CN109711535B - Method for predicting layer calculation time in deep learning model by using similar layer - Google Patents
Method for predicting layer calculation time in deep learning model by using similar layer Download PDFInfo
- Publication number
- CN109711535B CN109711535B CN201811574365.XA CN201811574365A CN109711535B CN 109711535 B CN109711535 B CN 109711535B CN 201811574365 A CN201811574365 A CN 201811574365A CN 109711535 B CN109711535 B CN 109711535B
- Authority
- CN
- China
- Prior art keywords
- layer
- calculation time
- similar
- estimated
- time
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Abstract
The invention provides a method for predicting layer calculation time in a deep learning model by using a similar layer. The method comprises the steps of evaluating similar layers through key attributes and non-key attributes of the layers, evaluating the similarity degree between the similar layers through the non-key attribute partial order relation of the layers, and further utilizing the calculation time of the similar layers in a historical log to estimate the calculation time of the layer to be predicted; for similar layers that cannot be evaluated (i.e., have no evaluation available in the history log) or are not acceptable, the layer is measured. The invention avoids repeated measurement work and reduces the system overhead by multiplexing the known layer calculation time. In addition, the invention also realizes the localization of prediction, and the prediction of the layer computation time of the model can be completed on one local device.
Description
Technical Field
The invention relates to the technical field of deep learning; and more particularly, to a method for predicting layer computation time in a deep learning model using a similar layer.
Background
Deep learning is currently widely used in industries such as financial insurance, security monitoring, and the like. The core approach to deep learning is to analyze and characterize the data using neural network models. Training a deep-learning model requires multiple iterations at a frequency on the order of seconds. Secondly, compared with a linear algorithm, deep learning often requires a more huge data training party to obtain an accurate model. Training a neural network model often takes days or weeks to complete.
For the above reasons, it obviously takes longer time to train the deep learning model by only a single node; in contrast, the current mainstream deep learning framework supports the realization of a distributed training model through parallel computing. Distributed training in the broad sense includes two types: one is data parallel and the other is model parallel. Under the scene of data parallel, by predicting the training time of the deep learning model, the method has important significance for resource allocation, model evaluation and the like in the deep learning model training process of many days, weeks and months. Especially, under the condition that a plurality of different model training tasks coexist, the training time is predicted in advance, resources are reasonably allocated, and the resource utilization rate is greatly improved. In contrast, through the measurement history logs of a small number of layers and the combination of the similar layer relation and the time of multiplexing the layers, the expenditure of the iterative time in the training of the prediction model is greatly reduced.
In addition, the calculation time of each layer in the deep learning model is measured/predicted, and the method is also significant for the targeted adjustment of the model in the layer granularity.
Disclosure of Invention
In view of the above, the present invention provides a method for predicting layer computation time in a deep learning model by using a similar layer. The method comprises the steps of evaluating similar layers through key attributes and non-key attributes of the layers, evaluating the similarity degree between the similar layers through the non-key attribute partial order relation of the layers, and further utilizing the calculation time of the similar layers in a historical log to estimate the calculation time of the layer to be predicted; for similar layers that cannot be evaluated (i.e., have no evaluation for similar layers in the history log) or are not acceptable, the layer computation time is measured.
In one aspect, the present invention provides a method for evaluating similar layers in a deep learning model using layer attributes.
The method for evaluating the similar layer in the deep learning model by using the layer attribute comprises the following steps:
acquiring layer attributes of each layer in the model, wherein the layer attributes comprise layer input and output scales;
for any two or more layers, if the input and output scales of the two or more layers are the same, the two or more layers are the same type of layer;
for any two or more layers of the same type, if all corresponding key attributes of the two or more layers are the same, the two or more layers are similar;
the key attribute mentioned above refers to a layer attribute having a non-linear influence on the computation time among all layer attributes. Correspondingly, the layer attribute having a linear influence on the computation time among all the layer attributes is a non-critical attribute.
Alternatively, in the methods described above, for any two or more similar layers, they are the same layer if all of their corresponding non-critical properties are also the same, i.e., all of their corresponding properties are the same.
In another aspect, the present invention provides a method for evaluating the degree of inter-level similarity in a deep learning model.
