Search

This study presents the first report on the development of an artificial intelligence (AI) for automatic region segmentation of four-dimensional computer tomography (4D-CT) images during swallowing. The material consists of 4D-CT images taken during swallowing. Additionally, data for verifying the practicality of the AI were obtained from 4D-CT images during mastication and swallowing. The ground truth data for the region segmentation for the AI were created from five 4D-CT datasets of swallowing. A 3D convolutional model of nnU-Net was used for the AI. The learning and evaluation method for the AI was leave-one-out cross-validation. The number of epochs for training the nnU-Net was 100. The Dice coefficient was used as a metric to assess the AI's region segmentation accuracy. Regions with a median Dice coefficient of 0.7 or higher included the bolus, bones, tongue, and soft palate. Regions with a Dice coefficient below 0.7 included the thyroid cartilage and epiglottis. Factors that reduced the Dice coefficient included metal artifacts caused by dental crowns in the bolus and the speed of movement for the thyroid cartilage and epiglottis. In practical verification of the AI, no significant misrecognition was observed for facial bones, jaw bones, or the tongue. However, regions such as the hyoid bone, thyroid cartilage, and epiglottis were not fully delineated during fast movement. It is expected that future research will improve the accuracy of the AI's region segmentation, though the risk of misrecognition will always exist. Therefore, the development of tools for efficiently correcting the AI's segmentation results is necessary. AI-based visualization is expected to contribute not only to the deepening of motion analysis of organs during swallowing but also to improving the accuracy of swallowing CT by clearly showing the current state of its precision., Comment: 8 pages, 5 figures, 1 table

Multi-channel acoustic signal processing is a well-established and powerful tool to exploit the spatial diversity between a target signal and non-target or noise sources for signal enhancement. However, the textbook solutions for optimal data-dependent spatial filtering rest on the knowledge of second-order statistical moments of the signals, which have traditionally been difficult to acquire. In this contribution, we compare model-based, purely data-driven, and hybrid approaches to parameter estimation and filtering, where the latter tries to combine the benefits of model-based signal processing and data-driven deep learning to overcome their individual deficiencies. We illustrate the underlying design principles with examples from noise reduction, source separation, and dereverberation., Comment: Accepted for IEEE Signal Processing Magazine

This paper addresses a new virtual try-on problem of fitting any size of clothes to a reference person in the image domain. While previous image-based virtual try-on methods can produce highly natural try-on images, these methods fit the clothes on the person without considering the relative relationship between the physical sizes of the clothes and the person. Different from these methods, our method achieves size-variable virtual try-on in which the image size of the try-on clothes is changed depending on this relative relationship of the physical sizes. To relieve the difficulty in maintaining the physical size of the closes while synthesizing the high-fidelity image of the whole clothes, our proposed method focuses on the residual between the silhouettes of the clothes in the reference and try-on images. We also develop a size-variable virtual try-on dataset consisting of 1,524 images provided by 26 subjects. Furthermore, we propose an evaluation metric for size-variable virtual-try-on. Quantitative and qualitative experimental results show that our method can achieve size-variable virtual try-on better than general virtual try-on methods.

While protoplanetary disks (PPDs) are generally thought to dissipate within several Myr, recent observations have revealed gas in debris disks. The origin of this gas remains uncertain, with one possibility being the unexpectedly long survival of PPDs (the primordial-origin scenario). To explore the plausibility of this scenario, we conduct 1D disk evolution simulations, varying parameters like stellar mass, disk mass, turbulent stress, and magnetohydrodynamic winds, while incorporating stellar evolution to account for time-varying photoevaporation rates. Our focus is on disks where small grains are depleted, as these are potentially long-lived due to reduced far-ultraviolet photoevaporation. Our results show that gas in these disks can survive beyond 10 Myr regardless of the stellar mass, provided they are initially massive ($M_{\mathrm{disk}}\approx 0.1M_*$) with relatively weak turbulent stress ($\alpha \ll 10^{-2}$). The longest lifetimes are consistently found for $M_* = 2 M_{\odot}$ across a wide parameter space, with gas typically persisting at $\sim 10$--$10^3 \mathrm{au}$. Roughly estimated CO masses for these disks fall within the observed range for the most massive gas-rich debris disks around early A stars. These alignments support the plausibility of the primordial-origin scenario. Additionally, our model predicts that accretion persists for as long as the disk survives, which could explain the accretion signatures detected in old disks hosted by low-mass stars, including Peter Pan disks. Our finding also suggests that ongoing accretion may exist in gas-rich debris disks. Thus, searching for accretion signatures could be a key factor to identifying the origin of gas in debris disks., Comment: 20 pages, 16 figures

