https://scholar.dgist.ac.kr/handle/20.500.11750/91 2026-09-30T23:11:52Z https://scholar.dgist.ac.kr/handle/20.500.11750/60905 Title: Translational Profiling of Drd2-Expressing Populations Reveals Molecular Heterogeneity of Dentate Gyrus Mossy Cells along the Dorsoventral Axis Author(s): Jeong, Minseok; Jang, Jin-Hyeok; Oh, Seo-Jin; Choi, Ji-Woong; Oh, Yong-Seok Abstract: Hilar mossy cells (MCs) are crucial for integrating and propagating signals across the hippocampal dorsoventral axis, mediating cognitive and affective processing. While MCs exhibit profound dorsoventral differences in their projections, physiology, and behavioral roles, the molecular basis underlying this functional specialization remains largely unexplored. To address this gap, we used translating ribosome affinity purification (TRAP) in male mice to systematically compare the translatome of Drd2-expressing, MC-enriched populations along the dorsoventral axis. This analysis revealed distinct translational signatures with 1,442 genes enriched in dorsal and 1,337 genes in ventral Drd2-expressing, MC-enriched populations. Pathway analysis demonstrated significant functional segregation along the dorsoventral axis. The dorsal population is notably enriched for genes linked to neuronal connectivity and synaptic transmission, whereas the ventral counterpart shows enrichment in genes associated with energy metabolism and cellular maintenance. Specifically, we identified a subset of dorsal enriched genes, including neurotransmitter receptors, ion channels, and axon guidance regulators, contrasting with ventral enriched genes highly related to glucose/fatty acid metabolism, oxidative phosphorylation, and exocytosis. We further predicted distinct sets of upstream transcriptional regulators activated in each subpopulation, providing insights into the regulatory networks that may drive molecular divergence. Our findings provide a translatomic basis for the dorsoventral heterogeneity of Drd2-expressing neurons that include MCs, offering molecular signatures associated with their differential contributions to hippocampal function. 2026-06-30T15:00:00Z https://scholar.dgist.ac.kr/handle/20.500.11750/60894 Title: Reinforcement Learning-Based Metro Train Tracking Control to Overcome Input Time Delay Author(s): Lee, Kyungbae; Lee, Seungyeop; Kim, Seunghyeon; Eun, Yongsoon Abstract: Speed tracking and precise stop control of metro trains play a key role in train operation. Reinforcement learning (RL) is one of the methods that can solve train control challenges and adapt to various environments. However, when applying RL to train control, the input time delay of the train presents particular difficulties. This paper proposes a RL design method that integrates prediction techniques and the deep deterministic policy gradient algorithm to overcome the input time delay. The training and performance evaluation of the proposed RL is conducted using the validated Automatic Train Operation (ATO) simulator for Seoul Metro Line 5. The results demonstrate that prediction-based RL (PRL) controllers outperform RL without prediction controllers. Additionally, the PRL controller robustness is verified through performance comparisons with a classical model-based approach across various realistic scenarios. 2026-08-31T15:00:00Z https://scholar.dgist.ac.kr/handle/20.500.11750/60890 Title: High-fidelity transcranial ultrasound multi-focal stimulation via physics-aware hologram technique Author(s): Lee, Moon Hwan; Khan, Mohd Afzal; Ashiquzzaman, Akm; Lee, Eunbin; Lee, Jonghun; Chung, Euiheon; Kwon, Hyuk-Sang; Hwang, Jae Youn Abstract: Introduction: Transcranial ultrasound stimulation (TUS) is an emerging non-invasive neuromodulation modality that offers deep brain access with high spatial precision. However, its broader application is limited by the difficulty of reliably generating complex transcranial acoustic fields, particularly for multi-target stimulation through the skull. These limitations can lead to focal distortion, off-target exposure, and reduced reliability of neuromodulation outcomes. Materials and methods: Here, we introduce a physics-aware thickness-only acoustic hologram (TOAH) technique for precise transcranial ultrasound neuromodulation. Unlike conventional approaches that rely on simplified phase-based approximations, TOAH directly generates fabrication-ready holographic implementations while preserving consistency between numerical field synthesis and physical acoustic realization. This enables accurate formation of single-, dual-, and tri-focal stimulation patterns under transcranial conditions. We validated TOAH through in silico simulations, ex vivo acoustic measurements through skulls, and in vivo experiments. Results: Compared with state-of-the-art methods, TOAH improved focal reconstruction, energy confinement, and multi-focal balance while reducing off-target acoustic leakage. Human-skull simulations further supported robust multi-focal reconstruction under clinically relevant transcranial conditions. In a neuropathic pain mouse model, bilateral thalamic stimulation induced measurable changes in neuronal activity, reflected by reduced c-Fos expression, together with preliminary improvements in pain-related behavioral responses. These findings support the capability of the proposed technique to enable spatially localized and reproducible neuromodulation in vivo. Conclusion: Collectively, this work provides a practical proof-of-concept strategy for achieving high-precision, multi-target transcranial neuromodulation and supports further investigation for neuroscience research and future therapeutic applications. 2026-06-30T15:00:00Z https://scholar.dgist.ac.kr/handle/20.500.11750/60874 Title: Social Reasoning-Aware Trajectory Prediction via Multimodal Language Model Author(s): Bae, Inhwan; Lee, Junoh; Jeon, Hae-Gon Abstract: Recent advancements in language models have demonstrated its capacity of context understanding and generative representations. Leveraged by these developments, we propose a novel multimodal trajectory predictor based on a vision-language model, named VLMTraj, which fully takes advantage of the prior knowledge of multimodal large language models and the human-like reasoning across diverse modality information. The key idea of our model is to reframe the trajectory prediction task into a visual question answering format, using historical information as context and instructing the language model to make predictions in a conversational manner. Specifically, we transform all the inputs into a natural language style: historical trajectories are converted into text prompts, and scene images are described through image captioning. Additionally, visual features from input images are also transformed into tokens via a modality encoder and connector. The transformed data is then formatted to be used in a language model. Next, in order to guide the language model in understanding and reasoning high-level knowledge, such as scene context and social relationships between pedestrians, we introduce an auxiliary multi-task question and answers. For training, we first optimize a numerical tokenizer with the prompt data to effectively separate integer and decimal parts, allowing us to capture correlations between consecutive numbers in the language model. We then train our language model using all the visual question answering prompts. During model inference, we implement both deterministic and stochastic prediction methods through beam-search-based most-likely prediction and temperature-based multimodal generation. Our VLMTrajvalidates that the language-based model can be a powerful pedestrian trajectory predictor, and outperforms existing numerical-based predictor methods. Extensive experiments show that VLMTrajcan successfully understand social relationships and accurately extrapolate the multimodal futures on public pedestrian trajectory prediction benchmarks. 2026-05-31T15:00:00Z