arXiv subjects

Yuji Matsumoto

Publications and source records attributed to Yuji Matsumoto.

At least 37 records · Page 2Linked to original sources

arXiv2023

An Earth-sized Planet around an M5 Dwarf Star at 22 pc

We report on the discovery of an Earth-sized transiting planet ($R_p=1.015\pm0.051\,R_\oplus$) in a $P=4.02$ day orbit around K2-415 (EPIC 211414619), an M5V star at 22 pc. The planet candidate was first identified by analyzing the light curve data by the K2 mission, and is here shown to exist in the most recent data from TESS. Combining the light curves with the data secured by our follow-up observations including high-resolution imaging and near infrared spectroscopy with IRD, we rule out false positive scenarios, finding a low false positive probability of $2\times 10^{-4}$. Based on IRD's radial velocities of K2-415, which were sparsely taken over three years, we obtain the planet mass of $3.0\pm 2.7\,M_\oplus$ ($M_p<7.5\,M_\oplus$ at $95\,\%$ confidence) for K2-415b. Being one of the lowest mass stars ($\approx 0.16\,M_\odot$) known to host an Earth-sized transiting planet, K2-415 will be an interesting target for further follow-up observations, including additional radial velocity monitoring and transit spectroscopy.

astro-ph.EP↗

arXiv2022

Switching to Discriminative Image Captioning by Relieving a Bottleneck of Reinforcement Learning

Discriminativeness is a desirable feature of image captions: captions should describe the characteristic details of input images. However, recent high-performing captioning models, which are trained with reinforcement learning (RL), tend to generate overly generic captions despite their high performance in various other criteria. First, we investigate the cause of the unexpectedly low discriminativeness and show that RL has a deeply rooted side effect of limiting the output words to high-frequency words. The limited vocabulary is a severe bottleneck for discriminativeness as it is difficult for a model to describe the details beyond its vocabulary. Then, based on this identification of the bottleneck, we drastically recast discriminative image captioning as a much simpler task of encouraging low-frequency word generation. Hinted by long-tail classification and debiasing methods, we propose methods that easily switch off-the-shelf RL models to discriminativeness-aware models with only a single-epoch fine-tuning on the part of the parameters. Extensive experiments demonstrate that our methods significantly enhance the discriminativeness of off-the-shelf RL models and even outperform previous discriminativeness-aware methods with much smaller computational costs. Detailed analysis and human evaluation also verify that our methods boost the discriminativeness without sacrificing the overall quality of captions.

cs.CV↗

arXiv2022

Exploring Optimal Granularity for Extractive Summarization of Unstructured Health Records: Analysis of the Largest Multi-Institutional Archive of Health Records in Japan

Automated summarization of clinical texts can reduce the burden of medical professionals. "Discharge summaries" are one promising application of the summarization, because they can be generated from daily inpatient records. Our preliminary experiment suggests that 20-31% of the descriptions in discharge summaries overlap with the content of the inpatient records. However, it remains unclear how the summaries should be generated from the unstructured source. To decompose the physician's summarization process, this study aimed to identify the optimal granularity in summarization. We first defined three types of summarization units with different granularities to compare the performance of the discharge summary generation: whole sentences, clinical segments, and clauses. We defined clinical segments in this study, aiming to express the smallest medically meaningful concepts. To obtain the clinical segments, it was necessary to automatically split the texts in the first stage of the pipeline. Accordingly, we compared rule-based methods and a machine learning method, and the latter outperformed the formers with an F1 score of 0.846 in the splitting task. Next, we experimentally measured the accuracy of extractive summarization using the three types of units, based on the ROUGE-1 metric, on a multi-institutional national archive of health records in Japan. The measured accuracies of extractive summarization using whole sentences, clinical segments, and clauses were 31.91, 36.15, and 25.18, respectively. We found that the clinical segments yielded higher accuracy than sentences and clauses. This result indicates that summarization of inpatient records demands finer granularity than sentence-oriented processing. Although we used only Japanese health records, it can be interpreted as follows: physicians extract "concepts of medical significance" from patient records and recombine them ...

cs.CL↗

arXiv2022

Unsupervised Lexical Substitution with Decontextualised Embeddings

We propose a new unsupervised method for lexical substitution using pre-trained language models. Compared to previous approaches that use the generative capability of language models to predict substitutes, our method retrieves substitutes based on the similarity of contextualised and decontextualised word embeddings, i.e. the average contextual representation of a word in multiple contexts. We conduct experiments in English and Italian, and show that our method substantially outperforms strong baselines and establishes a new state-of-the-art without any explicit supervision or fine-tuning. We further show that our method performs particularly well at predicting low-frequency substitutes, and also generates a diverse list of substitute candidates, reducing morphophonetic or morphosyntactic biases induced by article-noun agreement.

