CN109711535A - A method of the time is calculated using similar layer predetermined depth learning model middle layer - Google Patents
A method of the time is calculated using similar layer predetermined depth learning model middle layer Download PDFInfo
- Publication number
- CN109711535A CN109711535A CN201811574365.XA CN201811574365A CN109711535A CN 109711535 A CN109711535 A CN 109711535A CN 201811574365 A CN201811574365 A CN 201811574365A CN 109711535 A CN109711535 A CN 109711535A
- Authority
- CN
- China
- Prior art keywords
- layer
- time
- similar
- calculating
- estimation
- 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.)
- Granted
Links
Abstract
The present invention provides a kind of methods for calculating the time using similar layer predetermined depth learning model middle layer.This method goes using the calculating time of layer similar in history log to estimate the calculating time of layer to be predicted by the determinant attribute of layer, the similar layer of non-key attribute evaluation, the similarity degree for assessing by the non-key attribute partial ordering relation of layer similar interlayer;It is unacceptable to (without the similar layer for estimation i.e. in the history log) that can not estimate or estimation, then measure the layer.The present invention calculates the time by being multiplexed known layer, avoids some duplicate measurement work, reduces overhead.In addition, calculating the layer of model the prediction of time The invention also achieves the localization of prediction, can be completed in a local equipment.
Description
Technical field
The present invention relates to deep learning technology fields;Similar layer predetermined deep learning model is utilized in particular it relates to a kind of
The method of middle layer calculating time.
Background technique
Deep learning has been widely applied in the industry such as banking and insurance business, safety monitoring at present.Depth
The core methed of habit is the feature for being analyzed and being described data using neural network model.The model of one deep learning of training needs
More wheel iteration are carried out by the frequency of the order of magnitude of the second.Secondly, deep learning generally requires more huge compared to linear algorithm
Data training can obtain accurate model.One neural network model of training generally requires a couple of days or the time of several weeks could
It completes.
Exactly because reason above, when only deep learning model being trained to have clearly a need for spending longer by single node
Between;In this regard, the deep learning frame of current main-stream is supported to realize distributed training pattern by parallel computation.Sensu lato point
Cloth training includes two kinds: one is data parallel, another kind is that model is parallel.Under the scene of data parallel, pass through prediction
The training time of deep learning model, to a couple of days easily, several weeks, the several months deep learning model training during resource tune
Match, model evaluation etc. suffers from important meaning.Especially in multiple and different model training tasks and in the case of deposit, in advance
Predict the training time, rational allocation resource will greatly improve resource utilization.In this regard, pass through the measurement history log of small number of layers,
And it is multiplexed time of above-mentioned layer in conjunction with similar layer relationship, expense when will substantially reduce iteration time in prediction model training.
In addition, each layer calculates the time in measurement/predetermined deep learning model, specific aim adjustment is carried out with layer granularity to model
Also important in inhibiting.
Summary of the invention
In view of this, the present invention provides a kind of method for calculating the time using similar layer predetermined deep learning model middle layer.
This method assesses phase by the determinant attribute of layer, the similar layer of non-key attribute evaluation, by the non-key attribute partial ordering relation of layer
Like the similarity degree of interlayer, and then go using the calculating time of layer similar in history log to estimate the calculating time of layer to be predicted;
It is unacceptable to (without the similar layer for estimation i.e. in the history log) that can not estimate or estimation, then measure this layer calculating
Time.
On the one hand, the present invention provides a kind of method using similar layer in layer attribute evaluation deep learning model.
The above-mentioned method using similar layer in layer attribute evaluation deep learning model, comprising:
The layer attribute of each layer in model is obtained, above-mentioned layer attribute includes layer input, output scale;
It is same type if it is inputted, output scale is identical to any of them two or more layers
Layer;
To the layer of any two or more than two same types, if their whole corresponding determinant attribute all phases
Together, then they are similar layer;
Wherein, the determinant attribute that front is addressed refers to that there are the layer categories of non-linear effects to the time is calculated in whole layer attributes
Property.Accordingly, to the time is calculated, there are the layer attributes of linear effect in whole layer attributes, are non-key attribute.
Optionally, in the method that front is addressed, to any two or more than two similar layers, if their whole phases
The non-key attribute answered is also all identical, i.e. the respective attributes of their whole are all identical, then they are identical layer.
On the other hand, the present invention provide it is a kind of assessment deep learning model middle layer between similarity degree method.
With reference to first aspect, between above-mentioned assessment deep learning model middle layer similarity degree method, to first aspect
Any similar layer set that the method addressed is assessed is determined according to the partial ordering relation of interlayer all or part of in set
The similarity degree of these layers:
To arbitrary two layers or more than two layers in above-mentioned similar layer set, if there is following partial orders for they
Relationship:
The positive non-key attribute that influences layer calculate time any on its, if the non-key attribute makes it when being increased monotonically
Time monotone increasing is calculated,
And the non-key attribute that adversely affects layer calculate time any to its, if the non-key attribute monotone decreasing hour makes
It calculates time monotone increasing;
The similarity degree of these above-mentioned layers is then assessed according to the partial ordering relation.In general, neighbouring layer, they
Similarity degree is higher.
