arXiv subjects

Akiko Masaki

Publications and source records attributed to Akiko Masaki.

3 recordsLinked to original sources

arXiv2025

Dialect Identification Using Resource-Efficient Fine-Tuning Approaches

Dialect Identification (DI) is a task to recognize different dialects within the same language from a speech signal. DI can help to improve the downstream speech related tasks even when speakers have a strong dialect. However, fine-tuning a speech model for tasks like DI is expensive in terms of computation cost and memory requirement. Recent studies have explored fine-tuning pre-trained speech models for tasks like DI using Parameter-Efficient Fine-Tuning (PEFT) methods, which offer parameter efficiency but limited improvement in memory efficiency and training speed. To address these challenges, we explore Memory-Efficient Fine-Tuning (MEFT) methods, originally proposed for language processing, and apply them to the general-purpose pre-trained speech model. We then comprehensively analyze the GPU memory usage and fine-tuning speed based on various MEFT methods. As a case study, we fine-tune the Whisper model to identify six Mandarin subdialects from the KeSpeech dataset, reducing GPU memory usage by up to 73.25% and accelerating training speed by a factor of 2.1, while maintaining accuracy comparable to vanilla fine-tuning and PEFT methods.

cs.CL↗

arXiv2012

Quantum phases of hardcore bosons with long-range interactions on a square lattice

We study the ground-state phase diagrams of hardcore bosons with long-range interactions on a square lattice using the linear spin-wave theory and a cluster mean-field method. Specifically, we consider the two types of long-range interaction: One consists only of the nearest- and next-nearest-neighbor interactions, and the other is the dipole-dipole interaction that decays with the interparticle distance $r$ as $\sim r^{-3}$. It is known from previous analyses by quantum Monte Carlo methods that a checkerboard supersolid (CSS) is absent in the ground-state phase diagram of the former case while it is present in the latter. In the former, we find that quantum fluctuations around mean-field solutions are enhanced by the direct competition between the checkerboard and striped solid orders and that they destabilize the CSS phase. On the other hand, the emergence of the CSS phase in the latter case can be attributed to the absence of such a competition with other solid orders. We also show that the cluster mean-field method allows for the determination of phase boundaries in a precise quantitative manner when scaling with respect to the cluster size is taken into account. It is found that the phase transition between the superfluid and the solid (or CSS) is of the first order in the vicinity of the particle-hole symmetric line.

cond-mat.other↗

arXiv2008

Mott Plateau in Trapped Bose-Fermi Mixture

Boson-Fermion mixture confined on an optical lattice is numerically studied. Mott-plateau like behavior is observed under certain conditions in the density profile of the bosons and fermions as one sees in bosonic atoms in a trap potential. In the Mott plateau area, the total number of the bosons and fermions at each site is kept to an integer value but both local density of the bosons and that of the fermions take some intermediate values. Unlike bosonic or fermionic Mott insulators, the bosons and fermions can move even in the Mott plateau area by exchanging their positions. With Monte Carlo simulations we demonstrate the formation of the Mott plateau, examine the property of the bosons and fermions by measuring the densities and local density fluctuations, and give discussion on this rather new state.

quant-ph↗