Debris disks embrace the formation and evolution histories of planetary systems. Recent detections of gas in these disks have received considerable attention, as its origin ties up ongoing disk evolution and the present composition of planet-forming materials. Observations of the CO gas alone, however, cannot reliably differentiate between two leading, competing hypotheses: (1) the observed gas is the leftover of protoplanetary disk gas, and (2) the gas is the outcome of collisions between icy bodies. We propose that such differentiation may become possible by observing cold water vapor. Order-of-magnitude analyses and comparison with existing observations are performed. We show that different hypotheses lead to different masses of water vapor. This occurs because, for both hypotheses, the presence of cold water vapor is attributed to photodesorption from dust particles by attenuated interstellar UV radiation. Cold water vapor cannot be observed by current astronomical facilities as most of its emission lines fall in the far-IR (FIR) range. This work highlights the need for a future FIR space observatory to reveal the origin of gas in debris disks and the evolution of planet-forming disks in general., Comment: 9 pages, 5 figures, 1 table; accepted for publication in A&A

Synthesizers are essential in modern music production. However, their complex timbre parameters, often filled with technical terms, require expertise. This research introduces a method of timbre control in wavetable synthesis that is intuitive and sensible and utilizes semantic labels. Using a conditional variational autoencoder (CVAE), users can select a wavetable and define the timbre with labels such as bright, warm, and rich. The CVAE model, featuring convolutional and upsampling layers, effectively captures the wavetable nuances, ensuring real-time performance owing to their processing in the time domain. Experiments demonstrate that this approach allows for real-time, effective control of the timbre of the wavetable using semantic inputs and aims for intuitive timbre control through data-based semantic control., Comment: 6 pages, 4 figures, Accepted at APSIPA ASC 2024

We investigated some variational methods to compute a wavefunction based on antisymmetric product of geminals (APG). The Waring decomposition on the APG wavefunction leads a finite sum of antisymmetrized geminal power (AGP) wavefunctions, each for which the variational principle can be applied. We call this as AGP-CI method which provides a variational solution of the APG wavefunction efficiently. However, number of AGP wavefunctions in the exact AGP-CI formalism become exponentially large in case of many-electron systems. Therefore, we also investigate the low-rank APG wavefunction, in which the geminal matrices are factorized by the Schur decomposition. Interestingly, only a few non-zero eigenvalues (up to half number of electrons) were found from the Schur decomposition on the APG wavefunction. We developed some methods to approximate the APG wavefunction by lowering the ranks of geminal matrices, and demonstrate their performance. Our new geminal method based on the low-rank decomposition can drastically reduce the number of variational parameters, although there is no efficient algorithm so far, due to some mathematical complications., Comment: 21 pages, 7 figures

E-commerce stores enable multilingual product discovery which require accurate product title translation. Multilingual large language models (LLMs) have shown promising capacity to perform machine translation tasks, and it can also enhance and translate product titles cross-lingually in one step. However, product title translation often requires more than just language conversion because titles are short, lack context, and contain specialized terminology. This study proposes a retrieval-augmented generation (RAG) approach that leverages existing bilingual product information in e-commerce by retrieving similar bilingual examples and incorporating them as few-shot prompts to enhance LLM-based product title translation. Experiment results show that our proposed RAG approach improve product title translation quality with chrF score gains of up to 15.3% for language pairs where the LLM has limited proficiency., Comment: 6 Pages,In Proceedings of ACM CIKM Workshop on Data-Centric AI (CIKM DCAI 2024)