cs.CL↗

arXiv2022

A Super-Earth Orbiting Near the Inner Edge of the Habitable Zone around the M4.5-dwarf Ross 508

We report the near-infrared radial-velocity (RV) discovery of a super-Earth planet on a 10.77-day orbit around the M4.5 dwarf Ross 508 ($J_\mathrm{mag}=9.1$). Using precision RVs from the Subaru Telescope IRD (InfraRed Doppler) instrument, we derive a semi-amplitude of $3.92^{+0.60}_{-0.58}$ ${\rm m\,s}^{-1}$, corresponding to a planet with a minimum mass $m \sin i = 4.00^{+0.53}_{-0.55}\ M_{\oplus}$. We find no evidence of significant signals at the detected period in spectroscopic stellar activity indicators or MEarth photometry. The planet, Ross 508 b, has a semimajor-axis of $0.05366^{+0.00056}_{-0.00049}$ au. This gives an orbit-averaged insolation of $\approx$1.4 times the Earth's value, placing Ross 508 b near the inner edge of its star's habitable zone. We have explored the possibility that the planet has a high eccentricity and its host is accompanied by an additional unconfirmed companion on a wide orbit. Our discovery demonstrates that the near-infrared RV search can play a crucial role to find a low-mass planet around cool M dwarfs like Ross 508.

astro-ph.EP↗

arXiv2022

Global Entity Disambiguation with BERT

We propose a global entity disambiguation (ED) model based on BERT. To capture global contextual information for ED, our model treats not only words but also entities as input tokens, and solves the task by sequentially resolving mentions to their referent entities and using resolved entities as inputs at each step. We train the model using a large entity-annotated corpus obtained from Wikipedia. We achieve new state-of-the-art results on five standard ED datasets: AIDA-CoNLL, MSNBC, AQUAINT, ACE2004, and WNED-WIKI. The source code and model checkpoint are available at https://github.com/studio-ousia/luke.

cs.CL↗

arXiv2021

Learning Contextualised Cross-lingual Word Embeddings and Alignments for Extremely Low-Resource Languages Using Parallel Corpora

We propose a new approach for learning contextualised cross-lingual word embeddings based on a small parallel corpus (e.g. a few hundred sentence pairs). Our method obtains word embeddings via an LSTM encoder-decoder model that simultaneously translates and reconstructs an input sentence. Through sharing model parameters among different languages, our model jointly trains the word embeddings in a common cross-lingual space. We also propose to combine word and subword embeddings to make use of orthographic similarities across different languages. We base our experiments on real-world data from endangered languages, namely Yongning Na, Shipibo-Konibo, and Griko. Our experiments on bilingual lexicon induction and word alignment tasks show that our model outperforms existing methods by a large margin for most language pairs. These results demonstrate that, contrary to common belief, an encoder-decoder translation model is beneficial for learning cross-lingual representations even in extremely low-resource conditions. Furthermore, our model also works well on high-resource conditions, achieving state-of-the-art performance on a German-English word-alignment task.

cs.CL↗

arXiv2021

Size evolution of close-in super-Earths through giant impacts and photoevaporation

The Kepler transit survey with follow-up spectroscopic observations has discovered numerous super-Earth sized planets and revealed intriguing features of their sizes, orbital periods, and their relations between adjacent planets. For the first time, we investigate the size evolution of planets via both giant impacts and photoevaporation to compare with these observed features. We calculate the size of a protoplanet, which is the sum of its core and envelope sizes, by analytical models. $N$-body simulations are performed to evolve planet sizes during the giant impact phase with envelope stripping via impact shocks. We consider the initial radial profile of the core mass and the initial envelope mass fractions as parameters. Inner planets can lose their whole envelopes via giant impacts, while outer planets can keep their initial envelopes since they do not experience giant impacts. Photoevaporation is simulated to evolve planet sizes afterward. Our results suggest that the period-radius distribution of the observed planets would be reproduced if we perform simulations in which the initial radial profile of the core mass follows a wide range of power-law distributions and the initial envelope mass fractions are $\sim0.1$. Moreover, our model shows that the adjacent planetary pairs have similar sizes and regular spacings, with slight differences from detailed observational results such as the radius gap.

astro-ph.EP↗

arXiv2021

On the crystallinity of silicate dust in evolving protoplanetary disks due to magnetically driven disk winds

We present a novel mechanism for the outward transport of crystalline dust particles: the outward radial drift of pebbles. The dust ring structure is frequently observed in protoplanetary disks. One of the plausible mechanisms of the formation of dust rings is the accumulation of pebbles around the pressure maximum, which is formed by the mass loss due to magnetically driven disk winds. In evolving protoplanetary disks due to magnetically driven disk winds, dust particles can migrate outwardly from the crystallization front to the pressure maximum by radial drift. We found that the outward radial drift process can transport crystalline dust particles efficiently when the radial drift timescale is shorter than the advection timescale. Our model predicts that the crystallinity of silicate dust particles could be as high as 100% inside the dust ring position.

