A5010759340- Neural Networks and Applications
- Model Reduction and Neural Networks
- Advanced Graph Neural Networks
- Machine Learning in Materials Science
- RNA modifications and cancer
- Machine Learning in Healthcare
- Computational Drug Discovery Methods
- RNA and protein synthesis mechanisms
- Advanced Memory and Neural Computing
- Complex Network Analysis Techniques
- Advanced Neural Network Applications
- Generative Adversarial Networks and Image Synthesis
- Stock Market Forecasting Methods
- Domain Adaptation and Few-Shot Learning
- Nutritional Studies and Diet
- Machine Learning and Algorithms
- Finite Group Theory Research
- Dietary Effects on Health
- COVID-19 diagnosis using AI
- MicroRNA in disease regulation
- Cancer-related molecular mechanisms research
- Image and Signal Denoising Methods
- Chemical Synthesis and Analysis
- Machine Learning in Bioinformatics
- Obesity, Physical Activity, Diet
Preferred Networks (Japan)
2015-2024
Ehime University
2020
RIKEN
2020
The University of Tokyo
2019-2020
Graph Neural Networks (graph NNs) are a promising deep learning approach for analyzing graph-structured data. However, it is known that they do not improve (or sometimes worsen) their predictive performance as we pile up many layers and add non-lineality. To tackle this problem, investigate the expressive power of graph NNs via asymptotic behaviors layer size tends to infinity. Our strategy generalize forward propagation Convolutional Network (GCN), which popular NN variant, specific...
Automatic design with machine learning and molecular simulations has shown a remarkable ability to generate new promising drug candidates. Current models, however, still have problems in simulation concurrency diversity. Most methods one molecule at time do not allow multiple simulators run simultaneously. Additionally, better diversity could boost the success rate subsequent discovery process. We propose population-based approach using grammatical evolution named ChemGE. In our method,...
Abstract Background Noninvasive detection of early stage cancers with accurate prediction tumor tissue-of-origin could improve patient prognosis. Because miRNA profiles differ between organs, circulating miRNomics represent a promising method for cancers, but this has not been shown conclusively. Methods A serum profile (miRNomes)–based classifier was evaluated its ability to discriminate cancer types using advanced machine learning. The training set comprised 7931 samples from patients 13...
Data for healthcare applications are typically customized specific purposes but often difficult to access due high costs and privacy concerns. Rather than prepare separate datasets individual applications, we propose a novel approach: building general-purpose generative model applicable virtually any type of application. This encompasses broad range human attributes, including age, sex, anthropometric measurements, blood components, physical performance metrics, numerous healthcare-related...
Invertible neural networks based on coupling flows (CF-INNs) have various machine learning applications such as image synthesis and representation learning. However, their desirable characteristics analytic invertibility come at the cost of restricting functional forms. This poses a question power: are CF-INNs universal approximators for invertible functions? Without universality, there could be well-behaved transformation that CF-INN can never approximate, hence it would render model class...
Statistical generative models for molecular graphs attract attention from many researchers the fields of bio- and chemo-informatics. Among these models, invertible flow-based approaches are not fully explored yet. In this paper, we propose a powerful flow graphs, called graph residual (GRF). The GRF is based on flows, which known more flexible complex non-linear mappings than traditional coupling flows. We theoretically derive non-trivial conditions such that invertible, present way keeping...
The design of RNA plays a crucial role in developing vaccines, nucleic acid therapeutics, and innovative biotechnological tools. However, existing techniques frequently lack versatility across various tasks are dependent on pre-defined secondary structure or other prior knowledge. To address these limitations, we introduce GenerRNA, Transformer-based model inspired by the success large language models (LLMs) protein molecule generation. GenerRNA is pre-trained large-scale sequences capable...
Convolutional neural networks (CNNs) have been shown to achieve optimal approximation and estimation error rates (in minimax sense) in several function classes. However, previous analyzed CNNs are unrealistically wide difficult obtain via optimization due sparse constraints important classes, including the H\"older class. We show a ResNet-type CNN can attain these classes more plausible situations -- it be dense, its width, channel size, filter size constant with respect sample size. The key...
A bstract The design of RNA plays a crucial role in developing vaccines, nucleic acid therapeutics, and innovative biotechnological tools. Nevertheless, existing techniques lack versatility across various tasks frequently suffer from deficiency automated generation. Inspired by the remarkable success Large Language Models (LLMs) realm protein molecule design, we present GenerRNA, first large-scale pre-trained model for generation, aiming to further automate design. Our approach eliminates...
