Search

This paper describes RHP Friends, a social humanoid robot developed to enable assistive robotic deployments in human-coexisting environments. As a use-case application, we present its potential use in nursing by extending its capabilities to operate human devices and tools according to the task and by enabling remote assistance operations. To meet a wide variety of tasks and situations in environments designed by and for humans, we developed a system that seamlessly integrates the slim and lightweight robot and several technologies: locomanipulation, multi-contact motion, teleoperation, and object detection and tracking. We demonstrated the system's usage in a nursing application. The robot efficiently performed the daily task of patient transfer and a non-routine task, represented by a request to operate a circuit breaker. This demonstration, held at the 2023 International Robot Exhibition (IREX), conducted three times a day over three days., Comment: IEEE Robotics and Automation Magazine, In press

In this paper, we study the local boundedness of local weak solutions to the following parabolic equation associated with fractional $p$-Laplacian type operators $$ \partial_t u(t,x)-\text{p.v.}\int_{\R^d}|u(t,y)-u(t,x)|^{p-2}(u(t,y)-u(t,x))J(t;x,y)\,dy=0,\quad (t,x)\in \R\times \R^d, $$ where $\text{p.v.}$ means the integral in the principal value sense, $p\in(1,\infty)$ and $J(t;x,y)$ is comparable to the kernel of the fractional $p$-Laplacian operator $|x-y|^{-d-sp}$ with $s\in(0,1)$ and uniformly in $(t;x,y)\in\R\times\R^d\times\R^d$. Unlike existing results in the literature, the local boundedness of the solutions obtained in this paper extends the known results for the linear case (i.e., the case that $p=2$), in particular with a nonlocal parabolic tail that uses the $L^1$-norm in time for all $p\in (1,\infty)$. The proof is based on a new level set truncation in the De Giorgi-Nash-Moser iteration and a careful choice of iteration orders, as well as a general Caccioppoli-type inequality that is efficiently applied to fractional $p$-Laplacian type operators with all $p>1$., Comment: 25 pages

Thirteen years after the Fukushima Daiichi nuclear power plant accident, Japan's nuclear energy accounts for only approximately 6% of electricity production, as most nuclear plants remain shut down. To revitalize the nuclear industry and achieve sustainable development goals, effective communication with Japanese citizens, grounded in an accurate understanding of public sentiment, is of paramount importance. While nationwide surveys have traditionally been used to gauge public views, the rise of social media in recent years has provided a promising new avenue for understanding public sentiment. To explore domestic sentiment on nuclear energy-related issues expressed online, we analyzed the content and comments of over 3,000 YouTube videos covering topics related to nuclear energy. Topic modeling was used to extract the main topics from the videos, and sentiment analysis with large language models classified user sentiments towards each topic. Additionally, word co-occurrence network analysis was performed to examine the shift in online discussions during August and September 2023 regarding the release of treated water. Overall, our results provide valuable insights into the online discourse on nuclear energy and contribute to a more comprehensive understanding of public sentiment in Japan., Comment: 15 pages, 9 figures, 4 tables

We report on tip-enhanced Raman scattering (TERS) of H2 and D2 molecules physisorbed within a plasmonic picocavity at a cryogenic temperature (10 K). The intense Raman peaks resulting from the rotational and vibrational transitions are observed at sub-nanometer gap distances of the junction formed by a Ag tip and Ag(111) surface. We clarify that the predominant contribution of the electromagnetic field enhancement of the picocavity to the detection of a single hydrogen molecule. The gap-distance dependent TERS reveals not only the evolution of the picocavity field, but also the interaction between the molecule and tip/surface, which exhibit nontrivial isotope effects. A significant red-shift and peak broadening of the H-H stretching as the gap distance decreases, while the D-D stretching mode is unaffected. A combination of density functional theory and reduced-dimension models reveals that a distinct anharmonicity in the mode potential of H2 is one cause of the anomalous red-shift, whereas D2 has less anharmonicity due to the geometric isotope effect.