With reference to the first aspect, the method for evaluating the degree of similarity between layers in the deep learning model described above determines the degree of similarity between any similar layer set evaluated by the method described in the first aspect according to the partial order relationship between all or part of the layers in the set:
for any two or more layers in the above-mentioned similar layer set, if they all have the following partial order relationship:
for any non-critical attribute of the forward influence layer calculation time, if the non-critical attribute monotonically increases, the calculation time monotonically increases,
calculating the non-key attribute of the time of any reverse influence layer, and if the non-key attribute monotonically decreases, monotonically increasing the calculation time of the non-key attribute;
the similarity of the layers is evaluated according to the partial order relationship. Generally, the more adjacent layers, the higher their degree of similarity.
In yet another aspect, the present invention provides a method for estimating layer computation time in a model using computation time of similar layers.
With reference to the first and second aspects, the method for estimating layer computation time in a layer computation time estimation model using similarity layers includes:
for any of the layers to be evaluated,
estimating the calculation time of the layer according to the calculation time of the similar layer evaluated by the method mentioned in the first aspect; wherein, the time of the similarity layer is calculated and obtained by inquiring historical logs;
if there is no similar layer computation time in the history log (i.e. there is no similar layer in the history log for estimating the computation time of the layer), the layer is not estimated.
Optionally, the above-mentioned similar layer computation times are obtained from measurements rather than estimates to avoid errors in amplification (and potentially cascaded amplification) estimates in this way.
Optionally, the aforementioned estimating the computation time of the layer to be estimated based on the computation time of the similar layer includes:
if the same layer calculation time exists, multiplexing the same layer calculation time as the calculation time of the layer to be estimated;
otherwise, the similarity degree between the similar layer in the history log and the layer to be estimated is obtained by evaluation according to the second aspect and the method, a proper similar layer is selected according to the similarity degree between the layers, and the calculation time of the layer to be estimated is estimated according to the calculation time. For example, the similar layer with the closest similarity is selected, and the calculation time of the layer to be estimated is estimated based on the calculation time (one of them is simply estimation, i.e. the calculation time of the similar layer is directly used as the calculation time of the layer to be estimated).
However, in the method provided in the above embodiment, even if the closest similarity layer estimation calculation time is selected, only the most suitable layer is selected for estimation in terms of the degree of similarity between layers, and in terms of the calculation time obtained by estimation, an effective metric cannot be provided to evaluate the error caused by estimation.
Therefore, further, optionally, the aforementioned selecting a suitable similar layer computation time to estimate the layer computation time to be estimated includes:
according to a second aspect and the partial ordering relationship in the method, from similar levels in the history log,
selecting similar layers of which the calculation time is greater than that of the layer to be estimated in the partial order relation, sorting the calculation times of the similar layers (namely the calculation times in the historical logs) according to the size, and selecting the minimum value as the maximum calculation time of the layer to be estimated;
selecting similar layers of which the calculation time is less than that of the layer to be estimated in the partial order relation, sorting the calculation times of the similar layers according to the size, and selecting the maximum value as the minimum calculation time of the layer to be estimated;
estimating the calculation time of the layer to be estimated according to the maximum calculation time and the minimum calculation time; for example, the maximum computation time and the minimum computation time of the layer to be estimated are averaged to obtain the computation time of the layer to be estimated.
In yet another aspect, the present invention provides a method of evaluating whether an estimate of layer computation time is acceptable.
With reference to the second and third aspects, for the estimation error problem of the method of the third aspect, the method for evaluating whether the estimation of the layer calculation time is acceptable includes:
for any of the layers to be evaluated as described in the third aspect,
evaluating whether the estimation is acceptable according to the maximum calculation time and the minimum calculation time; wherein the content of the first and second substances,
if the difference between the maximum calculation time and the minimum calculation time is less than a set threshold value, the estimation is acceptable;
otherwise, the estimate is not acceptable;
wherein, based on the partial order relationship mentioned in the second aspect,
the maximum computation time is assigned according to the similar layer of which the computation time is greater than the layer to be estimated in the partial order relation in the historical log; the minimum computation time is assigned according to the similar layer of which the computation time is smaller than the layer to be estimated on the partial order relation in the history log. For example, two layers adjacent to the layer to be estimated in the history log in the upstream and downstream in the partial order relation are selected and correspondingly used as the maximum calculation time and the minimum calculation time.