astro-ph.EP↗

arXiv2021

Removing Word-Level Spurious Alignment between Images and Pseudo-Captions in Unsupervised Image Captioning

Unsupervised image captioning is a challenging task that aims at generating captions without the supervision of image-sentence pairs, but only with images and sentences drawn from different sources and object labels detected from the images. In previous work, pseudo-captions, i.e., sentences that contain the detected object labels, were assigned to a given image. The focus of the previous work was on the alignment of input images and pseudo-captions at the sentence level. However, pseudo-captions contain many words that are irrelevant to a given image. In this work, we investigate the effect of removing mismatched words from image-sentence alignment to determine how they make this task difficult. We propose a simple gating mechanism that is trained to align image features with only the most reliable words in pseudo-captions: the detected object labels. The experimental results show that our proposed method outperforms the previous methods without introducing complex sentence-level learning objectives. Combined with the sentence-level alignment method of previous work, our method further improves its performance. These results confirm the importance of careful alignment in word-level details.

cs.CL↗

arXiv2021

Formation of rims around chondrules via porous aggregate accretion

Chondrules are often surrounded by fine-grained rims or igneous rims. The properties of these rims reflect their formation histories. While the formation of fine-grained rims is modeled by the accretion of dust grains onto chondrules, the accretion should be followed by the growth of dust grains due to the shorter growth timescale than the accretion. In this paper, we investigate the formation of rims, taking into account the growth of porous dust aggregates. We estimate the rim thickness as a function of the chondrule fraction at a time when dust aggregate accretion onto chondrules is switched to collisions between these chondrules. Our estimations are consistent with the measured thicknesses of fine-grained rims in ordinary chondrites. However, those of igneous rims are thicker than our estimations. The thickness of igneous rims would be enlarged in remelting events.

astro-ph.EP↗

arXiv2020

Formation of multiple-planet systems in resonant chains around M dwarfs

Recent observations have revealed the existence of multiple-planet systems composed of Earth-mass planets around late M dwarfs. Most of their orbits are close to commensurabilities, which suggests that planets were commonly trapped in resonant chains in their formation around low-mass stars. We investigate the formation of multiple-planet systems in resonant chains around low-mass stars. A time-evolution model of the multiple-planet formation via pebble accretion in the early phase of the disk evolution is constructed based on the formation model for the TRAPPIST-1 system by Ormel et al. (2017). Our simulations show that knowing the protoplanet appearance timescale is important for determining the number of planets and their trapped resonances: as the protoplanet appearance timescale increases, fewer planets are formed, which are trapped in more widely separated resonances. We find that there is a range of the protoplanet appearance timescale for forming the stable multiple-planet systems in resonant chains. This range depends on the stellar mass and disk size. We suggest that the protoplanet appearance timescale is a key parameter for studying the formation of multiple-planet systems with planets in resonant chains around low-mass stars. The composition of the planets in our model is also discussed.

astro-ph.EP↗

arXiv2020

LUKE: Deep Contextualized Entity Representations with Entity-aware Self-attention

Entity representations are useful in natural language tasks involving entities. In this paper, we propose new pretrained contextualized representations of words and entities based on the bidirectional transformer. The proposed model treats words and entities in a given text as independent tokens, and outputs contextualized representations of them. Our model is trained using a new pretraining task based on the masked language model of BERT. The task involves predicting randomly masked words and entities in a large entity-annotated corpus retrieved from Wikipedia. We also propose an entity-aware self-attention mechanism that is an extension of the self-attention mechanism of the transformer, and considers the types of tokens (words or entities) when computing attention scores. The proposed model achieves impressive empirical performance on a wide range of entity-related tasks. In particular, it obtains state-of-the-art results on five well-known datasets: Open Entity (entity typing), TACRED (relation classification), CoNLL-2003 (named entity recognition), ReCoRD (cloze-style question answering), and SQuAD 1.1 (extractive question answering). Our source code and pretrained representations are available at https://github.com/studio-ousia/luke.

cs.CL↗

arXiv2020

Wikipedia2Vec: An Efficient Toolkit for Learning and Visualizing the Embeddings of Words and Entities from Wikipedia

