Haena Song, a master’s student in POSTECH’s Department of Electrical Engineering (coauthors: Dohun Kim, Jongho Yoon, and Eunji Kwon; advisors: Professors Seokhyeong Kang and Tae-Hyun Oh), won Bronze in the Computer Science & Engineering category at the 29th Samsung Humantech Paper Awards, receiving KRW 5 million.

Samsung Electronics holds the competition to identify and nurture outstanding scientific and technological talent and establish a foundation for world-leading competitiveness and technological capabilities.

The award-winning paper, “FPGA-Based Accelerator for Rank-Enhanced and Highly-Pruned Block-Circulant Neural Networks,” addressed limitations of block-circulant neural networks, a deep learning network compression technique, and proposed a more efficient compression method and accelerator hardware design.

The study interpreted the accuracy degradation of existing methods from a new perspective: rank deficiency in circulant matrices. To address this, it proposed a framework combining Hadamard parameterization with pruning.

The resulting method substantially improved accuracy over previous approaches while maintaining strong deep learning network compression. Its excellence was recognized with the Bronze award.