公开(公告)号：US12045961B2

公开(公告)日：2024-07-23

申请号：US18489876

申请日：2023-10-19

Applicant: ZHEJIANG LAB

Inventor： Jingsong Li ,  Jinnan Hu ,  Peijun Hu ,  Yu Tian ,  Tianshu Zhou

IPC: G06T5/70 ,  G06T5/10

CPC classification number: G06T5/70 ,  G06T5/10 ,  G06T2207/20064 ,  G06T2207/20081 ,  G06T2207/20084

Abstract: Disclosed is an image denoising method and apparatus based on wavelet high-frequency channel synthesis. Image data are expanded to a plurality of frequency-domain channels, a plurality of “less-noise” channels and a plurality of “more-noise” channels are grouped through a noise-sort algorithm, and a denoising submodule and a synthesis submodule based on style transfer are combined to form a generative network. A discriminative network is established to add a constraint to the global loss function. After iteratively training the GAN model described above, the denoised image data can be obtained through wavelet inverse transformation. The disclosed algorithm can effectively solve the problem of “blurring” and “loss of details” introduced by traditional filtering or CNN-based deep learning methods, which is especially suitable for noise-overwhelmed image data or high dimensional image data.

公开(公告)号：US12019571B1

公开(公告)日：2024-06-25

申请号：US18389783

申请日：2023-12-20

Applicant: ZHEJIANG LAB

Inventor： Li Yan ,  Songnan Ren ,  Zhiwei Liu ,  Tang Hu ,  Xiangdi Li ,  Jiani Gu ,  Chunling Hao ,  Xiao Yu

IPC: G06F13/20

CPC classification number: G06F13/20 ,  G06F2213/40

Abstract: A communication method for a multi-chip neural network algorithm based on a FPGA main control, which designs original data frames, status frames, layered data frames, layered weight frames, computation result frames, layered data request frames, layered weight request frames, computation result request frames and running status request frames, and then completes image processing based on the neural network algorithm according to the scheduling of transmitting and receiving processes. The present disclosure ensure that communication of multi-layer data structures and various data types based on the neural network algorithm, and accurately schedules the transmitting and receiving of data required by the main control and each chip in the multi-chip system, and sends out data request commands; it plays a very active role in receiving, transmitting and feeding back the running status of the chip and the errors and error types.

公开(公告)号：US12015548B2

公开(公告)日：2024-06-18

申请号：US18542823

申请日：2023-12-18

Applicant: ZHEJIANG LAB

Inventor： Congqi Shen ,  Huifeng Zhang ,  Tao Zou ,  Ruyun Zhang

IPC: H04L45/00 ,  H04L45/74 ,  H04L69/22

CPC classification number: H04L45/38 ,  H04L45/74 ,  H04L69/22

Abstract: A method and a device for identification management and optimized forwarding in a large-scale polymorphic network, the method comprising the follow steps: S1, constructing a polymorphic backbone network; S2, modality identification management; S3, determining a modality to be forwarded; S4, configuring a flow table for a switching node; S5, receiving a packet by a balanced distributor, and preliminarily parsing the type of the packet; S6, parsing key field information in the packet, determining the switching nodes to be allocated according to the key field information, and transmitting the key field information to the corresponding switching node; S7, the switching node matching the stored flow table according to the key field information to determine a correct forwarding action.

公开(公告)号：US20240161251A1

公开(公告)日：2024-05-16

申请号：US18489876

申请日：2023-10-19

Applicant: ZHEJIANG LAB

Inventor： Jingsong LI ,  Jinnan HU ,  Peijun HU ,  Yu TIAN ,  Tianshu ZHOU

IPC: G06T5/00 ,  G06T5/10

CPC classification number: G06T5/002 ,  G06T5/10 ,  G06T2207/20064 ,  G06T2207/20081 ,  G06T2207/20084

Abstract: Disclosed is an image denoising method and apparatus based on wavelet high-frequency channel synthesis. Image data are expanded to a plurality of frequency-domain channels, a plurality of “less-noise” channels and a plurality of “more-noise” channels are grouped through a noise-sort algorithm, and a denoising submodule and a synthesis submodule based on style transfer are combined to form a generative network. A discriminative network is established to add a constraint to the global loss function. After iteratively training the GAN model described above, the denoised image data can be obtained through wavelet inverse transformation. The disclosed algorithm can effectively solve the problem of “blurring” and “loss of details” introduced by traditional filtering or CNN-based deep learning methods, which is especially suitable for noise-overwhelmed image data or high dimensional image data.

公开(公告)号：US11983481B2

公开(公告)日：2024-05-14

申请号：US18351464

申请日：2023-07-12

Applicant: ZHEJIANG LAB

Inventor： Zhiquan Wan ,  Shunbin Li ,  Ruyun Zhang ,  Weihao Wang ,  Qingwen Deng

IPC: G06F30/398 ,  G06F30/392 ,  G06F117/12

CPC classification number: G06F30/398 ,  G06F30/392 ,  G06F2117/12

Abstract: The present disclosure relates to software-defined methods and apparatuses for designing a wafer-level switching system, including: determining wafer-level switching system layout constraints; constructing a target wafer-level switching system and determining parameters, and designing a logical topology of a switching network; designing a layout of the switching chiplets on the wafer substrate; respectively designing interface structures of external chiplets and internal chiplets; configuring a switching mode and an enable state of each port of the switching chiplets; ending the process when the target logical topology can be achieved by the wafer-level switching system; otherwise, reconstructing a logical topology of a switching network and mapping it to the substrate.