In addition, the threshold value may be set to be a minimum calculation time multiplied by a coefficient γ; the gamma value will define the error of the estimate for each layer.
If the layer to be estimated in the history log has a similar layer only on one side of the partial order relationship, that is, only the maximum computation time/the minimum computation time can be assigned, the evaluation cannot be performed by the above method, and the evaluation is considered to be unacceptable.
Optionally, a method for assigning and evaluating maximum and minimum computation times includes:
respectively initializing the maximum calculation time and the minimum calculation time of the layer to be estimated to be 0 and infinity;
according to a second aspect and the partial ordering relationship in the method, from similar levels in the history log,
selecting similar layers of which the calculation time is greater than that of the layer to be estimated in the partial order relation, sorting the calculation times of the similar layers (namely the calculation times in the historical logs) according to the size, and selecting the minimum value as the maximum calculation time of the layer to be estimated;
selecting similar layers of which the calculation time is less than that of the layer to be estimated in the partial order relation, sorting the calculation times of the similar layers according to the size, and selecting the maximum value as the minimum calculation time of the layer to be estimated;
and further, evaluating whether the estimation is acceptable according to the maximum calculation time and the minimum calculation time:
if the difference between the maximum calculation time and the minimum calculation time is less than a set threshold value, the estimation is acceptable;
otherwise, the estimate is not acceptable.
In yet another aspect, the present invention provides a method for predicting layer computation time in a deep learning model using a similar layer.
With reference to the third and fourth aspects, the method for predicting layer computation time in a deep learning model by using a similar layer includes:
for any of the layers to be predicted,
evaluating whether the evaluation for the layer to be predicted is acceptable in the method according to the fourth aspect,
if yes, estimating the computation time of the layer to be predicted by the method described in the third aspect;
and if not, or the layer to be predicted cannot be estimated, measuring the calculation time of the layer.
The technical scheme provided by the invention has a plurality of beneficial effects:
one, low overhead; the invention reuses the known layer calculation time as much as possible and avoids repeated measurement of the same layer or similar layers.
Secondly, forecasting localization; even for the model trained on the distributed machine learning platform or the model trained by utilizing the online resources, the method can be completely finished on one device when the calculation time is predicted.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings related to a part of the embodiments of the present invention will be briefly described below.
Fig. 1 is a flowchart illustrating a method for predicting layer computation time in a deep learning model by using a similar layer according to an embodiment of the present invention.
Detailed Description
The technical solution in the embodiments of the present invention is clearly and completely described below with reference to the drawings of the embodiments of the present invention. It is to be understood that the described embodiments are merely exemplary of a portion of the invention and not all embodiments. All other embodiments, which can be obtained by a person skilled in the art without any inventive step based on the embodiments of the present invention, are within the scope of the present invention.
The following is a preferred embodiment of the present invention.
Fig. 1 is a flowchart illustrating a method for predicting layer computation time in a deep learning model by using a similar layer according to this embodiment. The above embodiments show the prediction of layer computation time for a model implemented under the tenserflow framework.
As shown in figure 1 of the drawings, in which,
1) before prediction, obtaining a layer in the model through data flow graph analysis;
the data flow diagram description file describes each layer as a JSON entry, and the JSON entry comprises a layer name and a layer attribute;
the following is an example of a description file for one of the model dataflow graphs,
[
{
"layer_name":"conv1","layer_id":1,"tf_layer_name":"conv2d","params":{
"input_size":112,"kernel_size":3,"ch_in":3,"ch_out":64,"batch_size":32,"stride":1}
},
{
"layer_name":"fc1","layer_id":2,"input_id":1,"tf_layer_name":"dense","params":{"num_units":1000}
},
...
