-
1.发明申请
公开(公告)号：WO2023092008A1
公开(公告)日：2023-05-25
申请号：PCT/US2022/080046
申请日：2022-11-17
Applicant: UNIVERSITY OF SOUTHERN CALIFORNIA
Inventor： WEN, Shixian , ITTI, Laurent
Abstract: An apparatus and method for training a brain computer interface (BCI) decoder to output kinematics data from neural spike trains is disclosed. A neural spike synthesizer generator based on a LSTM neural network provides a direct mapping from input kinematics data from a first session with a first subject to a first synthesized spike train. The neural spike synthesizer generator is adapted to produce a second synthesized spike train suitable for a second subject based on input kinematics data from the second subject and fine tuning using observed real neural data of the second subject or a second session of the first subject based on input kinematics data from the second session and fine tuning using observed real neural data of the second session. The BCI decoder is trained for the second subject or the second session using an input of the observed real neural data and the second synthesized spike train and a ground set of kinematics data.
-
公开(公告)号：WO2023091970A1
公开(公告)日：2023-05-25
申请号：PCT/US2022/079989
申请日：2022-11-16
Applicant: THE GENERAL HOSPITAL CORPORATION , THE BROAD INSTITUTE, INC. , MASSACHUSETTS INSTITUTE OF TECHNOLOGY , COMITER, Charles
Inventor： SHU, Jian , REGEV, Aviv , KOBAYASHI-KIRSCHVINK, Koseki , BIANCALANI, Tommaso
Abstract: Computer-implemented methods, computer program products, and systems determine an omics profiles of a cell using microscopy imaging data. In one aspect, a computer-implemented method determines an omics profiles of a cell using microscopy imaging data by a) receiving microscopy imaging data of a cell or a population of cells; b) determining a targeted expression profile of a set of target genes from the microscopy imaging data using a first machine learning model, the target genes identifying a cell type or cell state of interest; and c) determining a singlecell omics profile for the population of cells using a second machine learning algorithm model. The targeted expression profile and a reference single-cell RNA-seq data set are used as inputs for the second machine learning model.
-
公开(公告)号：WO2023090967A1
公开(公告)日：2023-05-25
申请号：PCT/KR2022/018409
申请日：2022-11-21
Applicant: 에스케이플래닛 주식회사
Abstract: 본 발명의 결측값을 보간하기 위한 방법은 데이터처리부가 센서 신호를 구성하는 복수의 단위 신호 중 결측 부분 없이 온전한 전측 신호를 선별한 데이터 세트를 수집하는 단계와, 학습부가 상기 데이터 세트를 이용하여 센서 신호의 적어도 일부가 결측된 결측 부분을 가지는 결측 신호에서 결측 부분을 보간하는 보간망을 학습시키는 단계와, 보간부가 센서 신호의 적어도 일부가 결측된 결측 신호를 입력 받는 단계와, 상기 보간부가 상기 보간망을 이용하여 상기 결측 부분을 보간하여 보간 신호를 생성하는 단계를 포함한다.
-
公开(公告)号：WO2023090819A1
公开(公告)日：2023-05-25
申请号：PCT/KR2022/017997
申请日：2022-11-15
Applicant: 삼성전자 주식회사
Abstract: 본 개시는 영상에서 인지적 노이즈를 제거하는 영상 처리 방법 및 장치에 관한 것이다. 개시되는 영상 처리 방법은, 영상으로부터, 노이즈가 제거된 영상을 획득하는 단계; 상기 영상과 상기 노이즈가 제거된 영상의 차이로부터, 제거된 노이즈를 획득하는 단계; 상기 노이즈가 제거된 영상의 통계적 특징 및 상기 노이즈의 통계적 특징을 획득하는 단계; 상기 노이즈가 제거된 영상의 통계적 특징 및 상기 노이즈의 통계적 특징에 기초하여, 화소의 노이즈가 인지적 노이즈인지 또는 인지적 디테일인지에 관한 노이즈 특성을 판단하는 단계; 및 상기 화소의 노이즈 특성에 기초하여 상기 영상에서 인지적 노이즈를 제거하는 처리를 수행하는 단계;를 포함한다.