Scattering-type scanning near-field optical microscopy (s-SNOM) allows for the observation of the optical response of material surfaces with a resolution far below the diffraction limit. A spatial resolution of 10-100 nm is routinely achieved in s-SNOM based on amplitude-modulation atomic force microscopy (AM-AFM) with tapping amplitudes of tens of nanometers. However, optical imaging and spectroscopy of structures that are localized to the atomic scale remain a significant challenge. This can be overcome by combining the field enhancement localized at the atomic-scale structure of the tip apex, namely a plasmonic picocavity, with frequency-modulationAFM (FM-AFM), namely non-contact AFM, in a stable cryogenic ultrahigh vacuum environment. Here, we developed picocavityenhanced SNOM (PE-SNOM) under visible laser illumination based on the integration of a quartz tuning fork sensor with small-amplitude oscillations of 1 nm or less. In addition, the use of a focused ion beam-polished silver tip mounted on the sensor leads to strong field enhancement in the picocavity and ensures minimal background scattering from the tip shaft. PE-SNOM allows us to obtain a material-contrast image of silicon islands on a silver surface with 1-nm lateral resolution, which surpasses the conventional limits of s-SNOM. PE-SNOM paves the way for the acquisition of optical information from atomic-scale targets, such as single photo-active defects and molecules.

Two-dimensional (2D) $\beta$-TeO$_2$ has gained attention as a promising material for optoelectronic and power device applications, thanks to its transparency and high hole mobility. However, the underlying mechanism behind its $p$-type conductivity and dopability remains unclear. In this study, we investigate the intrinsic and extrinsic point defects in monolayer and bilayer $\beta$-TeO$_2$, the latter of which has been experimentally synthesized, using the HSE+D3 hybrid functional. Our results reveal that most intrinsic defects are unlikely to contribute to $p$-type doping in 2D $\beta$-TeO$_2$. Moreover, Si contamination could further impair $p$-type conductivity. Since the point defects do not contribute to $p$-type conductivity, we propose two possible mechanisms for hole conduction: hopping conduction via localized impurity states, and substrate effects. We also explored substitutional $p$-type doping in 2D $\beta$-TeO$_2$ with 10 trivalent elements. Among these, the Bi dopant is found to exhibit a relatively shallow acceptor transition level. However, most dopants tend to introduce deep localized states, where hole polarons become trapped at Te's lone pairs. Interestingly, monolayer $\beta$-TeO$_2$ shows potential advantages over bilayers due to reduced self-compensation effects for $p$-type dopants. These findings provide valuable insights into defect engineering strategies for future electronic applications involving 2D $\beta$-TeO$_2$.

This paper investigates test-time adaptation (TTA) for regression, where a regression model pre-trained in a source domain is adapted to an unknown target distribution with unlabeled target data. Although regression is one of the fundamental tasks in machine learning, most of the existing TTA methods have classification-specific designs, which assume that models output class-categorical predictions, whereas regression models typically output only single scalar values. To enable TTA for regression, we adopt a feature alignment approach, which aligns the feature distributions between the source and target domains to mitigate the domain gap. However, we found that naive feature alignment employed in existing TTA methods for classification is ineffective or even worse for regression because the features are distributed in a small subspace and many of the raw feature dimensions have little significance to the output. For an effective feature alignment in TTA for regression, we propose Significant-subspace Alignment (SSA). SSA consists of two components: subspace detection and dimension weighting. Subspace detection finds the feature subspace that is representative and significant to the output. Then, the feature alignment is performed in the subspace during TTA. Meanwhile, dimension weighting raises the importance of the dimensions of the feature subspace that have greater significance to the output. We experimentally show that SSA outperforms various baselines on real-world datasets., Comment: Accepted to ICLR 2025

Homogenization for non-local operators in periodic environments has been studied intensively. So far, these works are mainly devoted to the qualitative results, that is, to determine explicitly the operators in the limit. To the best of authors' knowledge, there is no result concerning the convergence rates of the homogenization for stable-like operators in periodic environments. In this paper, we establish a quantitative homogenization result for symmetric $\alpha$-stable-like operators on $\R^d$ with periodic coefficients. In particular, we show that the convergence rate for the solutions of associated Dirichlet problems on a bounded domain $D$ is of order $$ \varepsilon^{(2-\alpha)/2}\I_{\{\alpha\in (1,2)\}}+\varepsilon^{\alpha/2}\I_{\{\alpha\in (0,1)\}}+\varepsilon^{1/2}|\log \e|^2\I_{\{\alpha=1\}}, $$ while, when the solution to the equation in the limit is in $C^2_c(D)$, the convergence rate becomes $$ \varepsilon^{2-\alpha}\I_{\{\alpha\in (1,2)\}}+\varepsilon^{\alpha}\I_{\{\alpha\in (0,1)\}}+\varepsilon |\log \e|^2\I_{\{\alpha=1\}}. $$ This indicates that the boundary decay behaviors of the solution to the equation in the limit affects the convergence rate in the homogenization., Comment: 34 pages

