公开(公告)号：WO2022086728A1

公开(公告)日：2022-04-28

申请号：PCT/US2021/054142

申请日：2021-10-08

Applicant: NEC LABORATORIES AMERICA, INC.

Inventor： SUH, Yumin ,  YU, Xiang ,  FARAKI, Masoud ,  CHANDRAKER, Manmohan ,  DENG, Weijian

IPC: G06N3/08 ,  G06N3/04 ,  G06N3/063

Abstract: A method for multi-task learning via gradient split for rich human analysis is presented. The method includes extracting (1001) images from training data having a plurality of datasets, each dataset associated with one task, feeding (1003) the training data into a neural network model including a feature extractor and task- specific heads, wherein the feature extractor has a feature extractor shared component and a feature extractor task- specific component, dividing (1005) filters of deeper layers of convolutional layers of the feature extractor into N groups, N being a number of tasks, assigning (1007) one task to each group of the N groups, and manipulating (1009) gradients so that each task loss updates only one subset of filters.

公开(公告)号：WO2023086196A1

公开(公告)日：2023-05-19

申请号：PCT/US2022/047514

申请日：2022-10-24

Applicant: NEC LABORATORIES AMERICA, INC.

Inventor： FARAKI, Masoud ,  TSAI, Yi-Hsuan ,  YU, Xiang ,  SCHULTER, Samuel ,  SUH, Yumin ,  SIMON, Christian

IPC: G06N3/08 ,  G06N3/04

Abstract: A computer-implemented method for model training is provided. The method includes receiving, by a hardware processor, sets of images, each set corresponding to a respective task. The method further includes training, by the hardware processor, a task-based neural network classifier having a center and a covariance matrix for each of a plurality of classes in a last layer of the task-based neural network classifier and a plurality of convolutional layers preceding the last layer, by using a similarity between an image feature of a last convolutional layer from among the plurality of convolutional layers and the center and the covariance matrix for a given one of the plurality of classes, the similarity minimizing an impact of a data model forgetting problem.

公开(公告)号：WO2022103748A1

公开(公告)日：2022-05-19

申请号：PCT/US2021/058612

申请日：2021-11-09

Applicant: NEC LABORATORIES AMERICA, INC.

Inventor： YU, Xiang ,  TSAI, Yi-Hsuan ,  FARAKI, Masoud ,  MOSLEMI, Ramin ,  CHANDRAKER, Manmohan ,  LIU, Chang

IPC: G06N3/08 ,  G06N3/04 ,  G06K9/00

Abstract: A method for training a model for face recognition is provided. The method forward trains (610) a training batch of samples to form a face recognition model w(t), and calculates (620) sample weights for the batch. The method obtains (630) a training batch gradient with respect to model weights thereof and updates, using the gradient, the model w(t) to a face recognition model what(t). The method forwards (640) a validation batch of samples to the face recognition model what(t). The method obtains (650) a validation batch gradient, and updates, using the validation batch gradient and what(t), a sample-level importance weight of samples in the training batch to obtain an updated sample-level importance weight. The method obtains (660) a training batch upgraded gradient based on the updated sample-level importance weight of the training batch samples, and updates (660), using the upgraded gradient, the model w(t) to a trained model w(t+l) corresponding to a next iteration.

公开(公告)号：WO2022103682A1

公开(公告)日：2022-05-19

申请号：PCT/US2021/058404

申请日：2021-11-08

Applicant: NEC LABORATORIES AMERICA, INC.

Inventor： FARAKI, Masoud ,  YU, Xiang ,  TSAI, Yi-Hsuan ,  SUH, Yumin ,  CHANDRAKER, Manmohan

IPC: G06K9/00 ,  G06N3/08

Abstract: A method for improving face recognition from unseen domains including obtaining (1001) face images with associated identities from a plurality of datasets, randomly (1003) selecting two datasets of the plurality of datasets to train a model, sampling (1005) batch face images and their corresponding labels, sampling (1007) triplet samples including one anchor face image, a sample face image from a same identity, and a sample face image from a different identity than that of the one anchor face image, performing (1009) a forward pass by using the samples of the selected two datasets, finding (1011) representations of the face images by using a backbone convolutional neural network (CNN), generating (1013) covariances from the representations of the face images and the backbone CNN, the covariances made in different spaces by using positive pairs and negative pairs, and employing (1015) the covariances to compute a cross-domain similarity loss function.

公开(公告)号：WO2022103684A1

公开(公告)日：2022-05-19

申请号：PCT/US2021/058407

申请日：2021-11-08

Applicant: NEC LABORATORIES AMERICA, INC.

Inventor： SUH, Yumin ,  YU, Xiang ,  TSAI, Yi-Hsuan ,  FARAKI, Masoud ,  CHANDRAKER, Manmohan

IPC: G06K9/00

Abstract: A method for employing facial information in unsupervised person re-identification is presented. The method includes extracting (1001), by a body feature extractor, body features from a first data stream, extracting (1003), by a head feature extractor, head features from a second data stream, outputting (1005) a body descriptor vector from the body feature extractor, outputting (1007) a head descriptor vector from the head feature extractor, and concatenating (1009) the body descriptor vector and the head descriptor vector to enable a model to generate a descriptor vector.

公开(公告)号：WO2022072776A1

公开(公告)日：2022-04-07

申请号：PCT/US2021/053086

申请日：2021-10-01

Applicant: NEC LABORATORIES AMERICA, INC.

Inventor： YU, Xiang ,  TSAI, Yi-Hsuan ,  PITTALUGA, Francesco ,  FARAKI, Masoud ,  CHANDRAKER, Manmohan ,  ZHU, Yuqing

IPC: G06N20/20 ,  G06N5/02

Abstract: A method for employing a general label space voting-based differentially private federated learning (DPFL) framework is presented. The method includes labeling (1010) a first subset of unlabeled data from a first global server, to generate first pseudo-labeled data, by employing a first voting-based DPFL computation where each agent trains a local agent model by using private local data associated with the agent, labeling (1020) a second subset of unlabeled data from a second global server, to generate second pseudo-labeled data, by employing a second voting-based DPFL computation where each agent maintains a data- independent feature extractor, and training (1030) a global model by using the first and second pseudo-labeled data to provide provable differential privacy (DP) guarantees for both instance-level and agent- level privacy regimes.

公开(公告)号：WO2021243029A1

公开(公告)日：2021-12-02

申请号：PCT/US2021/034528

申请日：2021-05-27

Applicant: NEC LABORATORIES AMERICA, INC.

Inventor： CHANDRAKER, Manmohan ,  WANG, Ting ,  XU, Xiang ,  PITTALUGA, Francesco ,  SHARMA, Gaurav ,  TSAI, Yi-Hsuan ,  FARAKI, Masoud ,  CHEN, Yuheng ,  TIAN, Yue ,  HUANG, Ming-Fang ,  FANG, Jian

IPC: G06K9/00 ,  G06N3/08

Abstract: Methods and systems for training a neural network include generating (801) an image of a mask. A copy of an image is generated (302) from an original set of training data. The copy is altered (302) to add the image of a mask to a face detected within the copy. An augmented set of training data is generated (302) that includes the original set of training data and the altered copy. A neural network model is trained (304) to recognize masked faces using the augmented set of training data.