SearcharxivSearch

arXiv subjects

Kentaro Kaba

Publications and source records attributed to Kentaro Kaba.

3 recordsLinked to original sources

Simulation-free and finite-time diffusion model

The performance of generative diffusion models is determined by the choice of the reference diffusion process connecting the empirical and prior distributions. Conventional approaches typically trade off simulation-free training against finite-time generation. We propose a framework for designing the reference process that achieves both simultaneously. The key idea is to prescribe tractable time-dependent conditional distributions and then construct the reference process realizing them as its marginals. This framework reveals that score matching is not fundamental to diffusion-model training but instead emerges naturally through reversal of the reference process. We further show that conditional flow matching arises as the small-noise limit of the proposed framework.

cs.LG

Sch\"odinger Bridge Type Diffusion Models as an Extension of Variational Autoencoders

Generative diffusion models use time-forward and backward stochastic differential equations to connect the data and prior distributions. While conventional diffusion models (e.g., score-based models) only learn the backward process, more flexible frameworks have been proposed to also learn the forward process by employing the Schr\"odinger bridge (SB). However, due to the complexity of the mathematical structure behind SB-type models, we can not easily give an intuitive understanding of their objective function. In this work, we propose a unified framework to construct diffusion models by reinterpreting the SB-type models as an extension of variational autoencoders. In this context, the data processing inequality plays a crucial role. As a result, we find that the objective function consists of the prior loss and drift matching parts.

cs.LG

Possible Existence of Extremely Neutron-Rich Superheavy Nuclei in Neutron Star Crusts Under a Superstrong Magnetic Field

We investigate outer crust compositions for a wide range of magnetic field strengths, up to $B\simeq 4\times10^{18}$ G, employing the latest experimental nuclear masses supplemented with various mass models. The essential effects of the magnetic field are twoholds: 1) Enhancement of electron fraction, which is connected to that of protons via the charge neutrality condition, due to the Landau-Rabi quantization of electron motion perpendicular to the field, namely, neutron-richness is suppressed for a given pressure. As a result, 2) nuclei can exist at higher pressures without dripping out neutrons. By exploring optimal outer-crust compositions from all possible nuclei predicted by theoretical models, we find that neutron-rich heavy nuclei with neutron magic numbers 50, 82, 126, as well as 184, with various proton numbers emerge for $B\gtrsim 10^{17}$ G. Moreover, we show that superheavy nuclei with proton numbers $Z\ge104$, including unknown elements such as $Z=119, 120, 122$, and/or $124$, depending on mass models, may emerge as an equilibrium composition at bottom layers of the outer crust for $B\gtrsim 10^{18}$ G, which are extremely neutron-rich ($N\approx 260$-$287$, i.e., $N/Z\approx 2.2$-$2.4$). We point out that those extremely neutron-rich superheavy nuclei locate around the next neutron magic number $N=258$ after $N=184$, underlining importance of nuclear structure calculations under such really exotic, extreme conditions. We demonstrate how the superstrong magnetic field substantially alters crustal properties of neutron stars, which may have detectable consequences.

nucl-th