The embeddings of entities in a large knowledge base (e.g., Wikipedia) are highly beneficial for solving various natural language tasks that involve real world knowledge. In this paper, we present Wikipedia2Vec, a Python-based open-source tool for learning the embeddings of words and entities from Wikipedia. The proposed tool enables users to learn the embeddings efficiently by issuing a single command with a Wikipedia dump file as an argument. We also introduce a web-based demonstration of our tool that allows users to visualize and explore the learned embeddings. In our experiments, our tool achieved a state-of-the-art result on the KORE entity relatedness dataset, and competitive results on various standard benchmark datasets. Furthermore, our tool has been used as a key component in various recent studies. We publicize the source code, demonstration, and the pretrained embeddings for 12 languages at https://wikipedia2vec.github.io.

cs.CL↗

arXiv2020

A Fast and Easy Regression Technique for k-NN Classification Without Using Negative Pairs

This paper proposes an inexpensive way to learn an effective dissimilarity function to be used for $k$-nearest neighbor ($k$-NN) classification. Unlike Mahalanobis metric learning methods that map both query (unlabeled) objects and labeled objects to new coordinates by a single transformation, our method learns a transformation of labeled objects to new points in the feature space whereas query objects are kept in their original coordinates. This method has several advantages over existing distance metric learning methods: (i) In experiments with large document and image datasets, it achieves $k$-NN classification accuracy better than or at least comparable to the state-of-the-art metric learning methods. (ii) The transformation can be learned efficiently by solving a standard ridge regression problem. For document and image datasets, training is often more than two orders of magnitude faster than the fastest metric learning methods tested. This speed-up is also due to the fact that the proposed method eliminates the optimization over "negative" object pairs, i.e., objects whose class labels are different. (iii) The formulation has a theoretical justification in terms of reducing hubness in data.

cs.LG↗

arXiv2020

Ejection of close-in super-Earths around low-mass stars in the giant impact stage

Earth-sized planets were observed in close-in orbits around M dwarfs. While more and more planets are expected to be uncovered around M dwarfs, theories of their formation and dynamical evolution are still in their infancy. We investigate the giant impact growth of protoplanets, which includes strong scattering around low-mass stars. The aim is to clarify whether strong scattering around low-mass stars affects the orbital and mass distributions of the planets. We perform $N$-body simulation of protoplanets by systematically surveying the parameter space of the stellar mass and surface density of protoplanets. We find that protoplanets are often ejected after twice or three times close-scattering around late M dwarfs. The ejection sets the upper limit of the largest planet mass. Adopting the surface density scaling linearly with the stellar mass, we find that as the stellar mass decreases less massive planets are formed in orbits with higher eccentricities and inclinations. Under this scaling, we also find that a few close-in protoplanets are generally ejected. The ejection of protoplanets plays an important role in the mass distribution of super-Earths around late M dwarfs. The mass relation of observed close-in super-Earths and their central star mass is well reproduced by ejection.

astro-ph.EP↗

arXiv2020

Breaking Resonant Chains: Destabilization of Resonant Planets due to Long-term Mass Evolution

Recent exoplanet observations reported a large number of multiple-planet systems, in which some of the planets are in a chain of resonances. The fraction of resonant systems to non-resonant systems provides clues about their formation history. We investigated the orbital stability of planets in resonant chains by considering the long-term evolution of planetary mass and stellar mass and using orbital calculations. We found that while resonant chains were stable, they can be destabilized by a change of $\sim$10% in planetary mass. Such a mass evolution can occur by atmospheric escape due to photoevaporation. We also found that resonant chains can be broken by a stellar mass loss of $\lesssim1$%, which would be explained by stellar winds or coronal mass ejections. The long-term mass change of planets and stars plays an important role in the orbital evolutions of planetary systems including super-Earths.

astro-ph.EP↗

arXiv2020

Length-controllable Abstractive Summarization by Guiding with Summary Prototype

We propose a new length-controllable abstractive summarization model. Recent state-of-the-art abstractive summarization models based on encoder-decoder models generate only one summary per source text. However, controllable summarization, especially of the length, is an important aspect for practical applications. Previous studies on length-controllable abstractive summarization incorporate length embeddings in the decoder module for controlling the summary length. Although the length embeddings can control where to stop decoding, they do not decide which information should be included in the summary within the length constraint. Unlike the previous models, our length-controllable abstractive summarization model incorporates a word-level extractive module in the encoder-decoder model instead of length embeddings. Our model generates a summary in two steps. First, our word-level extractor extracts a sequence of important words (we call it the "prototype text") from the source text according to the word-level importance scores and the length constraint. Second, the prototype text is used as additional input to the encoder-decoder model, which generates a summary by jointly encoding and copying words from both the prototype text and source text. Since the prototype text is a guide to both the content and length of the summary, our model can generate an informative and length-controlled summary. Experiments with the CNN/Daily Mail dataset and the NEWSROOM dataset show that our model outperformed previous models in length-controlled settings.

cs.CL↗