Research trajectory, record, and how Yunsu Kim reports numbers.
I am a Ph.D. student in the Department of Information and Communication Engineering at Changwon National University, in SPAI-Lab.
The publications and the repositories on this site look like two different research programmes. They are one, and the through-line is measurement.
I came up through signal processing, where the object of study is a physical process you can only reach through an instrument. Active sonar target classification, then generative synthesis of sonar signals, then machine fault diagnosis from vibration, then horizon-adaptive forecasting for coastal metocean data. Different domains, one habit: the reading an instrument produces is not the quantity being measured, and the gap between them is where the work is.
Agentic systems break that habit by default. When a language model writes the code that computes a number, the number it reports is that code’s testimony about a measurement, not the measurement. The pipeline downstream has no way to tell the difference, and neither does the person reading the paper it writes.
So the current work is instrumentation: building the fixed layer that separates what a system reports from what can be independently checked, and documenting — stage by stage — what happens in a research pipeline when that layer is missing.
physical signals → generative modelling → time-series prediction → autonomous research systems → verification instrumentation
Ph.D. Candidate, Information and Communication Engineering @ Changwon National University · Changwon, South Korea
M.S. in Information and Communication Engineering @ Changwon National University · Changwon, South Korea
B.S. in Information and Communication Engineering @ Changwon National University · Changwon, South Korea
Course Instructor @ Industry-Academic Convergence Technology Education Center, Changwon National University · Changwon, South Korea
Metaverse Platform Development and AI Technology Training — fundamentals of programming, machine learning and deep learning in Python.
Course Instructor @ Industry-Academic Convergence Technology Education Center, Changwon National University · Changwon, South Korea
Smart Factory Supplier IT Specialist Training Program — fundamentals of programming, machine learning and deep learning in Python.
Course Instructor @ Industry-Academic Convergence Technology Education Center, Changwon National University · Changwon, South Korea
Fundamentals of Programming: C++.
Three rules, because the subject matter demands them.
Figures that cannot be recomputed are marked as such, not removed. Several numbers I published in an earlier write-up did not survive a later recount. The originals are still in the repository with the recount beside them, because silently correcting a figure destroys the only thing that makes a correction meaningful — a later reader’s ability to see that it happened.
Project pages state what can be verified, including when the answer is “not yet.” If a tool has not been carried through a real research cycle, its page says so. An empty verification line is information.
Claims are scoped to the evidence. One system and one operator is an existence proof, not a rate. Where a number describes my own archive rather than agent systems in general, the page says which.
There is at least one other researcher publishing in AI under the same name, working on machine translation and NLP. These are mine.