公开(公告)号：US20240211794A1

公开(公告)日：2024-06-27

申请号：US18079123

申请日：2022-12-12

申请人： INTERNATIONAL BUSINESS MACHINES CORPORATION ,  MASSACHUSETTS INSTITUTE OF TECHNOLOGY

发明人： Lam Minh Nguyen ,  Wang Zhang ,  Subhro Das ,  Alexandre Megretski ,  Luca Daniel

IPC分类号： G06N20/00

CPC分类号： G06N20/00

摘要： Providing a trained reinforcement learning (RL) model by formulating a decision process problem for the RL model, defining at least one of a logarithmic loss function for the RL model and defining an initiation point for the RL model according to an optimized spectral norm of the RL model, training the system according to the logarithmic loss function or from the initiation point, and providing the trained RL model.

公开(公告)号：US20240256837A1

公开(公告)日：2024-08-01

申请号：US18160737

申请日：2023-01-27

申请人： International Business Machines Corporation ,  MASSACHUSETTS INSTITUTE OF TECHNOLOGY

发明人： Lam Minh Nguyen ,  Wang Zhang ,  Subhro Das ,  Alexandre Megretski ,  Luca Daniel

IPC分类号： G06N3/0464 ,  G06N3/063

CPC分类号： G06N3/0464 ,  G06N3/063

摘要： One or more computer processors create a fully convolution network (FCN) comprising a plurality of 1×1 convolutions. The one or more computer processors append linear mapping layer (LM) to created FCN. The one or more computer processors capture a plurality of features utilizing multi-scale dilated convolutional kernels from the linear mapped FCN (LM-FCN). The one or more computer processors apply an average pool layer to the captured plurality of features along a temporal axis of a dilated convolutional kernel within the LM-FCN. The one or more computer processors predict a classification for subsequent time-series data utilizing the pooled plurality of features.

公开(公告)号：US20240096057A1

公开(公告)日：2024-03-21

申请号：US17933473

申请日：2022-09-19

申请人： INTERNATIONAL BUSINESS MACHINES CORPORATION ,  MASSACHUSETTS INSTITUTE OF TECHNOLOGY

发明人： Lam Minh Nguyen ,  Wang Zhang ,  Subhro Das ,  Pin-Yu Chen ,  Alexandre Megretski ,  Luca Daniel

IPC分类号： G06V10/764 ,  G06V10/774

CPC分类号： G06V10/764 ,  G06V10/774

摘要： A computer implemented method for certifying robustness of image classification in a neural network is provided. The method includes initializing a neural network model. The neural network model includes a problem space and a decision boundary. A processor receives a data set of images, image labels, and a perturbation schedule. Images are drawn from the data set in the problem space. A distance from the decision boundary is determined for the images in the problem space. A re-weighting value is applied to the images. A modified perturbation magnitude is applied to the images. A total loss function for the images in the problem space is determined using the re-weighting value. A confidence level of the classification of the images in the data set is evaluated for certifiable robustness.

公开(公告)号：US09595920B2

公开(公告)日：2017-03-14

申请号：US14760284

申请日：2014-03-13

申请人： Massachusetts Institute of Technology

发明人： Yan Li ,  Zhipeng Li ,  Alexandre Megretski ,  Vladimir Marko Stojanovic ,  Omer Tanovic ,  Yehuda Avniel

IPC分类号： H03F3/68 ,  H03F1/02 ,  H03F3/19 ,  H03F3/21 ,  H03F1/22 ,  H03F1/32 ,  H04L25/03 ,  H04L27/36

CPC分类号： H03F1/0205 ,  H03F1/0277 ,  H03F1/0294 ,  H03F1/223 ,  H03F1/3241 ,  H03F3/19 ,  H03F3/211 ,  H03F2200/451 ,  H03F2203/21112 ,  H04L25/03834 ,  H04L27/368

摘要： Digital compensators for use in outphasing-based power amplification systems (e.g., Linear Amplification using Nonlinear Components (LINC) amplifiers and Asymmetric Multilevel Outphasing (AMO) amplifiers) include a short memory nonlinear portion and a long memory linear time invariant (LTI) portion. In various embodiments, compensators are provided that are of relatively low complexity and that are capable of operation at throughputs exceeding a Gigasample per second.

摘要翻译： 用于基于偏移的功率放大系统（例如，使用非线性分量（LINC）放大器和非对称多电平外相（AMO）放大器的线性放大器）的数字补偿器包括短存储器非线性部分和长存储器线性时间不变量（LTI）部分。 在各种实施例中，提供了具有相对低的复杂性并且能够以超过每秒Gigasample的吞吐量进行操作的补偿器。