Target sound extraction (TSE) consists of isolating a desired sound from a mixture of arbitrary sounds using clues to identify it. A TSE system requires solving two problems at once, identifying the target source and extracting the target signal from the mixture. For increased practicability, the same system should work with various types of sound. The duality of the problem and the wide variety of sounds make it challenging to train a powerful TSE system from scratch. In this paper, to tackle this problem, we explore using a pre-trained audio foundation model that can provide rich feature representations of sounds within a TSE system. We chose the masked-modeling duo (M2D) foundation model, which appears especially suited for the TSE task, as it is trained using a dual objective consisting of sound-label predictions and improved masked prediction. These objectives are related to sound identification and the signal extraction problems of TSE. We propose a new TSE system that integrates the feature representation from M2D into SoundBeam, which is a strong TSE system that can exploit both target sound class labels and pre-recorded enrollments (or audio queries) as clues. We show experimentally that using M2D can increase extraction performance, especially when employing enrollment clues.

We present a distant automatic speech recognition (DASR) system developed for the CHiME-8 DASR track. It consists of a diarization first pipeline. For diarization, we use end-to-end diarization with vector clustering (EEND-VC) followed by target speaker voice activity detection (TS-VAD) refinement. To deal with various numbers of speakers, we developed a new multi-channel speaker counting approach. We then apply guided source separation (GSS) with several improvements to the baseline system. Finally, we perform ASR using a combination of systems built from strong pre-trained models. Our proposed system achieves a macro tcpWER of 21.3 % on the dev set, which is a 57 % relative improvement over the baseline., Comment: 5 pages, 4 figures, CHiME8 challenge

In the absence of laboratory data, state-of-the-art quantum chemical approaches can provide estimates of the binding energy (BE) of interstellar species with grains. Without BE values, contemporary astrochemical models are compelled to utilize wild guesses, often delivering misleading information. Here, we employed a fully quantum chemical approach to estimate the BE of seven diatomic radicals - CH, NH, OH, SH, CN, NS, and NO - that play a crucial role in shaping the interstellar chemical composition, using a suitable amorphous solid water model as a substrate since water is the principal constituent of interstellar ice in dense and shielded regions. While the BEs are compatible with physisorption, the binding of CH in some sites shows chemisorption, in which a chemical bond to an oxygen atom of a water molecule is formed. While no structural change has been observed for the CN radical, it is believed that the formation of a hemibonded system between the outer layer of the water cluster and the radical is the reason for the unusually large BE in one of the binding sites considered in our study. A significantly lower BE for NO, consistent with recent calculations, is obtained, which helps explain the recently observed HONO/NH$_2$OH and HONO/HNO ratios in the low-mass hot corino IRAS 16293-2422 B with chemical models., Comment: 11 pages, 6 figures, to be published in Astronomy & Astrophysics (A&A) Journal

The determination of material parameters is significantly important in material science, which is often a challenging task. Recently, advancements have shown that magnetic parameters, such as the Dzyaloshinskii-Moriya interaction (DMI), can be estimated from a magnetic domain image using machine learning (ML). This development suggests a potential shift in how magnetic parameters are determined, moving away from traditional measurement techniques to more innovative methods involving image-based inputs processed by ML. In previous studies, the test images used for estimation always matched the training images in size. However, since image size is contingent on the microscopy technique used, the ability to accurately estimate parameters from images of varying sizes is essential. Here, we investigated the feasibility of estimating the DMI constant and saturation magnetization from magnetic domain images of different sizes using ML. We successfully demonstrated that it is possible to estimate these parameters even when the imaging sizes differ between training and test datasets. Additionally, our comparison of the estimation accuracy for the DMI constant and saturation magnetization revealed that the tolerance for differences in image size varies depending on the specific parameter being estimated. These findings could have a significant impact on the future methods of determining magnetic parameters.

