One paper accepted to ICCD 2025

One papers was accepted to the International Conference on Computer Design (ICCD) 2025. Our paper proposes FINEA, an efficient neural network architecture exploiting factorized input features. FINEA leverages factorized dot-product operations by exploiting weight redundancy in quantized neural network models. FINEA employs a processing engine that can excutes both factorized and unfactorized dot-product operations. In order to implement efficient factorized and unfactorized dot-product computations, FINEA utilizes filter indexes derived from preprocessed weight tables.

Yujin Kim
Yujin Kim
PhD Student
Chanhun Jeong
Chanhun Jeong
Master Student
Gunjae Koo
Gunjae Koo
Associate Professor