arXiv subjects

Hyun Woo Kim

Publications and source records attributed to Hyun Woo Kim.

5 recordsLinked to original sources

arXiv2026

Leveraging Biokinetic Knowledge Priors for Data-Scarce Bioprocess Modeling

While deep learning has accelerated drug discovery, its impact on biomanufacturing has been considerably more limited. The reason is data scarcity. Bioreactor experiments are high-cost, take days to weeks, and are rarely shared in public form, leaving each research work with only a handful of experiments. The domain itself, however, is rich in prior knowledge. Biokinetic ordinary differential equation (ODE) models have described microbial growth for decades, yet how to inject this knowledge into a neural network has not been studied systematically. We present the first systematic study of how to inject this ODE knowledge into a neural network, comparing a data-level prior that pre-trains a generic decoder on simulated ODE curves against an architecture-level prior that embeds the ODE inside the decoder. Both consistently outperform no-prior baselines across 11 datasets and 7 microbial species. Our central finding is that the two are substitutable. A generic decoder pre-trained on simulation matches a fully bio-structured decoder trained on real data. Simulation pre-training therefore offers a simple, data-efficient recipe for deep learning under bioprocess data scarcity.

cs.LG↗

arXiv2026

Bayesian E(3)-Equivariant Interatomic Potential with Iterative Restratification of Many-body Message Passing

Machine learning potentials (MLPs) have become essential for large-scale atomistic simulations, enabling ab initio-level accuracy with computational efficiency. However, current MLPs struggle with uncertainty quantification, limiting their reliability for active learning, calibration, and out-of-distribution (OOD) detection. We address these challenges by developing Bayesian E(3) equivariant MLPs with iterative restratification of many-body message passing. Our approach introduces the joint energy-force negative log-likelihood (NLL$_\text{JEF}$) loss function, which explicitly models uncertainty in both energies and interatomic forces, yielding substantially improved accuracy compared to conventional NLL losses. We systematically benchmark multiple Bayesian approaches, including deep ensembles with mean-variance estimation, stochastic weight averaging Gaussian, improved variational online Newton, and Laplace approximation by evaluating their performance on uncertainty prediction, OOD detection, calibration, and active learning tasks. We further demonstrate that NLL$_\text{JEF}$ facilitates efficient active learning by quantifying energy and force uncertainties. Using Bayesian active learning by disagreement (BALD), our framework outperforms random sampling and energy-uncertainty-based sampling. Our results demonstrate that Bayesian MLPs achieve competitive accuracy with state-of-the-art models while enabling uncertainty-guided active learning, OOD detection, and energy/forces calibration. This work establishes Bayesian equivariant neural networks as a powerful framework for developing uncertainty-aware MLPs for atomistic simulations at scale.

cs.LG↗

arXiv2017

THz-Pump and X-Ray-Probe Sources Based on an Electron Linac

We describe a compact THz-pump and X-ray-probe beamline, based on an electron linac, for ultrafast time-resolved diffraction applications. Two high-energy electron ($γ>50$) bunches, $5~$ns apart, impinge upon a single-foil or a multifoil radiator and generate THz radiation and X-rays simultaneously. The THz pulse from the first bunch is synchronized to the X-ray beam of the second bunch by using an adjustable optical delay of THz pulse. The peak power of THz radiation from the multifoil radiator is estimated to be $0.14~$GW for a $200~$pC well-optimized electron bunch. GEANT4 simulations show a carbon foil with thickness of $0.5~-~1.0~$mm has the highest yield of $10~-~20~$keV hard X-rays for a $25~$MeV beam, which is approximately $10^3$ photons/(keV pC-electrons) within a few degrees of the polar angle. A carbon multifoil radiator with $35$ foils ($25~$$μm~$thick each) can generate close to $10^3$ hard X-rays/(keV pC-electrons) within a $2^\circ$ acceptance angle. With $200~$pC charge and $100~$Hz repetition rate, we can generate $10^7$ X-rays per $1~$keV energy bin per second or $10^5$ X-rays per $1~$keV energy bin per pulse. The longitudinal time profile of X-ray pulse ranges from $400~-~600~$fs depending on the acceptance angle. The broadening of the time duration of X-ray pulse is observed owing to its diverging effect. A double-crystal monochromator (DCM) will be used to select and transport the desired X-rays to the sample. The heating of the radiators by an electron beam is negligible because of the low beam current.

physics.acc-ph↗

arXiv2016

Beam Characterization at the KAERI UED Beamline

The UED (ultrafast electron diffraction) beamline of the KAERI's (the Korea Atomic Energy Research Institute's) WCI (World Class Institute) Center has been successfully commissioned. We have measured the beam emittance by using the quadrupole scan technique and the charge by using a novel measurement system we have developed. In the quadrupole scan, a larger drift distance between the quadrupole and the screen is preferred because it gives a better thin-lens approximation. A high bunch-charge beam, however, will undergo emittance growth in the long drift caused by the space-charge force. We present a method that mitigates this growth by introducing a quadrupole scan with a short drift and without using the thin-lens approximation. The quadrupole in this method is treated as a thick lens, and the emittance is extracted by using the thick-lens equations. Apart from being precise, our method can be readily applied without making any change to the beamline and has no need for a big drift space. For charge measurement, we have developed a system consisting of an in-air Faraday cup (FC) and a preamplifier. Tests performed utilizing 3.3-MeV electrons show that the system was able to measure bunches with pulse durations of tens of femtoseconds at 10 fC sensitivity.

physics.acc-ph↗

arXiv2016

Soft X-ray Absorption Spectroscopy Study of Multiferroic Bi-substituted Ba(1-x)Bi(x)Ti(0.9)Fe(0.1)O(3)

The electronic structures of multiferroic oxides of Ba(1-x)Bi(x)Ti(0.9)Fe(0.1)O(3) (0 < x < 0.12) have been investigated by employing photoemission spectroscopy and soft x-ray absorption spectroscopy (XAS). The measured Fe and Ti 2p XAS spectra show that Ti ions are in the Ti4+ states for all x and that Fe ions are Fe2+-Fe3+ mixed-valent for x > 0. The valence states of Fe ions are found to be nearly trivalent for x=0, and decreases with increasing x from being nearly trivalent (v(Fe)~ 3) for x=0 to v(Fe)~ 2.6 for x=0.12. The valence states of both Ti and Ba ions do not change for all x < 0.12. Based on the obtained valence states of Fe ions, the electronic and magnetic properties of Ba(1-x)Bi(x)Ti(0.9)Fe(0.1)O(3) are explored.

cond-mat.mtrl-sci↗