公开(公告)号：US11977865B2

公开(公告)日：2024-05-07

申请号：US17979752

申请日：2022-11-02

Applicant: ZHEJIANG LAB

Inventor： Lei Xue ,  Tao Zou ,  Ruyun Zhang

IPC: G06F8/30 ,  G06F8/41 ,  G06F9/48

CPC classification number: G06F8/37 ,  G06F8/447 ,  G06F9/4881

Abstract: A software and hardware collaborative compilation processing method and system. The system comprises an environment configurator, a command parser, a code filler, a scheduler and a heterogeneous target system, wherein the code filler is configured for obtaining the source program path of a user, reading source codes and identifying the heterogeneous target system according to a macro definition, complementing the codes related to the heterogeneous target system, carrying out primary filling and secondary filling on the source codes; the scheduler is configured for realizing compilation scheduling and execution scheduling functions respectively; the heterogeneous target system is configured for compiling and processing user modal data, and comprises at least two heterogeneous target subsystems; each target subsystem comprises a target-related middle-end compiler, a back-end compiler and a target-related running environment.

公开(公告)号：US20240111586A1

公开(公告)日：2024-04-04

申请号：US18472648

申请日：2023-09-22

Applicant: ZHEJIANG LAB

Inventor： Shiqiang ZHU ,  Aimin PAN ,  Feng GAO

IPC: G06F9/50

CPC classification number: G06F9/5027

Abstract: The present disclosure belongs to the field of intelligent computing technologies, and relates to a multi-policy intelligent scheduling methods and apparatuses oriented to heterogeneous computing power. The method includes: step 1, setting an execution policy of a task based on heterogeneity of computing clusters, differences of computing tasks and a user requirement, and constructing a Markov decision process model by adopting a reinforcement learning method combined with the execution policy; step 2, adopting a proximal policy optimization to solve an optimal task scheduling policy of the task input by the user based on the constructed Markov decision process model; step 3, scheduling the task to a corresponding computing cluster for execution based on the optimal task scheduling policy.

公开(公告)号：US20240104016A1

公开(公告)日：2024-03-28

申请号：US18071958

申请日：2022-11-30

Applicant: ZHEJIANG LAB

Inventor： Hongsheng WANG ,  Aimin PAN ,  Guang CHEN

IPC: G06F12/0802 ,  G06N3/063

CPC classification number: G06F12/0802 ,  G06N3/063

Abstract: The disclosure discloses an intermediate representation method for compiling computation graphs, including: step 1: compiling a neural network into a computation graph for neural network computation; step 2: constructing a node for each tensor variable in the computation graph; step 3: associating the node representing the tensor variable in the computation graph to a set of pointers to the tensor variable; step 4: analyzing constraint relationships between the tensor variables in the computation graph; step 5: iteratively constructing a topological graph of the intermediate representation based on the constraint relationships between the tensor variables in the computation graph; and step 6: analyzing the tensor variables with different aliases pointing to a same memory location based on the intermediate representation, and allocating a register for the tensor variables with different aliases. The method optimizes the compilation efficiency of the tensor variables pointing to the same memory location in the computation graph.

公开(公告)号：US20240073096A1

公开(公告)日：2024-02-29

申请号：US17767531

申请日：2021-11-10

Applicant: ZHEJIANG LAB

Inventor： Peilei WANG ,  Wenjiao YANG ,  Wei LV ,  Xingming ZHANG

IPC: H04L41/12 ,  H04L45/02

CPC classification number: H04L41/12 ,  H04L45/02

Abstract: The present disclosure discloses a design method for lightweight wired Mesh networking. The design method includes: designing a protocol format of a lightweight wired Mesh network message; designing an input parsing module of the lightweight wired Mesh network message; designing a heartbeat logic module between devices of the lightweight wired Mesh networking; designing a route management module of a lightweight wired Mesh network; designing a forwarding and processing module of the lightweight wired Mesh network message; and designing an underlying hardware interface management module.

公开(公告)号：US20240062006A1

公开(公告)日：2024-02-22

申请号：US18297610

申请日：2023-04-08

Applicant: ZHEJIANG LAB

Inventor： Taihao LI ,  Jiantao HUANG

IPC: G06F40/284 ,  G06F40/30 ,  G06F40/268

CPC classification number: G06F40/284 ,  G06F40/30 ,  G06F40/268

Abstract: An open domain dialog reply method and a system based on thematic enhancement are provided. The method includes: collecting and pre-processing text corpuses to obtain Chinese dialog corpus dataset, performing sentence breaking, word separation, and lexical annotation of dialogs and extracting noun words, performing enhancement of semantic and thematic information on each sentence, and learning vector representations of original sentences and enhanced sentences by a pre-trained sentence representation model, performing thematic aggregation enhancement by a graph convolutional neural network, and inputting the sentence vector after the thematic aggregation enhancement into a pre-trained generative model, generating a candidate set of dialog replies, and training a reply ranking selection model with a contrast learning manner to select the most suitable reply.