-
公开(公告)号：WO2023090499A1
公开(公告)日：2023-05-25
申请号：PCT/KR2021/017227
申请日：2021-11-23
Applicant: 서울과학기술대학교 산학협력단
Abstract: 딥러닝 네트워크의 프루닝 방법에 있어서, 복수의 연속적인 컨볼루션 레이어인 제1 레이어 및 제2 레이어 중 상기 제1 레이어에서, 배치 정규화 감마 파라미터 값에 근거하여 상기 제1 레이어에 포함되는 복수의 커널 중 적어도 하나의 프루닝 대상을 선택하는 단계; 상기 선택된 프루닝 대상 중 어느 하나의 커널을 트렁크로 설정하는 단계; 트렁크로 설정된 어느 하나의 커널을 소정의 수학식에 근거하여 업데이트하는 단계; 및 상기 선택된 프루닝 대상 중 트렁크로 설정되지 않은 나머지 커널과, 상기 제2 레이어에서 상기 나머지 커널과 대응되는 커널을 제거하는 단계를 포함한다.
-
Patent Agency Ranking6.发明申请
公开(公告)号：WO2023088665A1
公开(公告)日：2023-05-25
申请号：PCT/EP2022/080267
申请日：2022-10-28
Applicant: ABB SCHWEIZ AG
Inventor： ABUKWAIK, Hadil , SHARMA, Divyasheel , KLOEPPER, Benjamin , KOTRIWALA, Arzam Muzaffar , RODRIGUEZ, Pablo , SCHMIDT, Benedikt , TAN, Ruomu , K R, Chandrika , BORRISON, Reuben , DIX, Marcel , DOPPELHAMER, Jens
Abstract: A method (100) for training a prediction model (1) for predicting the likelihood that at least one predetermined undesired event will occur during execution of a process (2) using training samples (3), wherein each training sample (3) comprises data that characterizes a state of the process (2), and the method (100) comprises the steps of: obtaining (110) training samples (3) representing states of the process (2) that do not cause the undesired event, and labelling these training samples with a pre-set low likelihood of the undesired event occurring; obtaining (120), based at least in part on a process model (2a) and a set of predetermined rules (2b) that stipulate in which states of the process (2) there is an increased likelihood of the undesired event occurring, further training samples (4) representing states of the process (2) with an increased likelihood to cause the undesired event, and labelling these training samples (4) with said increased likelihood; providing (130) training samples (3, 4) to the to-be-trained prediction model (1), so as to obtain, from the prediction model (1), a prediction (5) of the likelihood for occurrence of the undesired event in a state of the process (2) represented by the respective sample (3, 4); rating (140) a difference between the prediction (5) and the label of the respective sample (3, 4) by means of a predetermined loss function (6); and optimizing (150) parameters (1a) that characterize the behavior of the prediction model (1), such that, when predictions (5) on further samples (3, 4) are made, the rating (6a) by the loss function (6) is likely to improve.
-
公开(公告)号：WO2023088562A1
公开(公告)日：2023-05-25
申请号：PCT/EP2021/082306
申请日：2021-11-19
Applicant: TELEFONAKTIEBOLAGET LM ERICSSON (PUBL)
Inventor： LI, Yun
Abstract: A computer-implemented method of training an autoencoder for data compression is provided. The method comprises inputting training data into a neural network to generate a plurality of first latent variables, using a probability estimator to obtain probability mass functions for the plurality of latent variables, and determining a rate measure based on the probability mass functions and a plurality of coded latent variables. The plurality of coded latent variables is based on one-hot encoding of plurality of quantized latent variables obtained by quantizing the plurality of latent variables. A gradient of the rate measure is determined based on the probability mass functions and a plurality of approximately coded latent variables obtained by applying a differentiable function to the plurality of latent variables to approximate one-hot encoding. One or more parameters of the autoencoder is updated based on the rate measure and the gradient of the rate measure.