Binaural target sound extraction (TSE) aims to extract a desired sound from a binaural mixture of arbitrary sounds while preserving the spatial cues of the desired sound. Indeed, for many applications, the target sound signal and its spatial cues carry important information about the sound source. Binaural TSE can be realized with a neural network trained to output only the desired sound given a binaural mixture and an embedding characterizing the desired sound class as inputs. Conventional TSE systems are trained using signal-level losses, which measure the difference between the extracted and reference signals for the left and right channels. In this paper, we propose adding explicit spatial losses to better preserve the spatial cues of the target sound. In particular, we explore losses aiming at preserving the interaural level (ILD), phase (IPD), and time differences (ITD). We show experimentally that adding such spatial losses, particularly our newly proposed ITD loss, helps preserve better spatial cues while maintaining the signal-level metrics., Comment: Accepted in the International Workshop on Acoustic Signal Enhancement (IWAENC 2024)

We present an analysis of a stellar occultation event caused by a near-Earth asteroid (98943) 2001 CC21, an upcoming flyby target in the Hayabusa2 extended mission, on March 5, 2023. To accurately determine the asteroid's shape from diffraction-affected light curves, we developed a novel data reduction technique named the Diffracted Occultation's United Simulator for Highly Informative Transient Explorations (DOUSHITE). Using DOUSHITE-generated synthetic models, we derived constraints on (98943) 2001 CC21's shadow shape from the single-chord occultation data. Our results suggest a significant elongation of the shadow with an axis ratio of $b/a = 0.37\pm0.09$. This shape can be crucial for planning Hayabusa2's high-speed flyby to optimise the limited imaging opportunities., Comment: 10 pages, 9 figures, PASJ accepted

Photoevaporation driven by hydrogen-ionizing radiation, also known as extreme-ultraviolet (EUV), profoundly shapes the lives of diverse astrophysical objects. Focusing here mainly on the dispersal of protoplanetary disks, we construct an analytical model accounting for the finite timescales of photoheating and photoionization. The model offers improved estimates for the ionization, temperature, and velocity structures versus distance from the central source, for a given EUV emission rate and spectral hardness. Compared to the classical picture of fully-ionized and isothermal winds with temperatures $\approx 10^4{\rm \,K}$ and speeds $\approx 10{\rm \,km\,s^{-1}}$, our model unveils broader hydrodynamical and thermochemical states of photoevaporative winds. In contrast to the classical picture, T~Tauri stars with EUV luminosities around $10^{30}{\rm \,erg\,s^{-1}}$ have non-isothermal ionized winds at lower temperatures than the classical value if the spectrum is soft, with an average deposited energy per photoionization less than about 3.7\,eV. Conversely, if the spectrum is hard, the winds tend to be atomic and isothermal at most radii in the disk. For lower EUV intensities, even with soft spectra, atomic winds can emerge beyond $\sim 10{\, \rm au}$ through advection. We demonstrate that the analytical model's predictions are in general agreement with detailed radiation-hydrodynamics calculations. The model furthermore illustrates how the energy efficiency of photoevaporation varies with the intensity and spectral hardness of the EUV illumination, as well as addressing discrepancies in the literature around the effectiveness of X-ray photoevaporation. These findings highlight the importance of considering the finite timescales of photoheating and photoionization, both in modeling and in interpreting observational data., Comment: Submitted to ApJ

To utilize half-metallic Heusler alloys in practical spintronic devices, such as magnetic sensors and magnetic memories, the key is to realize highly textured and structurally ordered polycrystalline thin films. In this study, we fabricated polycrystalline Co2FeGa0.5Ge0.5 (CFGG) Heusler alloy films deposited on a [001]-oriented Ag buffer layer, which was achieved by introducing N2 into Ar during the sputtering process, on a thermally oxidized Si substrate. We obtained strongly [001]-oriented CFGG films with B2 ordering and a high saturation magnetization close to the theoretical value, which can provide highly spin-polarized electric and spin current sources in spintronic devices with industrial viability., Comment: 16 papers, 5 figures