Neural ordinary differential equations (NODEs) is an invertible neural network architecture promising for its free-form Jacobian and the availability of a tractable determinant estimator. Recently, representation power NODEs has been partly uncovered: they form $L^p$-universal approximator continuous maps under certain conditions. However, $L^p$-universality may fail to guarantee approximation entire input domain as it still hold even if largely differs from target function on small region...
With the rapid increase of compound databases available in medicinal and material science, there is a growing need for learning representations molecules semi-supervised manner. In this paper, we propose an unsupervised hierarchical feature extraction algorithm (or more generally, graph-structured objects with fixed number types nodes edges), which applicable to both tasks. Our method extends recently proposed Paragraph Vector incorporates neural message passing obtain subgraphs. We applied...
Invertible neural networks (INNs) are network architectures with invertibility by design. Thanks to their and the tractability of Jacobian, INNs have various machine learning applications such as probabilistic modeling, generative representation learning. However, attractive properties often come at cost restricting layer designs, which poses a question on power: can we use these models approximate sufficiently diverse functions? To answer this question, developed general theoretical...
It is known that the current graph neural networks (GNNs) are difficult to make themselves deep due problem as over-smoothing. Multi-scale GNNs a promising approach for mitigating over-smoothing problem. However, there little explanation of why it works empirically from viewpoint learning theory. In this study, we derive optimization and generalization guarantees transductive algorithms include multi-scale GNNs. Using boosting theory, prove convergence training error under weak learning-type...
A graph neural network (GNN) is a good choice for predicting the chemical properties of molecules. Compared with other deep networks, however, current performance GNN limited owing to "curse depth." Inspired by long-established feature engineering in field chemistry, we expanded an atom representation using Weisfeiler-Lehman (WL) embedding, which designed capture local atomic patterns dominating molecule. In terms representability, show WL embedding can replace first two layers ReLU --...
Abstract We study testing properties of functions on finite groups. First we consider the form , where G is a group. show that conjugate invariance, homomorphism, and property being proportional to an irreducible character testable with constant number queries f crucial notion in representation theory. Our proof relies theory harmonic analysis Next d fixed family by matrices each element . For function unitary isomorphism g say are isomorphic if there exists matrix U such for any © 2016...
We study testing properties of functions on finite groups. First we consider the form $f:G \to \mathbb{C}$, where $G$ is a group. show that conjugate invariance, homomorphism, and property being proportional to an irreducible character testable with constant number queries $f$, crucial notion in representation theory. Our proof relies theory harmonic analysis Next $f: G M_d(\mathbb{C})$, $d$ fixed $M_d(\mathbb{C})$ family by matrices each element $\mathbb{C}$. For function $g:G unitary...
<sec> <title>BACKGROUND</title> Human health status can be measured on the basis of many different parameters. Statistical relationships among these parameters will enable several possible care applications and an approximation current individuals, which allow for more personalized preventive by informing potential risks developing interventions. Furthermore, a better understanding modifiable risk factors related to lifestyle, diet, physical activity facilitate design optimal treatment...
We present \emph{TabRet}, a pre-trainable Transformer-based model for tabular data. TabRet is designed to work on downstream task that contains columns not seen in pre-training. Unlike other methods, has an extra learning step before fine-tuning called \emph{retokenizing}, which calibrates feature embeddings based the masked autoencoding loss. In experiments, we pre-trained with large collection of public health surveys and fine-tuned it classification tasks healthcare, achieved best AUC...
Human health status can be measured on the basis of many different parameters. Statistical relationships among these parameters will enable several possible care applications and an approximation current individuals, which allow for more personalized preventive by informing potential risks developing interventions. Furthermore, a better understanding modifiable risk factors related to lifestyle, diet, physical activity facilitate design optimal treatment approaches individuals.This study...
Variational autoencoders (VAEs) are one of the deep generative models that have experienced enormous success over past decades. However, in practice, they suffer from a problem called posterior collapse, which occurs when encoder coincides, or collapses, with prior taking no information latent structure input data into consideration. In this work, we introduce an inverse Lipschitz neural network decoder and, based on architecture, provide new method can control simple and clear manner degree...