정석송 교수

We propose the Augmented State Neural-Transition EKF (ANT-EKF), a hybrid filtering framework that embeds a learnable neural transition model inside the extended Kalman filter. By augmenting the state with the neural network’s parameters, ANT-EKF jointly tracks the physical state and adapts the dynamics model online using Jacobians obtained via autograd. We instantiate the method for (i) object trajectory tracking and (ii) RIS-assisted mobile channel tracking, where the measurement remains linear

Abstract Background: Invasive lobular carcinoma (ILC) is typically characterized by lobular morphology, CDH1 inactivation and reduced E-cadherin expression. However, discordance between morphologic and immunohistochemical (IHC) findings often complicates diagnosis. To address this issue, we performed whole transcriptome sequencing (WTS) on selected breast cancer cases exhibiting variable E-cadherin IHC patterns and morphology, including cases with and without CDH1 mutations identified by prior t