Another aspect, the present invention provide a kind of side for calculating time estimation model middle layer and calculating the time using similar layer
Method.
In conjunction with the first, second aspect, the above-mentioned side for calculating time estimation model middle layer and calculating the time using similar layer
Method, comprising:
To any layer to be evaluated,
According to the calculating time of calculating this layer of time estimation for addressing the similar layer that method is assessed with first aspect;Its
In, above-mentioned similar layer calculates the time, is obtained by query history log;
If calculating the time (without for estimating that this layer calculates the time i.e. in history log without above-mentioned similar layer in history log
Similar layer situation), then this layer can not be estimated.
Optionally, the similar layer that front is addressed calculates the time, is to be obtained according to measurement, rather than estimate acquisition, to keep away
Exempt to amplify error when (and potentially Cascaded amplification) is estimated in this way.
Optionally, the calculating time for the calculating time estimation layer to be estimated according to similar layer that front is addressed, comprising:
If it exists identical layer calculate the time, then be multiplexed the identical layer and calculate the time, when calculating as the layer to be evaluated
Between;
Otherwise, method is addressed with second aspect and assesses to obtain similar layer journey similar to layer to be evaluated in history log
Degree therefrom selects suitable similar layer according to its interlayer similarity degree, and the meter of the time estimation layer to be evaluated is calculated according to it
Evaluation time.For example, the selection immediate similar layer of similarity degree, when calculating the calculating of the above-mentioned layer to be evaluated of time estimation with it
Between (one such simple estimation, i.e., directly using the calculating time of the similar layer as the calculating time of layer to be evaluated).
But in the method provided in above-described embodiment, even if having selected to calculate the time closest to the estimation of similar layer,
It is only to have selected most suitable layer for estimating on interlayer similarity degree, it, can not and on the calculating time that estimation obtains
Effective measurement is provided to assess estimation bring error.
Therefore further, optionally, the suitably similar layer of the selection that front is addressed calculates time estimation when estimating layer calculating
Between, comprising:
The partial ordering relation in method is addressed according to second aspect, from the similar layer in history log,
It selects to calculate similar layer of the time greater than the layer to be evaluated on partial ordering relation, these similar layers is calculated into times
(the calculating time i.e. in history log) sorts by size, when selecting max calculation of the minimum value therein as the layer to be evaluated
Between;
It selects to calculate similar layer of the time less than the layer to be evaluated on partial ordering relation, these similar layers calculating times is pressed
Size sequence, selects minimum of computation time of the maximum value therein as the layer to be evaluated;
According to above-mentioned max calculation time, the calculating time of the minimum of computation time estimation layer to be evaluated;For example, to this
Max calculation time, the minimum of computation time of layer to be evaluated average, the calculating time as layer to be evaluated.
Another aspect, the present invention provide a kind of whether acceptable method of estimation assessed and calculate layer the time.
In conjunction with second, third aspect, the estimation error problem of method is addressed for the third aspect, above-mentioned assessment is to layer meter
The whether acceptable method of the estimation of evaluation time, comprising:
For any layer to be evaluated that the third aspect is addressed,
Whether it is subjected to according to its max calculation time, the assessment estimation of minimum of computation time;Wherein,
If the difference of its max calculation time, minimum of computation time are less than the threshold value of setting, the estimation is acceptable;
Otherwise, the estimation is unacceptable;
Wherein, the partial ordering relation addressed based on second aspect,
The above-mentioned max calculation time is according to the calculating time is greater than the layer to be evaluated on partial ordering relation in history log
Similar layer assignment;The minimum of computation time is according to the calculating time is less than the phase of the layer to be evaluated on partial ordering relation in history log
Like layer assignment.For example, two layers adjacent with layer to be evaluated upstream and downstream on partial ordering relation in selection history log, corresponding
Ground is as its max calculation time, minimum of computation time.
In addition, above-mentioned threshold value, which can be set as the minimum of computation time, is multiplied by coefficient gamma;γ value will limit the mistake of every layer of estimation
Difference.
If in history log, only there are similar layers in the side of partial ordering relation for the layer to be evaluated, that is, it is only capable of the assignment most matter of fundamental importance
Evaluation time/minimum of computation time, then can not in the above way assess, and also think to estimate unacceptable.
