All first-author. Journal venues are labeled by index; two manuscripts are in preparation.
Manuscripts in preparation
The Calibration Law: Diagnosing and Repairing Answer-Position Bias in Mixture-of-Experts Vision-Language Models
Indra Imanuel, et al.
In preparation for ICML 2027 · independent project; I proposed and lead the project
MoE vision–language models fail badly at fine-grained visual
discrimination. On the VisMin benchmark we show much of the underlying
failure is a severe answer-position bias, introduce a zero-cost
diagnostic, and establish, across eight open MoE
VLMs, a "calibration law": training-free
interventions succeed in proportion to how well they calibrate
the bias. A label-free recipe greatly improves the overall VisMin
group score and transfers zero-shot to Winoground and ColorSwap.
Second-Pass Document Hierarchy Repair for PDF Extraction (working title)
Indra Imanuel, et al.
In preparation for ICDAR 2027 · Laputa, Inc.; I proposed and lead the project
PDF extractors recover text well and document hierarchy poorly.
This project builds an end-to-end evaluation on three public benchmarks (Comp-HRDoc, HiPS, DocHieNet) and Korean textbooks, a self-supervised pointer model trained on pseudo-labels mined from section numbering and embedded PDF outlines with no human annotation, and LLM/VLM baselines.
Electronics Letters, vol. 58, pp. 47–49, 2022 SCIE, Scopus Q2 · M.Sc. thesis work
A GAN training scheme that applies adversarial loss on feature maps of the generated high-resolution image via pre-trained networks; outperformed the state-of-the-art CVPR method on a real-world low-resolution benchmark at the time of publication.
Theses
Ph.D. — Image Compression and Demosaicing based on Denoising Diffusion Null-Space Model. Dongseo University, 2024. Advisor: Prof. Suk-Ho Lee.
M.Sc. — Super-Resolution with Adversarial Loss on the Feature Maps of the Generated High-Resolution Image. Dongseo University, 2021. Advisor: Prof. Suk-Ho Lee.
B.Comp. — Committee Management System for Dept. of Informatics in University of Surabaya. University of Surabaya, 2019. Advisor: Fitri Dwi Kartikasari, M.Si. & Dr. Daniel Soesanto.