]
2) after the layers of the model are obtained, inquiring whether similar layers exist in a historical log or not for each layer to be predicted;
wherein, the definition of the similar layer and the related concept is as follows:
definition of A symbol
Layer input scale: sI(ii) a Layer output scale: sO(ii) a Layer properties: p; layer attribute set: pL(ii) a Key attributes are as follows: p is a radical ofK(ii) a Key attribute set: pL K(ii) a Non-critical property set PL；-PL K(ii) a Wherein the layer is input with a scale SILayer output scale SOEach one of the above-mentioned layer properties p.
In this embodiment, the influence of kernel _ size and the like on the calculation time is non-linear and is a key attribute, and the influence of batch _ size, channels and the like on the calculation time is linear and is a non-key attribute;
b same type layer
For any two or more layers, if their layer is input into the scale SILayer output scale SOAll are the same, then they are considered to be the same type of layer;
c similar layer
For any two or more layers of the same type, if any of their corresponding key attributes P (P ∈ P)L K) All are the same, they are considered to be similar layers;
d same layer
For any two or more similar layers, if any of them corresponds to a non-critical property P (P ∈ P)L-PL K) And are all the same, they are considered to be the same layer;
e, preliminarily evaluating the partial order relation of similarity degree of each layer in the calculation time
For any similar layer set, if any two or more layers exist, the following partial order relationship exists:
for any non-critical attribute of the forward influence layer calculation time, if the non-critical attribute monotonically increases, the calculation time monotonically increases,
calculating the non-key attribute of the time of any reverse influence layer, and if the non-key attribute monotonically decreases, monotonically increasing the calculation time of the non-key attribute;
namely, the partial order relationship of the similarity degree of each layer in the calculation time can be preliminarily evaluated.
In this embodiment, if similar layers liAnd a layer ljThe above partial order relationship exists in all the non-critical attributes (for example, the non-critical attributes of channel and batch size, both of which positively affect the layer computation time, i.e., the computation time increases as the attribute value increases, channel (l)i)＜channels(lj)，batch size(li)＜batch size(lj) And other non-key attributes also have a partial ordering relationship with them), the two are considered to have a partial ordering relationship in the calculation time, li＜lj。
3) If the similar layer of the layer to be predicted exists, estimating the calculation time of the layer to be predicted by using the similar layer and evaluating whether the estimation is acceptable:
firstly, further judging whether the same layer of the layer to be predicted exists in the history log,
if so, multiplexing the same layer of calculation time as the calculation time;
if it is not present, the first layer of the film,
then the calculation is carried out again, and the calculation time of the layer to be predicted is estimated by using the calculation time of the similar layer;
before estimation, whether the estimation is acceptable is evaluated according to the maximum and minimum calculation time of the layer to be predicted:
respectively initializing the maximum calculation time and the minimum calculation time of the layer to be estimated to be 0 and infinity;
according to the above-mentioned partial order relationship, from the similar layer,
selecting similar layers of which the calculation time is greater than that of the layer to be estimated in the partial order relation, sorting the calculation times of the similar layers according to the magnitude, and selecting the minimum value as the maximum calculation time of the layer to be estimated;
selecting similar layers of which the calculation time is less than that of the layer to be estimated in the partial order relation, sorting the calculation times of the similar layers according to the size, and selecting the maximum value as the minimum calculation time of the layer to be estimated;
if the difference between the maximum calculation time and the minimum calculation time is less than the threshold value Ts(Ts＝γtminWherein t isminThe minimum computation time of the layer to be predicted, and the value of gamma in the embodiment is 20%), the estimation is acceptable;
and averaging the maximum calculation time and the minimum calculation time to obtain the calculation time of the layer to be predicted.
Otherwise, the estimate is not acceptable (or cannot be estimated);
the calculated time for that layer is measured.
The above description is only a specific embodiment of the present invention, but the scope of the present invention is not limited thereto.
