Self-attention-based confidence estimation of language models

    公开(公告)号:US12124814B2

    公开(公告)日:2024-10-22

    申请号:US17720912

    申请日:2022-04-14

    申请人: NAVER CORPORATION

    摘要: A confidence estimation system includes: a neural network including at least one an attention module including N heads configured to: generate attention matrices based on interactions between tokens for words in an input sequence of words, the input sequence of words including a word that is obscured; and determine the word that is obscured in the input sequence; and a confidence module configured to determine a confidence value indicative of a probability of the neural network correctly determining the word that is obscured, the confidence module determining the confidence value of the word that is obscured using a convolutional neural network that projects the attention matrices generated by the attention module over a multi-dimensional space, the attention matrices recording interactions between the tokens in the input sequence of words without information regarding the tokens for the words and the word that is obscured.

    Embedding Concealed Meta-Data Into Deep Neural Networks (DNNs)

    公开(公告)号:US20240330417A1

    公开(公告)日:2024-10-03

    申请号:US18738379

    申请日:2024-06-10

    申请人: Ciena Corporation

    IPC分类号: G06F21/16 G06N3/10

    CPC分类号: G06F21/16 G06N3/10

    摘要: Systems and methods for embedding concealed meta-data into DNNs. Specifically, the system and method presented consists of receiving a trained neural network that includes one or more layers each having weights. The disclosed process includes transforming at least one layer to a transformed domain, adding information to the layer(s) in the transformed domain, and performing an inverse domain transform on at least one layer such that the layer(s) has new weights with an embedded watermark.

    Programmable multi-level data access address generator

    公开(公告)号:US12105625B2

    公开(公告)日:2024-10-01

    申请号:US17588240

    申请日:2022-01-29

    申请人: Ceremorphic, Inc.

    IPC分类号: G06F12/04 G06N3/0464 G06N3/10

    摘要: A programmable address generator has an iteration variable generator for generation of an ordered set of iteration variables, which are re-ordered by an iteration variable selection fabric, which delivers the re-ordered iteration variables to one or more address generators. A configurator receives an instruction containing fields which provide configuration constants to the address generator, iteration variable selection fabric, and address generators. After configuration, the address generators provide addresses coupled to a memory. In one example of the invention, the address generators generate an input address, a coefficient address, and an output address for performing convolutional neural network inferences.

    MUTABLE PARAMETERS FOR MACHINE LEARNING MODELS DURING RUNTIME

    公开(公告)号:US20240296346A1

    公开(公告)日:2024-09-05

    申请号:US18383858

    申请日:2023-10-25

    申请人: Apple Inc.

    IPC分类号: G06N3/10 G06F8/41 G06F17/16

    CPC分类号: G06N3/10 G06F8/41 G06F17/16

    摘要: The subject technology receives code corresponding to a neural network (NN) model and a set of weights for the NN model. The subject technology determines a set of layers that are mutable in the NN model. The subject technology determines information for mapping a second set of weights to the set of weights for the NN model. The subject technology generates metadata corresponding to the set of layers that are mutable, and the information for mapping the second set of weights to the set of weights for the NN model, wherein the generated metadata enables updating the set of layers that are mutable during execution of the NN model.

    Optimizing neural network structures for embedded systems

    公开(公告)号:US12079723B2

    公开(公告)日:2024-09-03

    申请号:US18183515

    申请日:2023-03-14

    申请人: Tesla, Inc.

    摘要: A model training and implementation pipeline trains models for individual embedded systems. The pipeline iterates through multiple models and estimates the performance of the models. During a model generation stage, the pipeline translates the description of the model together with the model parameters into an intermediate representation in a language that is compatible with a virtual machine. The intermediate representation is agnostic or independent to the configuration of the target platform. During a model performance estimation stage, the pipeline evaluates the performance of the models without training the models. Based on the analysis of the performance of the untrained models, a subset of models is selected. The selected models are then trained and the performance of the trained models are analyzed. Based on the analysis of the performance of the trained models, a single model is selected for deployment to the target platform.

    Programmable multi-level data access address generator

    公开(公告)号:US12072799B2

    公开(公告)日:2024-08-27

    申请号:US18121294

    申请日:2023-03-14

    申请人: CEREMORPHIC, INC.

    IPC分类号: G06F12/04 G06N3/0464 G06N3/10

    摘要: A programmable address generator has an iteration variable generator for generation of an ordered set of iteration variables, which are re-ordered by an iteration variable selection fabric, which delivers the re-ordered iteration variables to one or more address generators. A configurator receives an instruction containing fields which provide configuration constants to the address generator, iteration variable selection fabric, and address generators. After configuration, the address generators provide addresses coupled to a memory. In one example of the invention, the address generators generate an input address, a coefficient address, and an output address for performing convolutional neural network inferences.

    Embedding concealed meta-data into deep neural networks (DNNs)

    公开(公告)号:US12056220B2

    公开(公告)日:2024-08-06

    申请号:US17893648

    申请日:2022-08-23

    申请人: Ciena Corporation

    IPC分类号: G06F21/16 G06N3/10

    CPC分类号: G06F21/16 G06N3/10

    摘要: The present disclosure relates to systems and methods for embedding concealed meta-data into DNNs. Specifically, the system and method presented consists of receiving a trained neural network that includes one or more layers each having weights. The disclosed process includes transforming at least one layer to a transformed domain, adding information to the layer(s) in the transformed domain, and performing an inverse domain transform on at least one layer such that the layer(s) has new weights with an embedded watermark.