https://scholar.dgist.ac.kr/handle/20.500.11750/1194 2026-08-25T22:44:00Z https://scholar.dgist.ac.kr/handle/20.500.11750/60671 Title: Hyperbolic Prototype-Residual Autoencoder for Interpretable Generative Latent Organization Author(s): Lee, Hyunjong; Kwak, Jeongho Abstract: Interpretable organization of latent spaces remains a central challenge in generative modeling. Most generative models rely on continuous latent variables, but the learned space often lacks an explicit structure that explains how samples are organized or how semantic variation can be controlled. This paper proposes a hyperbolic prototype-residual autoencoder that organizes generative latent representations using a fixed prototype tree embedded in the Poincar & eacute; ball. Each encoded sample is assigned to a prototype by hyperbolic distance, and the decoder reconstructs or generates images from a prototype-residual representation. The prototype serves as a semantic anchor, while the residual captures local instance-level variation around the selected prototype. The framework combines prototype semantic learning, MMD-based latent spreading, and structural regularizers to align encoder-decoder behavior with the predefined hyperbolic hierarchy. Prototype semantic learning encourages fixed prototypes to decode into representative visual anchors, while MMD encourages encoded samples to occupy broad regions of the hyperbolic latent space. Experiments on MNIST and CelebA show interpretable coarse-to-fine behavior through radial decoding and prototype decoding, suggesting that fixed hyperbolic prototype trees provide an effective scaffold for prototype-guided image reconstruction and generation. 2026-07-31T15:00:00Z https://scholar.dgist.ac.kr/handle/20.500.11750/60670 Title: LanthanideDoping and Dual-Site Switching on AmorphousHigh-Entropy Metallene Oxides Boost Acidic Water Oxidation Author(s): Li, Yinghao; Sun, Yuntong; Chen, Mengshan; Zhou, Yingtang; Fan, Wenjun; Lee, Jong-Min Abstract: High-entropy oxides (HEOs) are promising electrocatalysts for breaking the Sabatier principle and maximizing their oxygen evolution reaction (OER) efficiency. Herein, the effect of lanthanide doping on RuIr-based HEOs is investigated through theoretical and experimental studies. The optimized amorphous RuIrMnCoGd high-entropy metallene oxides (a-Gd_HEMOs) with moderate intermediate affinity exhibit a small overpotential of 211 mV and an ultralong stability of 700 h at 10 mA cm(-2) in acidic media. Integrated into a proton exchange membrane electrolyzer, they deliver 3.0 A cm(-2) at 1.872 V and maintain stable operation for 300 h at 1.0 A cm(-2). Operando characterizations reveal the changes in electronic configurations and coordination environments of Ru and Ir sites during OER. Theoretical simulations further clarify the adsorbate evolution mechanism and the distinct role of the constituent elements. Specifically, the switching of Ru and activated Ir as primary active sites at low and high potentials, respectively, enhances robustness for oxygen generation. 2026-07-31T15:00:00Z https://scholar.dgist.ac.kr/handle/20.500.11750/60630 Title: Ultrafast Multilevel Switching and Synaptic Behavior in a Planar Quantum Topological Memristor Author(s): Rashid, Mamoon Ur; Safder, Usman; Khan, Sobia Ali; Pham, Anh-Tuan; Sheeraz, Muhammad; Dang, Nguyen-Hoang; Thanh, Duy Le; Tahir, Zeeshan; Maqbool, Faisal; Chung, Koo-Hyun; Cho, Sunglae; Kim, Jungdae; Kim, Yong Soo Abstract: The rapid increase in data driven by analytics and Internet of Things demands innovation in both device architecture and materials to meet the growing need for fast and efficient computing. Here we report an ultrafast planar quantum topological memristor (PQTM), comprised of bismuth-telluride (Bi2Te3) thin film transferred onto pre-patterned electrodes. Owing to the planar architecture, the device connects both electrodes to the surface states of Bi2Te3, offering a platform to directly benefit from the characteristic features of topological surface states, such as low-dissipation and scattering-resistant channels essential for ultrafast- and efficient-charge transport. Pertinently, PQTM presents a forming-free bipolar-resistive switching behavior with an ultrafast-switching similar to 15 +/- 5 ns and low-energy consumption similar to 14.5 nJ, which is a record high among the topological insulator-based memristors. Moreover, the endurance evaluation over 103 consecutive DC-switching cycles demonstrates superior stability in both high and low resistive states, while the retention tests display an excellent longevity of similar to 105 s, signifying reliable non-volatile operation. Finally, PQTM reproducibility is established via comparison with 24 other devices, presenting multilevel resistive switching exhibiting both digital and analog switching modes together with long-term potentiation, depression, and persistent image-recognition performance, corroborated via 1D-convolutional layers with four LeNet models. Thus, our work emphasizes the critical role of device architecture in harnessing material properties for advanced-memory and neuromorphic applications. 2026-03-31T15:00:00Z https://scholar.dgist.ac.kr/handle/20.500.11750/60629 Title: Serpina1e mediates the exercise-induced enhancement of hippocampal memory in male mice Author(s): Kim, Hyunyoung; 대학원 비전임; Han, Jeongho; Yun, Kangseok; Kim, Jong-Seo; Park, Hyungju Abstract: Exercise enhances learning and memory, not only through improved cardiometabolic but also through body-brain interactions mediated by secreted factors. Given the prominent role of skeletal muscle during exercise, muscle-derived factors, myokines, are believed to mediate the exercise-induced cognitive enhancements. Here, we demonstrate that intramuscular Serpina1e is upregulated following exercise in male mice. Systemic delivery of recombinant Serpina1e or intramuscular overexpression of Serpina1e reproduces exercise-induced memory enhancements in sedentary male mice. Conversely, muscle-specific depletion of Serpina1e abolishes hippocampal memory enhancement, indicating a requirement of muscle-derived Serpina1e for these cognitive benefits. Mechanistically, elevated plasma Serpina1e stimulates neurogenesis, brain-derived neurotrophic factor (BDNF) expression, and neurite growth in the hippocampus by crossing the blood-cerebrospinal fluid (CSF) and blood-brain barrier. Our findings identify Serpina1e as a key mediator of skeletal muscle-brain interaction that enables the beneficial effects of exercise on cognitive function. 2026-03-31T15:00:00Z