The progress in text summarization techniques has been remarkable. However the task of accurately extracting and summarizing necessary information from highly specialized documents such as research papers has not been sufficiently investigated. We are focusing on the task of extracting research questions (RQ) from research papers and construct a new dataset consisting of machine learning papers, RQ extracted from these papers by GPT-4, and human evaluations of the extracted RQ from multiple perspectives. Using this dataset, we systematically compared recently proposed LLM-based evaluation functions for summarizations, and found that none of the functions showed sufficiently high correlations with human evaluations. We expect our dataset provides a foundation for further research on developing better evaluation functions tailored to the RQ extraction task, and contribute to enhance the performance of the task. The dataset is available at https://github.com/auto-res/PaperRQ-HumanAnno-Dataset.

In the context of a metric measure space $(X,d,\mu)$, we explore the potential-theoretic implications of having a finite-dimensional Besov space. We prove that if the dimension of the Besov space $B^\theta_{p,p}(X)$ is $k>1$, then $X$ can be decomposed into $k$ number of irreducible components (Theorem 1.1). Note that $\theta$ may be bigger than $1$, as our framework includes fractals. We also provide sufficient conditions under which the dimension of the Besov space is $1$. We introduce critical exponents $\theta_p(X)$ and $\theta_p^{\ast}(X)$ for the Besov spaces. As examples illustrating Theorem 1.1, we compute these critical exponents for spaces $X$ formed by glueing copies of $n$-dimensional cubes, the Sierpi\'{n}ski gaskets, and of the Sierpi\'{n}ski carpet., Comment: 24 pages, 3 figures

We consider the bifurcation diagram of radial solutions for the Gelfand problem with a positive radially symmetric weight in the unit ball. We deal with the exponential nonlinearity and a power-type nonlinearity. When the weight is constant, it is well-known that the bifurcation curve exhibits three different types depending on the dimension and the exponent of power for higher dimensions, while the curve exhibits only one type in two dimensions. In this paper, we succeed in realizing in two dimensions a phenomenon such that the bifurcation curve exhibits all of the three types, by choosing the weight appropriately. In particular, to the best of the author's knowledge, it is the first result to establish in two dimensions the bifurcation curve having no turning points., Comment: 24 pages

Designing a safe policy for uncertain environments is crucial in real-world control applications. However, this challenge remains inadequately addressed within the Markov decision process (MDP) framework. This paper presents the first algorithm capable of identifying a near-optimal policy in a robust constrained MDP (RCMDP), where an optimal policy minimizes cumulative cost while satisfying constraints in the worst-case scenario across a set of environments. We first prove that the conventional Lagrangian max-min formulation with policy gradient methods can become trapped in suboptimal solutions by encountering a sum of conflicting gradients from the objective and constraint functions during its inner minimization problem. To address this, we leverage the epigraph form of the RCMDP problem, which resolves the conflict by selecting a single gradient from either the objective or the constraints. Building on the epigraph form, we propose a binary search algorithm with a policy gradient subroutine and prove that it identifies an $\varepsilon$-optimal policy in an RCMDP with $\tilde{\mathcal{O}}(\varepsilon^{-4})$ policy evaluations.

Although $\beta$-TeO$_2$ is a promising $p$-type transparent conducting oxide (TCO) due to the large optical gap ($\sim$ 3.7 eV) and a light effective hole mass, its hole dopability still remains unexplored. In this work, electronic structure of $\beta$-TeO$_2$ and its point defects are investigated using the HSEsol functional with the band-gap-tuned mixing parameter. Our calculations reveal that $\beta$-TeO$_2$ exhibits a significant difference between the fundamental and optical band gaps because lower energy optical transitions are dipole forbidden. Additionally, it has a low hole effective mass, especially in-plane. The point defect calculations show that $\beta$-TeO$_2$ is intrinsically an insulator. From systematic calculations of the trivalent dopants as well as hydrogen, Bi doping is suggested as the best candidate as an acceptor dopant. This work paves the way for the material design of the $p$-type $\beta$-TeO$_2$.