Claims (6)
1. A method for estimating layer computation time in a model using computation time of a similarity layer, the method comprising:
the model is a neural network model;
obtaining the model information and the historical log; wherein the model information comprises information of each layer in the model;
according to the information of each layer in the model,
for any of the layers to be evaluated,
if the historical log has the calculation time of the similar layer, estimating the calculation time of the layer to be estimated by using the calculation time of the similar layer in the historical log according to the similarity degree of the layer to be estimated;
if the similar layer calculation time does not exist, measuring the calculation time of the layer to be estimated; wherein the content of the first and second substances,
the similar layer refers to a layer which has the same key attributes with all the corresponding layers to be estimated and is of the same type as the layer to be estimated; wherein, the layers of the same type refer to the layers with the same input and output scales as the layer to be estimated; the key attribute refers to a layer attribute having nonlinear influence on the calculation time in all layer attributes; the non-key attribute refers to a layer attribute having a linear influence on the calculation time in all layer attributes; the similarity degree is determined according to the partial order relation between the similar layer and the layer to be estimated, and the similar layer is considered to be most similar to the layer adjacent to the layer to be estimated in the partial order relation; the partial order relation satisfies the following conditions: for any non-critical attribute of the forward influence layer calculation time, if the non-critical attribute monotonically increases, the calculation time monotonically increases, and for any non-critical attribute of the reverse influence layer calculation time, if the non-critical attribute monotonically decreases, the calculation time monotonically increases.
2. The method of estimating layer computation time in a model using computation time of similar layer as claimed in claim 1, wherein the computation time of similar layer in the history log is obtained from measurement.
3. The method for estimating layer computation time in a model using computation time of similar layers as set forth in claim 1,
if the historical logs have the calculation time of the same layer, taking the calculation time of the same layer as the calculation time of the layer to be estimated; wherein, the same layer refers to a similar layer which has the same non-key properties with all the corresponding non-key properties of the layer to be evaluated;
if there is no same layer calculation time, the estimation or measurement is performed based on the calculation time of other similar layers.
4. The method of claim 3, wherein the estimating of the computation time using the similarity layer comprises:
selecting similar layers of which the calculation time is greater than that of the layer to be estimated in the partial order relation, sorting the calculation times of the similar layers according to the magnitude, and selecting the minimum value as the maximum calculation time of the layer to be estimated; selecting similar layers of which the calculation time is less than that of the layer to be estimated in the partial order relation, sorting the calculation times of the similar layers according to the size, and selecting the maximum value as the minimum calculation time of the layer to be estimated; and estimating the calculation time of the layer to be estimated according to the maximum calculation time and the minimum calculation time.
5. A method of assessing whether an estimation of layer computation time in a neural network model is acceptable,
for any layer to be evaluated according to claims 1 to 4,
respectively setting the maximum calculation time and the minimum calculation time of the layer to be estimated to be initialized to 0 and infinity;
selecting similar layers with the calculation time larger than that of the layer to be estimated in the partial order relation from a historical log, sorting the calculation times of the similar layers according to the size, and selecting the minimum value to be reassigned as the maximum calculation time of the layer to be estimated; selecting similar layers of which the calculation time is less than that of the layer to be estimated in the partial order relation, sorting the calculation times of the similar layers according to the size, and selecting the maximum value to be reassigned as the minimum calculation time of the layer to be estimated;
evaluating whether the estimation is acceptable by using the maximum calculation time and the minimum calculation time:
if the difference between the maximum calculation time and the minimum calculation time is less than a set threshold value, the estimation is acceptable; otherwise, the estimation is not acceptable.
