https://scholar.dgist.ac.kr/handle/20.500.11750/91 2026-10-22T09:15:31Z https://scholar.dgist.ac.kr/handle/20.500.11750/61013 Title: Optically bandlimited coherent lateral resampling for optical coherence tomography Author(s): Lee, Changhoe; Um, Taeguk; Jeong, Sunhong; Yang, Dongjae; Oh, Wang-Yuhl; Lee, Byungkun Abstract: We present an optically bandlimited coherent lateral resampling framework for densely and nonuniformly sampled optical coherence tomography (OCT). Fourier-domain group-delay stabilization is combined with bandlimited DFT/NUDFT reconstruction and evaluated using retinal OCT acquired at a 4.73 MHz A-scan rate. When the target-grid Nyquist remains at or above the retained optical cutoff, reconstruction density can be selected retrospectively without changing the retained spectral support. Experiments demonstrate alias suppression, correction of nonuniform scan distortions, and improved SNR associated with dense acquisition relative to the retained optical bandwidth. The framework enables densely sampled high-speed OCT data to be acquired once and subsequently reconstructed at application-specific lateral sampling densities while explicitly controlling the retained spatial-frequency bandwidth and preserving the SNR benefits of oversampled acquisition. 2026-09-30T15:00:00Z https://scholar.dgist.ac.kr/handle/20.500.11750/61003 Title: Thalamofrontal synaptic weakening underlies short-term memory deficits from adolescent NMDAR hypofunction Author(s): Yu, Jinseon; Choi, In Sun; Kim, Gyu Hyun; Bahn, Sangkyu; Kim, Jinmo; Bae, Sungwon; Lee, Taekwan; Choi, Joon Ho; Kwon, Jun Soo; Kim, Minah; Choi, Ji-Woong; Lee, Kea Joo; Rah, Jong-Cheol Abstract: Thalamofrontal (TF) dysconnectivity is one of the most consistent circuit-level abnormalities reported in patients with schizophrenia and is linked to deficits in short-term memory (STM). However, the biological mechanisms underlying TF weakening remain unclear. Here, we show that repeated adolescent N-methyl-d-aspartate receptor (NMDAR) antagonism produces STM deficits by impairing TF synaptic transmission. In mice repeatedly exposed to ketamine, STM impairment coincided with reduced release probability and attenuated short-term depression at mediodorsal thalamus (MD) -> dorsomedial prefrontal cortex (dmPFC) synapses, without detectable changes in corticocortical synaptic release probability, intrinsic excitability, or gross synaptic ultrastructure. These presynaptic deficits were accompanied by diminished direction-selective population coding in the dmPFC and impaired delayed alternation performance. Chemogenetically strengthening MD -> dmPFC projections restored both neural selectivity and behavior. These findings identify a projection-specific presynaptic mechanism through which adolescent NMDAR hypofunction weakens TF communication and produces cognitive impairment, offering a biological explanation for clinically observed TF dysconnectivity and establishing TF synapses as a therapeutic target. 2026-08-31T15:00:00Z https://scholar.dgist.ac.kr/handle/20.500.11750/60985 Title: Twin-mediated orientation engineering for high-mobility Ge thin film via laser crystallization Author(s): Jeong, Heejae; Kwon, Hyojeong; Won, Jiwoo; Choi, Woong; Kwon, Hyuk-Jun Abstract: High-crystallinity germanium (Ge) thin films are crucial for advanced electronic applications, particularly for monolithic three-dimensional (M3D) integration under strict thermal budget constraints. However, conventional solid-phase crystallization (SPC) typically yields low crystallinity and randomly oriented grains. In this study, Ge thin films with pronounced orientation toward (1 1 0) are achieved by optimizing the power of continuous-wave (CW) laser crystallization to 1.8 W. This orientation was accompanied by an increased density of {1 1 1}-related twin traces and stable directional growth along the scan direction, suggesting a twin-related contribution to the orientation evolution during laser crystallization. The optimized Ge thin films exhibit improved crystallinity and reduced tensile stress compared to those crystallized via SPC or under non-optimized laser conditions. Hall-effect measurements indicate that the (1 1 0)-oriented Ge thin films achieve a low carrier concentration (2.21 x 1016 cm-3) and a high mobility (552 cm2V-1s-1), comparing favorably with many previously reported Ge thin films. Notably, among Ge films with comparable grain sizes, the (1 1 0)-oriented Ge thin films show significantly improved crystallinity and electrical properties. This suggests that grain size is not the only factor influencing electrical behavior and that crystallographic orientation substantially contributes to the observed performance enhancement. These findings provide a practical foundation for laser-based orientation engineering of Ge electronic materials for M3D integrated device applications. 2026-08-31T15:00:00Z https://scholar.dgist.ac.kr/handle/20.500.11750/60984 Title: Optimized layerwise approximation for efficient private inference on fully homomorphic encryption Author(s): Lee, Junghyun; Lee, Joon-Woo; Lee, Eunsang; Kim, Young-Sik; Lee, Yongwoo; Kim, Yongjune; No, Jong-Seon Abstract: Recent studies have explored the deployment of privacy-preserving deep neural networks utilizing homomor phic encryption (HE), especially for private inference (PI). Many works have attempted the approximation-aware training (AAT) approach in PI, changing the activation functions of a model to low-degree polynomials that are easier to compute on HE by allowing model retraining. However, due to constraints in the training environ ment, it is often necessary to consider post-training approximation (PTA), using the pre-trained parameters of the existing plaintext model without retraining. Existing PTA studies have uniformly approximated the activation function in all layers to a high degree to mitigate accuracy loss from approximation, leading to significant time consumption. This study proposes an optimized layerwise approximation (OLA), a systematic framework that opti mizes both accuracy loss and time consumption by using different approximation polynomials for each layer in the PTA scenario. For efficient approximation, we reflect the layerwise impact on the classification accuracy by considering the actual input distribution of each activation function while constructing the optimization problem. Additionally, we provide a dynamic programming technique to solve the optimization problem and achieve the optimized layerwise degrees in polynomial time. As a result, we successfully approximated the ReLU and GELU functions, significantly reducing time latency while maintaining classification performance. Especially, the OLA method reduces inference times for the ResNet-20 model and the ResNet-32 model by 3.02 times and 2.82 times, respectively, compared to prior state-of-the-art implementations employing uniform degree polynomials. 2026-08-31T15:00:00Z