In this paper, we consider the self-affinity of planar curves. It is regarded as an important property to characterize the log-aesthetic curves which have been studied as reference curves or guidelines for designing aesthetic shapes in CAD systems. We reformulate the two different self-affinities proposed in the development of log-aesthetic curves. We give rigorous proof that one self-affinity actually characterizes log-aesthetic curves, while another one characterizes parabolas. We then propose a new self-affinity which, in equiaffine geometry, characterizes the constant curvature curves (the quadratic curves). It integrates the two self-affinities, by which constant curvature curves in similarity and equiaffine geometries are characterized in a unified manner., Comment: 19 pages, 5 figures

We propose a meta-learning method for positive and unlabeled (PU) classification, which improves the performance of binary classifiers obtained from only PU data in unseen target tasks. PU learning is an important problem since PU data naturally arise in real-world applications such as outlier detection and information retrieval. Existing PU learning methods require many PU data, but sufficient data are often unavailable in practice. The proposed method minimizes the test classification risk after the model is adapted to PU data by using related tasks that consist of positive, negative, and unlabeled data. We formulate the adaptation as an estimation problem of the Bayes optimal classifier, which is an optimal classifier to minimize the classification risk. The proposed method embeds each instance into a task-specific space using neural networks. With the embedded PU data, the Bayes optimal classifier is estimated through density-ratio estimation of PU densities, whose solution is obtained as a closed-form solution. The closed-form solution enables us to efficiently and effectively minimize the test classification risk. We empirically show that the proposed method outperforms existing methods with one synthetic and three real-world datasets., Comment: 21 pages

Semi-supervised anomaly detection, which aims to improve the performance of the anomaly detector by using a small amount of anomaly data in addition to unlabeled data, has attracted attention. Existing semi-supervised approaches assume that unlabeled data are mostly normal. They train the anomaly detector to minimize the anomaly scores for the unlabeled data, and to maximize those for the anomaly data. However, in practice, the unlabeled data are often contaminated with anomalies. This weakens the effect of maximizing the anomaly scores for anomalies, and prevents us from improving the detection performance. To solve this problem, we propose the positive-unlabeled autoencoder, which is based on positive-unlabeled learning and the anomaly detector such as the autoencoder. With our approach, we can approximate the anomaly scores for normal data using the unlabeled and anomaly data. Therefore, without the labeled normal data, we can train the anomaly detector to minimize the anomaly scores for normal data, and to maximize those for the anomaly data. In addition, our approach is applicable to various anomaly detectors such as the DeepSVDD. Experiments on various datasets show that our approach achieves better detection performance than existing approaches., Comment: Under revirew. Code is available at https://github.com/takahashihiroshi/puae

External pressure and associated changes in lattice structures are key to realizing exotic quantum phases such as high-$T_{\rm c}$ superconductivity. While applying external pressure is a standard method to induce novel lattice structures, its impact on organic molecular crystals has been less explored. Here we report a unique structural phase transition in (DMET-TTF)$_2$AuBr$_2$ under pressure. By combining advanced high-pressure techniques and $ab$ $initio$ calculations, we elucidate that (DMET-TTF)$_2$AuBr$_2$ undergoes a transition from a quasi-one-dimensional lattice to a nearly perfect rectangular lattice at 0.9 GPa. This transition leads to the realization of an antiferromagnetic Mott insulator with $T_{\rm N}=66$ K, the highest $T_{\rm N}$ in low-dimensional molecular crystal solids to date. Upon increasing the pressure, the antiferromagnetic ordering is suppressed, and a superconducting phase with $T_{\rm c}=4.8$ K emerges around 6 GPa. Our study reveals the significant impact of external pressure on lattice structures of organic molecular crystals and highlights the intricate relationship between geometrical frustration and superconductivity. Our findings also pave the way for realizing functional organic molecular crystals through changes in lattice structures by pressure., Comment: 11 pages, 7 figures, 3 Tables

We consider the Gelfand problem with general supercritical nonlinearities in the two-dimensional unit ball. In this paper, we prove the non-existence of an unstable solution for any positive small parameter $\lambda$. The result implies that once the bifurcation curve emanates from the starting point, then the curve never approaches $\lambda=0$. As a result, we obtain the existence of a radial singular solution. In addition, we prove the uniformly boundedness of finite Morse index solutions. As a result, we prove that the bifurcation curve has infinitely many turning points. We remark that these properties are well-known in $N$ dimensions with $3\le N \le 9$ and less known in two dimensions. Our results clarify that the bifurcation structure is solely determined by the supercriticality of the nonlinearities if $2\le N\le 9$., Comment: 16 pages