This paper proposes Group Activity Feature (GAF) learning in which features of multi-person activity are learned as a compact latent vector. Unlike prior work in which the manual annotation of group activities is required for supervised learning, our method learns the GAF through person attribute prediction without group activity annotations. By learning the whole network in an end-to-end manner so that the GAF is required for predicting the person attributes of people in a group, the GAF is trained as the features of multi-person activity. As a person attribute, we propose to use a person's action class and appearance features because the former is easy to annotate due to its simpleness, and the latter requires no manual annotation. In addition, we introduce a location-guided attribute prediction to disentangle the complex GAF for extracting the features of each target person properly. Various experimental results validate that our method outperforms SOTA methods quantitatively and qualitatively on two public datasets. Visualization of our GAF also demonstrates that our method learns the GAF representing fined-grained group activity classes. Code: https://github.com/chihina/GAFL-CVPR2024., Comment: Accepted to CVPR2024

The aging law and the slip law are two representative evolution laws of the rate- and state-dependent friction (RSF) law, based on canonical behaviors in three types of laboratory experiments: slide-hold-slide (SHS), velocity-step (VS), and steady-state (SS) tests. The aging law explains the SHS canon but contradicts the VS canon, and vice versa for the slip law. The later proposed composite law, which switches these two laws according to the slip rate $V$, explains both canons but contradicts the SS canon. The present study constructs evolution laws satisfying all three canons throughout the range of variables where experiments have confirmed the canons. By recompiling the three canons, we have derived constraints on the evolution law and found that the evolution rates in the strengthening phases of the SHS and VS canons are so different that complete reconciliation throughout the entire range of variables is mathematically impossible. However, for the limited range of variables probed by experiments so far, we have found that the SHS and VS canons can be reconciled without violating the SS canon by switching the evolution function according to $\Omega$, the ratio of the state $\theta$ to its steady-state value $\theta_{\rm SS}$ for the instantaneous slip rate. We could generally show that, as long as the state evolution rate $\dot \theta$ depends only on the instantaneous values of $V$ and $\theta$, simultaneous reproduction of the three canons, throughout the experimentally confirmed range, requires the aging-law-like evolution for $\Omega$ sufficiently below a threshold $\beta$ and the slip-law-like evolution for $\Omega$ sufficiently above $\beta$. The validity of the canons in existing experiments suggests $\beta \lesssim 0.01$., Comment: 106 pages, 14 figures

Although mask-based beamforming is a powerful speech enhancement approach, it often requires manual parameter tuning to handle moving speakers. Recently, this approach was augmented with an attention-based spatial covariance matrix aggregator (ASA) module, enabling accurate tracking of moving speakers without manual tuning. However, the deep neural network model used in this module is limited to specific microphone arrays, necessitating a different model for varying channel permutations, numbers, or geometries. To improve the robustness of the ASA module against such variations, in this paper we investigate three approaches: training with random channel configurations, employing the transform-average-concatenate method to process multi-channel input features, and utilizing robust input features. Our experiments on the CHiME-3 and DEMAND datasets show that these approaches enable the ASA-augmented beamformer to track moving speakers across different microphone arrays unseen in training., Comment: accepted at Interspeech 2024

Recent advances in the study of materials with topological electronic band structures have revealed magnetic materials exhibiting giant anomalous Hall effects (AHE). The giant AHE has not only attracted the research interest in its mechanism but also opened up the possibility of practical application in magnetic sensors. In this article, we describe simulation-based investigations of AHE magnetic sensors for the applications to read head sensors (readers) of hard disk drives. With the shrinking of magnetic recording patterns, the reader technology, which currently uses multilayer-based tunnel magnetoresistance (TMR) devices, is associated with fundamental challenges, such as insufficient spatial resolution and signal-to-noise ratio (SNR) in sensors with dimensions below 20 nm. The structure of an AHE-based device composed of a single ferromagnetic material is advantageous for magnetic sensors with nanoscale dimensions. We found that AHE readers using topological ferromagnets with giant AHE, such as Co2MnGa, can achieve a higher SNR than current TMR readers. The higher SNR originates from the large output signal of the giant AHE as well as from the reduced thermal magnetic noise, which is the dominant noise in TMR readers. We highlight a major challenge in the development of AHE readers: the reduction in the output signal due to the shunting of the bias current and the leakage of the Hall voltage through the soft magnetic shields surrounding the AHE reader. We propose reader structures that overcome this challenge. Finally, we discuss the scope for future research to realize AHE readers., Comment: Revised version after peer-review