Institution profile

University of Ulsan

Academic institutionasia · kr
Official website
Research library9linked papers
Opportunities0open roles
Selected work

Representative Papers

ZeroCode: On-demand Error-Correcting Code Construction from the Zero Matrix via Reinforcement Learning

Sep 28, 2026

This study addresses the challenge of adapting parity-check matrices of error-correcting codes to diverse scenario-specific requirements by proposing a unified construction framework based on reinforcement learning. The matrix generation process is formulated as a discrete sequential decision-making problem, where edges are incrementally added to an all-zero matrix. An action masking mechanism is introduced to flexibly embed structural constraints, enabling the generation of code libraries across multiple complexities through a single policy rollout without retraining. Experimental results demonstrate that for (32,16) short codes at a bit error rate of $10^{-4}$, the proposed method achieves approximately 1 dB gain over existing genetic, differentiable, and classical approaches, thereby realizing on-demand design of high-performance error-correcting codes.

0 citationsRead paper

Uni-SFU: Algorithm-HW Co-Design for Universal SFUs via Mixed-Degree Piecewise Approximation

Aug 11, 2026

This work addresses the challenge of balancing computational efficiency, area overhead, and numerical accuracy in hardware implementations of activation functions. The authors propose a novel algorithm-hardware co-design framework that introduces, for the first time, a cross-activation mixed-order, non-uniform piecewise polynomial approximation strategy. This approach is jointly optimized with an RTL-level area cost model to achieve high accuracy and hardware reuse under a unified configuration. Experimental results demonstrate that the proposed design attains a mean squared error below 8.22×10⁻⁸ and incurs no more than a 1.02% Top-1 accuracy loss across over 700 neural network variants and three NLP models. Implemented in 22 nm technology, the hardware footprint is only 6,800 μm².

0 citationsRead paper
Recent publications

Latest Papers

ZeroCode: On-demand Error-Correcting Code Construction from the Zero Matrix via Reinforcement Learning

Sep 28, 2026

This study addresses the challenge of adapting parity-check matrices of error-correcting codes to diverse scenario-specific requirements by proposing a unified construction framework based on reinforcement learning. The matrix generation process is formulated as a discrete sequential decision-making problem, where edges are incrementally added to an all-zero matrix. An action masking mechanism is introduced to flexibly embed structural constraints, enabling the generation of code libraries across multiple complexities through a single policy rollout without retraining. Experimental results demonstrate that for (32,16) short codes at a bit error rate of $10^{-4}$, the proposed method achieves approximately 1 dB gain over existing genetic, differentiable, and classical approaches, thereby realizing on-demand design of high-performance error-correcting codes.

0 citationsRead paper

Uni-SFU: Algorithm-HW Co-Design for Universal SFUs via Mixed-Degree Piecewise Approximation

Aug 11, 2026

This work addresses the challenge of balancing computational efficiency, area overhead, and numerical accuracy in hardware implementations of activation functions. The authors propose a novel algorithm-hardware co-design framework that introduces, for the first time, a cross-activation mixed-order, non-uniform piecewise polynomial approximation strategy. This approach is jointly optimized with an RTL-level area cost model to achieve high accuracy and hardware reuse under a unified configuration. Experimental results demonstrate that the proposed design attains a mean squared error below 8.22×10⁻⁸ and incurs no more than a 1.02% Top-1 accuracy loss across over 700 neural network variants and three NLP models. Implemented in 22 nm technology, the hardware footprint is only 6,800 μm².

0 citationsRead paper