We address the problem of finite-horizon control of a discrete-time linear system, where the initial state distribution follows a Gaussian mixture model, the terminal state must follow a specified Gaussian distribution, and the state and control inputs must obey chance constraints. We show that, throughout the time horizon, the state and control distributions are fully characterized by Gaussian mixtures. We then formulate the cost, distributional terminal constraint, and affine/2-norm chance constraints on the state and control, as convex functions of the decision variables. This is leveraged to formulate the chance-constrained path planning problem as a single convex optimization problem. A numerical example demonstrates the effectiveness of the proposed method., Comment: Accepted to 2024 Conference on Decision and Control (CDC)

Let $V$ be an affine algebraic variety, and let $p\in V$ be a singular point. For a regular function $g$ on $V$ such that $g(p)=0$ and for a positive integer $n$, we consider the cyclic covering $\phi_n\: V_n \to V$ of degree $n$ branched along the hypersurface defined by $g$. We will prove that for sufficiently large $n$, the tangent cone of $V_n$ at $\phi_n^{-1}(p)$ is, as an affine variety, the product of the tangent cone of the branch locus and the affine line. In particular, the multiplicity of the singularity $\phi_n^{-1}(p) \in V_n$, which is a function of $n$ determined by $V$ and $g$, remains constant for sufficiently large $n$. This result generalizes Tomaru's theorem for normal surface singularities., Comment: 8 pages; To appear in Tsukuba J. Math

We establish the quenched local limit theorem for reversible random walk on $\Z^d$ (with $d\ge 2$) among stationary ergodic random conductances that permit jumps of arbitrary length. The proof is based on the weak parabolic Harnack inequalities and on-diagonal heat-kernel estimates for long-range random walks on general ergodic environments. In particular, this partly solves \cite[Open Problem 2.7]{BCKW}, where the quenched invariance principle was obtained. As a byproduct, we prove the maximal inequality with an extra tail term for long-range reversible random walks, which in turn yields the everywhere sublinear property for the associated corrector., Comment: 37 pages

Recently, Ainsworth et al. showed that using weight matching (WM) to minimize the $L_2$ distance in a permutation search of model parameters effectively identifies permutations that satisfy linear mode connectivity (LMC), where the loss along a linear path between two independently trained models with different seeds remains nearly constant. This paper analyzes LMC using WM, which is useful for understanding stochastic gradient descent's effectiveness and its application in areas like model merging. We first empirically show that permutations found by WM do not significantly reduce the $L_2$ distance between two models, and the occurrence of LMC is not merely due to distance reduction by WM itself. We then demonstrate that permutations can change the directions of the singular vectors, but not the singular values, of the weight matrices in each layer. This finding shows that permutations found by WM primarily align the directions of singular vectors associated with large singular values across models. This alignment brings the singular vectors with large singular values, which determine the model's functionality, closer between the original and merged models, allowing the merged model to retain functionality similar to the original models, thereby satisfying LMC. This paper also analyzes activation matching (AM) in terms of singular vectors and finds that the principle of AM is the same as that of WM. Finally, we analyze the difference between WM and the straight-through estimator (STE), a dataset-dependent permutation search method, and show that WM can be more advantageous than STE in achieving LMC among three or more models., Comment: 36 pages

We study a primal-dual (PD) reinforcement learning (RL) algorithm for online constrained Markov decision processes (CMDPs). Despite its widespread practical use, the existing theoretical literature on PD-RL algorithms for this problem only provides sublinear regret guarantees and fails to ensure convergence to optimal policies. In this paper, we introduce a novel policy gradient PD algorithm with uniform probably approximate correctness (Uniform-PAC) guarantees, simultaneously ensuring convergence to optimal policies, sublinear regret, and polynomial sample complexity for any target accuracy. Notably, this represents the first Uniform-PAC algorithm for the online CMDP problem. In addition to the theoretical guarantees, we empirically demonstrate in a simple CMDP that our algorithm converges to optimal policies, while baseline algorithms exhibit oscillatory performance and constraint violation.

Because of their high photon flux, X-ray free-electron lasers (FEL) allow to resolve the structure of individual nanoparticles via coherent diffractive imaging (CDI) within a single X-ray pulse. Since the inevitable rapid destruction of the sample limits the achievable resolution, a thorough understanding of the spatiotemporal evolution of matter on the nanoscale following the irradiation is crucial. We present a technique to track X-ray induced structural changes in time and space by recording two consecutive diffraction patterns of the same single, free-flying nanoparticle, acquired separately on two large-area detectors opposite to each other, thus examining both the initial and evolved particle structure. We demonstrate the method at the extreme ultraviolet (XUV) and soft X-ray Free-electron LASer in Hamburg (FLASH), investigating xenon clusters as model systems. By splitting a single XUV pulse, two diffraction patterns from the same particle can be obtained. For focus intensities of about $2\cdot10^{12}\,\text{W/cm}^2$ we observe still largely intact clusters even at the longest delays of up to 650 picoseconds of the second pulse, indicating that in the highly absorbing systems the damage remains confined to one side of the cluster. Instead, in case of five times higher flux, the diffraction patterns show clear signatures of disintegration, namely increased diameters and density fluctuations in the fragmenting clusters. Future improvements to the accessible range of dynamics and time resolution of the approach are discussed.

