen input types without retraining. Evaluations based on pitch accuracy, cosine similarity, and Fréchet Audio Distance show that our model achieves 96% pitch accuracy and high fidelity in Erhu timbre reproduction. This study demonstrates the potential of diffusion-based timbre transfer models in music generation and provides new directions for future work on both music generation and timbre transfer.
- Open Access
- 18 December 2023
- Published by Warsaw University of Life Sciences - SGGW Press in Machine Graphics and Vision
- Vol. 32 (3/4) , 125-145
Rule-based explaining module: Enhancing the interpretability of recurrent relational network in Sudoku solving
Computer vision has gained significant attention in the field of information technology due to its widespread application that addresses real-world challenges, surpassing human intelligence in tasks such as image recognition, classification, natural language processing, and even game playing. Sudoku, a challenging puzzle that has captivated many people, exhibits a complexity that has attracted researchers to leverage deep learning techniques for its solution. However, the reliance on black-box neural networks has raised concerns about transparency and explainability. In response to this challenge, we present the Rule-based Explaining Module (REM), which is designed to provide explanations of the decision-making processes using Recurrent Relational Networks (RRN). Our proposed methodology is to bridge the gap between complex RRN models and human understanding by unveiling the specific rules applied by the model at each stage of the Sudoku puzzle solving process. Evaluating REM on the Minimum Sudoku dataset, we achieved an accuracy of over 98.00%.
- 18 December 2023
- Published by Warsaw University of Life Sciences - SGGW Press in Machine Graphics and Vision
- Vol. 32 (3/4) , 125-145
Open AccessRule-based explaining module: Enhancing the interpretability of recurrent relational network in Sudoku solving
Computer vision has gained significant attention in the field of information technology due to its widespread application that addresses real-world challenges, surpassing human intelligence in tasks such as image recognition, classification, natural language processing, and even game playing. Sudoku, a challenging puzzle that has captivated many people, exhibits a complexity that has attracted researchers to leverage deep learning techniques for its solution. However, the reliance on black-box neural networks has raised concerns about transparency and explainability. In response to this challenge, we present the Rule-based Explaining Module (REM), which is designed to provide explanations of the decision-making processes using Recurrent Relational Networks (RRN). Our proposed methodology is to bridge the gap between complex RRN models and human understanding by unveiling the specific rules applied by the model at each stage of the Sudoku puzzle solving process. Evaluating REM on the Minimum Sudoku dataset, we achieved an accuracy of over 98.00%.