A5101463446- Domain Adaptation and Few-Shot Learning
- Multimodal Machine Learning Applications
- Advanced Image and Video Retrieval Techniques
- Cancer-related molecular mechanisms research
- Advanced Neural Network Applications
- COVID-19 diagnosis using AI
- Radiomics and Machine Learning in Medical Imaging
- Glioma Diagnosis and Treatment
- Adversarial Robustness in Machine Learning
- Lung Cancer Diagnosis and Treatment
- Generative Adversarial Networks and Image Synthesis
- Cancer Immunotherapy and Biomarkers
- Anomaly Detection Techniques and Applications
- Immunotherapy and Immune Responses
- Video Surveillance and Tracking Methods
- Immune cells in cancer
- Topic Modeling
- AI in cancer detection
- Ferroptosis and cancer prognosis
- Advanced Vision and Imaging
- Medical Imaging Techniques and Applications
- Remote-Sensing Image Classification
- Neurofibromatosis and Schwannoma Cases
- Human Pose and Action Recognition
- Speech Recognition and Synthesis
Omron (Japan)
2024
Boston University
2019-2023
Google (United States)
2023
Fujita Health University
2020-2022
Fujita Health University Hospital
2019-2021
The University of Tokyo
2013-2020
Kyorin University
2018-2019
University of Tokyo Hospital
2018
Jichi Medical University
2018
Kyoto University
2017
In this work, we present a method for unsupervised domain adaptation. Many adversarial learning methods train classifier networks to distinguish the features as either source or target and feature generator network mimic discriminator. Two problems exist with these methods. First, only tries thus does not consider task-specific decision boundaries between classes. Therefore, trained can generate ambiguous near class boundaries. Second, aim completely match distributions different domains,...
We propose an approach for unsupervised adaptation of object detectors from label-rich to label-poor domains which can significantly reduce annotation costs associated with detection. Recently, approaches that align distributions source and target images using adversarial loss have been proven effective adapting classifiers. However, detection, fully matching the entire each other at global image level may fail, as could distinct scene layouts different combinations objects. On hand, strong...
Contemporary domain adaptation methods are very effective at aligning feature distributions of source and target domains without any supervision. However, we show that these techniques perform poorly when even a few labeled examples available in the domain. To address this semi-supervised (SSDA) setting, propose novel Minimax Entropy (MME) approach adversarially optimizes an adaptive few-shot model. Our base model consists encoding network, followed by classification layer computes features'...
Deep-layered models trained on a large number of labeled samples boost the accuracy many tasks. It is important to apply such different domains because collecting in various expensive. In unsupervised domain adaptation, one needs train classifier that works well target when provided with source and unlabeled samples. Although methods aim match distributions samples, simply matching distribution cannot ensure domain. To learn discriminative representations for domain, we assume artificially...
We present a method for transferring neural representations from label-rich source domains to unlabeled target domains. Recent adversarial methods proposed this task learn align features across by fooling special domain critic network. However, drawback of approach is that the simply labels generated as in-domain or not, without considering boundaries between classes. This can lead ambiguous being near class boundaries, reducing classification accuracy. propose novel approach, Adversarial...
Unsupervised domain adaptation methods traditionally assume that all source categories are present in the target domain. In practice, little may be known about category overlap between two domains. While some address settings with either partial or open-set categories, they particular setting is a priori. We propose more universally applicable framework can handle arbitrary shift, called Domain Adaptative Neighborhood Clustering via Entropy optimization (DANCE). DANCE combines novel ideas:...
Mutations in H3F3A, which encodes histone H3.3, commonly occur pediatric glioblastoma. Additionally, H3F3A K27M substitutions gliomas that arise at midline locations (eg, pons, thalamus, spine); moreover, this substitution occurs mainly tumors children and adolescents. Here, we sought to determine the association between mutations adult thalamic glioma. Genomic was sequenced from 20 separate gliomas. for 14 of gliomas, 639 genes—including cancer-related genes chromatin-modifier genes—were...
Universal Domain Adaptation (UNDA) aims to handle both domain-shift and category-shift between two datasets, where the main challenge is transfer knowledge while rejecting "unknown" classes which are absent in labeled source data but present unlabeled target data. Existing methods manually set a threshold reject samples based on validation or pre-defined ratio of samples, this strategy not practical. In paper, we propose method learn thresh-old using adapt it domain. Our idea that minimum...