-
公开(公告)号：WO2023088131A1
公开(公告)日：2023-05-25
申请号：PCT/CN2022/130549
申请日：2022-11-08
Applicant: 中移（上海）信息通信科技有限公司 , 中移智行网络科技有限公司 , 中国移动通信集团有限公司
Inventor： 鱼一帆
Abstract: 一种交通状态预测方法、装置、设备、介质及程序，涉及智能交通技术领域，所述方法包括：生成多个染色体单元（101），每个染色体单元用于表征一类时空卷积网络模型；基于样本集分别计算多个染色体单元中每个染色体单元对应的时空卷积网络模型的损失值（102）；依据多个染色体单元对应的时空卷积网络模型的损失值更新多个染色体单元，并返回执行基于样本集分别计算多个染色体单元中每个染色体单元对应的时空卷积网络模型的损失值的步骤，直至确定满足预设条件的目标染色体单元（103）；基于目标染色体单元对应的时空卷积网络模型确定预先训练的时空卷积网络模型（104）；基于预先训练的时空卷积网络模型预测交通状态（105）。上述方法能够提高预测交通状态的准确性。
-
公开(公告)号：WO2023087558A1
公开(公告)日：2023-05-25
申请号：PCT/CN2022/076475
申请日：2022-02-16
Applicant: 重庆邮电大学
IPC: G06V10/764 , G06V10/774 , G06V10/82 , G06V20/13 , G06K9/62 , G06N3/04 , G06N3/08
Abstract: 本发明涉及一种基于嵌入平滑图神经网络的小样本遥感图像场景分类方法，属于遥感图像识别领域。该方法首先将场景图片输入到嵌入学习模块中，通过一个卷积神经网络提取场景嵌入特征；再将嵌入平滑引入到场景分类中，在无监督的情况下捕获嵌入特征之间的相似性与差异性，提高嵌入特征的可区分性，扩展决策边界，降低无关特征的影响；同时通过注意力机制采用任务级关系来构建图矩阵，将目标样本与任务中的所有样本关联起来，并在不同场景类别之间产生更具有分辨力的关系表示；然后根据样本间的内在联系构造图；标签匹配模块可以根据构造的图，通过直推式学习迭代生成测试集中样本的预测标签，直到得到最优解。本发明能够实现图像的精确分类。
-
公开(公告)号：WO2023087303A1
公开(公告)日：2023-05-25
申请号：PCT/CN2021/132082
申请日：2021-11-22
Applicant: ROBERT BOSCH GMBH , TSINGHUA UNIVERSITY
Inventor： KHARLAMOV, Evgeny , TANG, Jie , FENG, Wenzheng
Abstract: The present disclosure provides a method for training a Graph Neural Network (GNN) model to perform a task of classifying nodes of a graph based on semi-supervised learning. The method comprises: sampling a batch of labeled nodes and a batch of unlabeled nodes from the nodes of the graph, wherein the graph comprising nodes represented by a feature matrix and edges represented by an adjacency matrix, each of the nodes of graph being represented by a corresponding feature vector of the feature matrix; obtaining a plurality of augmented feature vectors for each node in the batch of labeled nodes and the batch of unlabeled nodes by randomly propagating feature vectors of neighboring nodes of the node based on an adjacency vector of the node; obtaining a plurality of classification predictions for each node in the batch of labeled nodes and the batch of unlabeled nodes by respectively applying the plurality of augmented feature vectors of the node to the GNN model; obtaining a loss based on the classification predictions of the nodes in the batch of labeled nodes and the batch of unlabeled nodes; and updating parameters of the GNN model based on the loss.
-
-
-
-
-
-
-
-
-
Search Results
8065
records
(took 0.049 seconds)