Although Gaussian processes (GPs) with deep kernels have been successfully used for meta-learning in regression tasks, its uncertainty estimation performance can be poor. We propose a meta-learning method for calibrating deep kernel GPs for improving regression uncertainty estimation performance with a limited number of training data. The proposed method meta-learns how to calibrate uncertainty using data from various tasks by minimizing the test expected calibration error, and uses the knowledge for unseen tasks. We design our model such that the adaptation and calibration for each task can be performed without iterative procedures, which enables effective meta-learning. In particular, a task-specific uncalibrated output distribution is modeled by a GP with a task-shared encoder network, and it is transformed to a calibrated one using a cumulative density function of a task-specific Gaussian mixture model (GMM). By integrating the GP and GMM into our neural network-based model, we can meta-learn model parameters in an end-to-end fashion. Our experiments demonstrate that the proposed method improves uncertainty estimation performance while keeping high regression performance compared with the existing methods using real-world datasets in few-shot settings.

Here, using low-temperature optical scanning tunneling microscopy (STM), we investigate inelastic light scattering (ILS) in the vicinity of a single-atom quantum point contact (QPC). A vibration mode localized at the single Ag adatom on the Ag(111) surface is resolved in the ILS spectrum, resulting from tip-enhanced Raman scattering (TERS) by the atomically-confined plasmonic field in the STM junction. Furthermore, we trace how TERS from the single adatom evolves as a function of the gap distance. The exceptional stability of the low-temperature STM allows to examine distinctly different electron transport regimes of the picocavity, namely in the tunneling and quantum point contact (QPC) regimes. This measurement shows that the vibration mode localized at the adatom and its TERS intensity exhibits a sharp change upon the QPC formation, indicating that the atomic-level structure has a crucial impact on the plasmonic properties. To gain microscopic insights into picocavity optomechanics, we scrutinize the structure and plasmonic field in the STM junction using time-dependent density functional theory. The simulations reveal that atomic-scale structural relaxation at the single-atom QPC results in a discrete change of the plasmonic field strength, volume, and distribution as well as the vibration mode localized at the single atom. These findings give a qualitative explanation for the experimental observations. Furthermore, we demonstrate that strong ILS is a characteristic feature of QPC by continuously forming, breaking, and reforming the atomic contact, and how the plasmonic resonance evolves throughout the non-tunneling, tunneling, and QPC regimes., Comment: 23 pages, 5 figures

Research automation efforts usually employ AI as a tool to automate specific tasks within the research process. To create an AI that truly conduct research themselves, it must independently generate hypotheses, design verification plans, and execute verification. Therefore, we investigated if an AI itself could autonomously generate and verify hypothesis for a toy machine learning research problem. We prompted GPT-4 to generate hypotheses and Python code for hypothesis verification with limited methodological guidance. Our findings suggest that, in some instances, GPT-4 can autonomously generate and validate hypotheses without detailed guidance. While this is a promising result, we also found that none of the verifications were flawless, and there remain significant challenges in achieving autonomous, human-level research using only generic instructions. These findings underscore the need for continued exploration to develop a general and autonomous AI researcher.

We propose a meta-learning method for semi-supervised learning that learns from multiple tasks with heterogeneous attribute spaces. The existing semi-supervised meta-learning methods assume that all tasks share the same attribute space, which prevents us from learning with a wide variety of tasks. With the proposed method, the expected test performance on tasks with a small amount of labeled data is improved with unlabeled data as well as data in various tasks, where the attribute spaces are different among tasks. The proposed method embeds labeled and unlabeled data simultaneously in a task-specific space using a neural network, and the unlabeled data's labels are estimated by adapting classification or regression models in the embedding space. For the neural network, we develop variable-feature self-attention layers, which enable us to find embeddings of data with different attribute spaces with a single neural network by considering interactions among examples, attributes, and labels. Our experiments on classification and regression datasets with heterogeneous attribute spaces demonstrate that our proposed method outperforms the existing meta-learning and semi-supervised learning methods.