6. A method for predicting layer computation time in a deep learning model using a similarity layer, the method comprising:
the model is a neural network model;
obtaining the model information and the historical log; wherein the model information comprises information of each layer in the model;
according to the information of each layer in the model,
for any of the layers to be predicted,
evaluating whether the evaluation of the layer to be predicted is acceptable according to the method of claim 5, if so, evaluating the computation time of the layer to be predicted by the method of any one of claims 1 to 4;
and if not, measuring the calculation time of the layer to be predicted.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811574365.XA CN109711535B (en) | 2018-12-21 | 2018-12-21 | Method for predicting layer calculation time in deep learning model by using similar layer |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811574365.XA CN109711535B (en) | 2018-12-21 | 2018-12-21 | Method for predicting layer calculation time in deep learning model by using similar layer |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109711535A CN109711535A (en) | 2019-05-03 |
| CN109711535B true CN109711535B (en) | 2021-05-11 |
Family
ID=66257187
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811574365.XA Active CN109711535B (en) | 2018-12-21 | 2018-12-21 | Method for predicting layer calculation time in deep learning model by using similar layer |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109711535B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101039178A (en) * | 2007-04-18 | 2007-09-19 | 华中师范大学 | Method for building hierachical trust model in open system |
| CN105303896A (en) * | 2015-09-07 | 2016-02-03 | 合肥飞友网络科技有限公司 | Method for precisely pre-estimating estimated arrival time of flight |
| CN106408562A (en) * | 2016-09-22 | 2017-02-15 | 华南理工大学 | Fundus image retinal vessel segmentation method and system based on deep learning |
-
2018
- 2018-12-21 CN CN201811574365.XA patent/CN109711535B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101039178A (en) * | 2007-04-18 | 2007-09-19 | 华中师范大学 | Method for building hierachical trust model in open system |
| CN105303896A (en) * | 2015-09-07 | 2016-02-03 | 合肥飞友网络科技有限公司 | Method for precisely pre-estimating estimated arrival time of flight |
| CN106408562A (en) * | 2016-09-22 | 2017-02-15 | 华南理工大学 | Fundus image retinal vessel segmentation method and system based on deep learning |
Non-Patent Citations (1)
| Title |
|---|
| Hike: A Hybrid Human-Machine Method for Entity Alignment in Large-Scale Knowledge Bases;Yan Zhuang等;《CIKM’17》;20171110;第1-10页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109711535A (en) | 2019-05-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108364195B (en) | User retention probability prediction method and device, prediction server and storage medium | |
| CN109922032B (en) | Method, device, equipment and storage medium for determining risk of logging in account | |
| CN107153874B (en) | Water quality prediction method and system | |
| CN110569166A (en) | Abnormality detection method, abnormality detection device, electronic apparatus, and medium | |
| CN110633194B (en) | Performance evaluation method of hardware resources in specific environment | |
| WO2017071369A1 (en) | Method and device for predicting user unsubscription | |
| US8180716B2 (en) | Method and device for forecasting computational needs of an application | |
| CN112445699A (en) | Strategy matching method and device, electronic equipment and storage medium | |
| CN104581749A (en) | A method and device for predicting service amount of data service of mobile network | |
| KR20220115357A (en) | A method and apparatus for generating future demand forecast data based on attention mechanism | |
| Almeida et al. | The impact of uncertainty in the measurement of progress in earned value analysis | |
| CN110083518B (en) | AdaBoost-Elman-based virtual machine software aging prediction method | |
| CN109670593B (en) | Method for evaluating and predicting layer calculation time in deep learning model | |
| CN111144673A (en) | Method, device and equipment for evaluating structure of organization personnel and computer readable medium | |
| CN109711555B (en) | Method and system for predicting single-round iteration time of deep learning model | |
| CN109711535B (en) | Method for predicting layer calculation time in deep learning model by using similar layer | |
| CN111784173B (en) | AB experiment data processing method, device, server and medium | |
| Attarzadeh et al. | Software development cost and time forecasting using a high performance artificial neural network model | |
| US9188968B2 (en) | Run-time characterization of on-demand analytical model accuracy | |
| CN111277710A (en) | Telephone traffic prediction method, system, computer device and storage medium | |
| CN116340883A (en) | Power distribution network data resource fusion method, device, equipment and storage medium | |
| CN113780666B (en) | Missing value prediction method and device and readable storage medium | |
| CN111612226B (en) | Group daily average arrival number prediction method and device based on hybrid model | |
| Inoue et al. | An all-stage truncated multiple change point model for software reliability assessment | |
| JP2008165412A (en) | Performance-computing device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 518000 a2405, building 9, zone 2, Shenzhen Bay science and technology ecological park, 3609 Baishi Road, community, Yuehai street, Nanshan District, Shenzhen City, Guangdong Province Applicant after: Shenzhen Zhixing Technology Co., Ltd Address before: Room 408, Building 3, 4 Chegongzhuang Street, Xicheng District, Beijing 100044 Applicant before: BEIJING HANHAI CLUSTAR TECHNOLOGY Co.,Ltd. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |