公开(公告)号：US10211733B1

公开(公告)日：2019-02-19

申请号：US15796488

申请日：2017-10-27

Applicant: Google LLC

Inventor： Shuai Jiang ,  Xin Li

IPC: H02M3/158 ,  H02M1/15

Abstract: An apparatus that includes a resonant DC-DC converter with switching frequencies based on stray inductances of the physical components used to construct the apparatus. This results in a relatively high efficiency and high density DC-DC converter that, in some implementations, does not require a discrete inductor component.

公开(公告)号：US20240404238A1

公开(公告)日：2024-12-05

申请号：US18698997

申请日：2022-10-05

Applicant: Google LLC

Inventor： Jiahui Yu ,  Vijay Vasudevan ,  Alexander Yeong-Shiuh Ku ,  Yonghui Wu ,  Jason Michael Baldridge ,  Yuanzhong Xu ,  Jing Yu Koh ,  Thang Minh Luong ,  Gunjan Baid ,  Zirui Wang ,  Han Zhang ,  Xin Li

IPC: G06V10/28 ,  G06F40/284 ,  G06V10/764 ,  G06V10/766 ,  G06V10/82

Abstract: Systems and methods are provided for vector-quantized image modeling using vision transformers and improved codebook handling. In particular, the present disclosure provides a Vector-quantized Image Modeling (VIM) approach that involves pre-training a machine learning model (e.g., Transformer model) to predict rasterized image tokens autoregressively. The discrete image tokens can be encoded from a learned Vision-Transformer-based VQGAN (example implementations of which can be referred to as ViT-VQGAN). The present disclosure proposes multiple improvements over vanilla VQGAN from architecture to codebook learning, yielding better efficiency and reconstruction fidelity. The improved ViT-VQGAN further improves vector-quantized image modeling tasks, including unconditional image generation, conditioned image generation (e.g., class-conditioned image generation), and unsupervised representation learning.

公开(公告)号：US20240112088A1

公开(公告)日：2024-04-04

申请号：US18520083

申请日：2023-11-27

Applicant: Google LLC

Inventor： Jiahui Yu ,  Xin Li ,  Han Zhang ,  Vijay Vasudevan ,  Alexander Yeong-Shiuh Ku ,  Jason Michael Baldridge ,  Yuanzhong Xu ,  Jing Yu Koh ,  Thang Minh Luong ,  Gunjan Baid ,  Zirui Wang ,  Yonghui Wu

IPC: G06N20/00

CPC classification number: G06N20/00

Abstract: Systems and methods are provided for vector-quantized image modeling using vision transformers and improved codebook handling. In particular, the present disclosure provides a Vector-quantized Image Modeling (VIM) approach that involves pretraining a machine learning model (e.g., Transformer model) to predict rasterized image tokens autoregressively. The discrete image tokens can be encoded from a learned Vision-Transformer-based VQGAN (example implementations of which can be referred to as ViT-VQGAN). The present disclosure proposes multiple improvements over vanilla VQGAN from architecture to codebook learning, yielding better efficiency and reconstruction fidelity. The improved ViT-VQGAN further improves vector-quantized image modeling tasks, including unconditional image generation, conditioned image generation (e.g., class-conditioned image generation), and unsupervised representation learning.

公开(公告)号：US20240094799A1

公开(公告)日：2024-03-21

申请号：US17894600

申请日：2022-08-24

Applicant: Google LLC

Inventor： Xiong Li ,  Xin Li ,  Qiong Wang ,  Kaushik Vaidyanathan ,  Chenhao Nan ,  Robert Ashby Armistead, III

IPC: G06F1/3296

CPC classification number: G06F1/3296

Abstract: The technology is generally directed to a coordinated power throttling mechanism for a payload using power provided by a rack such that the rack power does not exceed a threshold amount for greater than a predetermined period of time. The coordinated power throttling mechanism includes the rack providing a power throttling signal to the payload and the payload executing the power throttling upon detection of the throttling signal. The payload may detect the throttling signal and, after a delay, execute the power throttling. The delay may ensure that all payloads within the rack have detected the power throttling signal.

公开(公告)号：US20240070193A1

公开(公告)日：2024-02-29

申请号：US17902485

申请日：2022-09-02

Applicant: GOOGLE LLC

Inventor： Pengfei Wu ,  Xin Li

IPC: G06F16/683 ,  G10L15/22

CPC classification number: G06F16/683 ,  G10L15/22 ,  G10L2015/223

Abstract: Implementations relate to maintaining metadata, of a client device, stored on a remote computing device, to reduce overall network utilization and/or latency in responding to automated assistant requests received at the client device. The metadata of a client device is identified and provided to a remote computing device with a request that requires additional processing by the remote computing device. Current metadata is compared to previously provided metadata that was provided to the remote computing device with a previous request and one or more operations are generated based on the difference between the current metadata and the previous metadata. The remote computing device executes the one or more operations on a remote version of the metadata to generate a new remote version of the metadata of the client device. The remote computing device further processes the request with the new remote version of the metadata.

公开(公告)号：US20220156340A1

公开(公告)日：2022-05-19

申请号：US17098016

申请日：2020-11-13

Applicant: Google LLC

Inventor： Xin Li ,  Yixin Wang ,  Benedict Liang ,  Dharminder Singh

IPC: G06F16/957 ,  G06F16/9032 ,  G06F16/903 ,  G06F16/9035

Abstract: Techniques of this disclosure are directed to enable a computing device to process voice queries and provide query answers even when the computing device and vehicle do not have internet connectivity. According to the disclosed techniques, a computing device may detect a query via input devices of the computing device and output a query answer determined based on the detected query. Rather than directly querying a remote computing system, various aspects of the techniques of this disclosure may enable the computing device to use a query answer cache to generate the query answer. The query answer cache may include predicted queries and query answers retrieved from a query answer cache of a remote computing system, thereby enabling the computing device to respond to the detected queries while experiencing unreliable internet connection.

公开(公告)号：US10958098B1

公开(公告)日：2021-03-23

申请号：US16594704

申请日：2019-10-07

Applicant: Google LLC

Inventor： Xiong Li ,  Xin Li ,  Philippe H. Gow ,  Qiong Wang

IPC: H02J9/06 ,  H02M3/335 ,  H02J3/38 ,  H02J7/00 ,  H02J1/10 ,  H02J3/00

Abstract: Methods, systems, and apparatus, for powering AC and DC loads in the event of a power failure. In one aspect, a system includes one or more DC loads that are powered by DC power, one or more AC loads that are powered by AC power, and an uninterruptible power supply (UPS). The UPS includes AC terminals connected to an AC power source and to the one or more AC loads to power the AC loads, DC terminals connected to a backup battery and to the one or more DC loads to power the DC loads a bidirectional AC/DC converter connected between the AC terminals and the DC terminals, and a controller that selectively switches the bidirectional AC/DC converter to the first mode when the AC power source is available and to the second mode when the AC power source is not available.

公开(公告)号：US20190197077A1

公开(公告)日：2019-06-27

申请号：US15318229

申请日：2016-10-17

Applicant: GOOGLE LLC

Inventor： Xin Li ,  Fang Yang

IPC: G06F16/955 ,  G06N20/00 ,  H04L29/08 ,  H04L12/24

CPC classification number: G06F16/9566 ,  G06F16/9558 ,  G06N20/00 ,  H04L41/16 ,  H04L67/02 ,  H04L67/22 ,  H04L67/36 ,  H04L69/329

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for identifying broken network connections. In one aspect, a system includes front-end server(s) that receive data specifying, for multiple different user interactions with one or more application links that link to a given application, presentation durations specifying how long application content linked to by the application link was presented following the multiple different user interactions with the application link(s). Back-end server(s) that communicate with the front end server(s) can classify each application link as broken or working based on application of a machine learning model to the presentation durations for the application link. The machine learning model can be generated using labeled training data. The back-end server(s) can generate and output an alert identifying an application link as a broken link based on the application link being classified as broken by the machine learning model.

公开(公告)号：US10312813B2

公开(公告)日：2019-06-04

申请号：US16170452

申请日：2018-10-25

Applicant: Google LLC

Inventor： Shuai Jiang ,  Chee Yee Chung ,  Xin Li

IPC: H02M3/158

Abstract: An apparatus that includes first and second parallel converter branches, each parallel converter branch including an input node, N output nodes, a plurality of switches, a converter output node, and control logic. The control logic generates a first set of switch signals to control the switches of the first parallel converter branch and a second set of switch signals to control the second parallel converter branch, the first set switch signals and the second set of switch signals having respective duty cycles to cause each of the first and second parallel converter branches to output the DC output voltage on each of the N output nodes.

公开(公告)号：US12159738B2

公开(公告)日：2024-12-03

申请号：US17358354

申请日：2021-06-25

Applicant: Google LLC

Inventor： Chenhao Nan ,  Houle Gan ,  Runruo Chen ,  Qiong Wang ,  Xin Li

IPC: H01F27/28 ,  H01F27/06 ,  H01F27/24

Abstract: The subject matter described herein provides systems and techniques for the integration of TLVR technology in a vertical power VR module. A multiple-secondary TLVR topology using a controlled leakage inductance in the place of a separate compensation inductor, Lc, may be employed for the vertical power VR module. In addition, the capacitance inside the device to which the TLVR based vertical power VR module supplies power, rather than an output capacitance board, may be used in order to allow the module to be a single layer. Example structures that may include one or more primary windings and/or one or more secondary windings for each of possibly multiple linked phases of the TLVR based module are provided. The windings may be formed using traditional copper windings or printed circuit board (PCB) copper trace winding.