Lung cancer is a leading cause of death worldwide. Although computed tomography (CT) examinations are frequently used for lung diagnosis, it can be difficult to distinguish between benign and malignant pulmonary nodules on the basis CT images alone. Therefore, bronchoscopic biopsy may conducted if malignancy suspected following examinations. However, biopsies highly invasive, patients with undergo many unnecessary biopsies. To prevent this, an imaging diagnosis high classification accuracy...
In Composed Image Retrieval (CIR), a user combines query image with text to describe their intended target. Existing methods rely on supervised learning of CIR models using labeled triplets consisting the image, specification, and target image. Labeling such is expensive hinders broad applicability CIR. this work, we propose study an important task, Zero-Shot (ZS-CIR), whose goal build model without requiring for training. To end, novel method, called Pic2Word, that requires only weakly...
Abstract BACKGROUND Wearable devices with heads-up displays or smart glasses can overlay images onto the sight of wearer. This technology has never been applied to surgical navigation. OBJECTIVE To assess applicability and accuracy for augmented reality (AR)-based neurosurgical METHODS Smart were AR-based Three-dimensional computer graphics created based on preoperative magnetic resonance visualized in see-through glasses. Optical markers attached patient's head accurate Two motion capture...
In this paper, we address the problem of spatio-temporal person retrieval from videos using a natural language query, in which output tube (i.e., sequence bounding boxes) encloses described by query. For problem, introduce novel dataset consisting containing people annotated with boxes for each second and five descriptions. To retrieve given design model that combines methods human detection multimodal retrieval. We conduct comprehensive experiments to compare variety text representations...
Immune-based therapies have shown limited efficacy in glioma thus far. This might be at least part due to insufficient numbers of neoantigens, thought targets immune attack. In addition, we hypothesized that dynamic genetic and epigenetic tumor evolution gliomas also affect the mutation/neoantigen landscape contribute treatment resistance through evasion. Here, investigated changes neoantigen immunologic features during progression using exome RNA-seq paired primary recurrent samples...
Artificial intelligence (AI) applications in medical imaging continue facing the difficulty collecting and using large datasets. One method proposed for solving this problem is data augmentation fictitious images generated by generative adversarial networks (GANs). However, applying a GAN as technique has not been explored, owing to quality diversity of images. To promote such generating diverse images, study aims generate free-form lesion from tumor sketches pix2pix-based model, which an...
Visual question answering (VQA) tasks use two types of images: abstract (illustrations) and real. Domain-specific differences exist between the images with respect to "objectness," "texture," "color." Therefore, achieving similar performance by applying methods developed for real images, vice versa, is difficult. This a critical problem in VQA, because image features are crucial clues correctly questions about images. However, an effective, domain-invariant method can provide insight into...
Unsupervised transfer of object recognition models from synthetic to real data is an important problem with many potential applications. The challenge how "adapt" a model trained on simulated images so that it performs well real-world without any additional supervision. Unfortunately, current benchmarks for this are limited in size and task diversity. In paper, we present new large-scale benchmark called Syn2Real, which consists domain rendered 3D two real-image domains containing the same...
We present a two-stage pre-training approach that improves the generalization ability of standard single-domain pre-training. While on single large dataset (such as ImageNet) can provide good initial representation for transfer learning tasks, this may result in biased representations impact success with new multi-domain data (e.g., different artistic styles) via methods like domain adaptation. propose novel called Cross-Domain Self-supervision (CDS), which directly employs unlabeled...
Existing vision-language methods typically support two languages at a time most. In this paper, we present modular approach which can easily be incorporated into existing in order to many languages. We accomplish by learning single shared Multimodal Universal Language Embedding (MULE) has been visually-semantically aligned across all Then learn relate MULE visual data as if it were language. Our method is not architecture specific, unlike prior work learned separate branches for each...
Existing unsupervised domain adaptation methods aim to transfer knowledge from a label-rich source an unlabeled target domain. However, obtaining labels for some domains may be very expensive, making complete labeling as used in prior work impractical. In this work, we investigate new scenario with sparsely labeled data, where only few examples the have been labeled, while is unlabeled. We show that when are limited, existing often fail learn discriminative features applicable both and...