It is optionally, a kind of to maximum, minimum of computation time assignment and appraisal procedure, comprising:
The max calculation time for the layer to be evaluated that front is addressed, minimum of computation time are initialized as 0 and infinity respectively;
The partial ordering relation in method is addressed according to second aspect, from the similar layer in history log,
It selects to calculate similar layer of the time greater than the layer to be evaluated on partial ordering relation, these similar layers is calculated into times
(the calculating time i.e. in history log) sorts by size, when selecting max calculation of the minimum value therein as the layer to be evaluated
Between;
It selects to calculate similar layer of the time less than the layer to be evaluated on partial ordering relation, these similar layers calculating times is pressed
Size sequence, selects minimum of computation time of the maximum value therein as the layer to be evaluated;
In turn, whether it is subjected to according to above-mentioned max calculation time, the assessment estimation of minimum of computation time:
If the difference of its max calculation time, minimum of computation time are less than the threshold value of setting, the estimation is acceptable;
Otherwise, the estimation is unacceptable.
Another aspect, the present invention provide a kind of method for calculating the time using similar layer predetermined deep learning model middle layer.
In conjunction with third, fourth aspect, the above-mentioned method for calculating the time using similar layer predetermined deep learning model middle layer,
Include:
To any layer to be predicted,
Whether the layer estimation to be predicted is subjected to the method assessment that fourth aspect is addressed,
If so, estimating the calculating time of the layer to be predicted in the method that the third aspect is addressed;
If not or the layer to be predicted can not be estimated, then measure the calculating time of this layer.
Technical solution provided by the invention, have it is many utility model has the advantages that
One, low overhead;The present invention has been multiplexed known layer as much as possible and has calculated the time, avoids the weight of identical layer or similar layer
Repetition measurement amount.
Two, prediction localization;Even to the model of training on distributed machines learning platform or utilizing online money
The model of source training can be completed on one device completely when predicting that it calculates the time through the invention.
Detailed description of the invention
It, below will be in the present invention one clearly to illustrate the embodiment of the present invention or technical solution in the prior art
The attached drawing that section Example is related to does simple introduction.
What Fig. 1 was that one embodiment of the invention provides a kind of utilizes similar layer predetermined deep learning model middle layer to calculate the time
The flow diagram of method.
Specific embodiment
Below with reference to the attached drawing of the embodiment of the present invention, technical solution in the embodiment of the present invention is carried out clearly and completely
Description.Obviously, described embodiment is only the embodiment of a part of the invention, instead of all the embodiments.Based on this hair
Embodiment in bright, those of ordinary skill in the art's every other implementation obtained under the premise of being not necessarily to creative work
Example, shall fall within the protection scope of the present invention.
The following are a preferred embodiment of the present invention.
Fig. 1 is a kind of method that the time is calculated using similar layer predetermined deep learning model middle layer that the embodiment provides
Flow diagram.The layer that above embodiment illustrates the model realized under tensorflow frame calculates the prediction of time.
As shown in Figure 1,
1) it before prediction, is parsed by data flow diagram, obtains the layer in model;
Wherein, flowing water line chart is defined using JSON, which describes file and each layer is described as JSON item
Mesh, including layer name and layer attribute;
The following are the description file example of one of model data flow graph,
[
{
"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 layer for obtaining model, to wherein each layer to be predicted, it whether there is its similar layer in query history log;
Wherein, as follows to similar layer and its related concept definitions:
A symbol definition
Layer input size: SI;Layer output scale: SO;Layer attribute: p;Layer attribute set: PL;Determinant attribute: pK;Key belongs to
Property set: PL K;Non-key attribute set PL;-PL K;Wherein, layer input size SI, layer export scale SOIt is in above-mentioned layer attribute p
One kind.
In the present embodiment, kernel_size etc. is nonlinear to time effects are calculated, and is determinant attribute, and
Batch_size, channels etc. are linear to the time is calculated, and are non-key attribute;
B same type layer
To any two or more than two layers, if their layer input size SI, layer export scale SOIt is all the same, then recognize
It is the layer of same type for them;
The similar layer of C
To the layer of any two or more than two same types, if their any corresponding determinant attribute p (p ∈ PL K)
It is all the same, then it is assumed that they are similar layer;
D identical layer
To any two or more than two similar layers, if their any corresponding non-key attribute p (p ∈ PL-PL K)
Also all the same, then it is assumed that they are identical layer;
E is capable of the partial ordering relation of each interlayer of entry evaluation similarity degree on calculating the time
To in any similar layer set, if any of them two layers or more than two layers, if they there is
Following partial ordering relation:
The positive non-key attribute that influences layer calculate time any on its, if the non-key attribute makes it when being increased monotonically
Time monotone increasing is calculated,
And the non-key attribute that adversely affects layer calculate time any to its, if the non-key attribute monotone decreasing hour makes
It calculates time monotone increasing;
As it is capable of the partial ordering relation of each interlayer of entry evaluation similarity degree on calculating the time.
In the present embodiment, if similar layer liWith layer ljWhole non-key attributes in there is above-mentioned partial ordering relation
(such as non-key attribute channels and batch size therein, the positive layer that influences calculate the time, i.e., increase with attribute value
And cause to calculate time increase, channels (li) < channels (lj), batch size (li) < batch size (lj),
And there is also corresponding partial ordering relations for other non-key attributes), then it is assumed that the two on calculating the time there are partial ordering relation,
li< lj。
3) the similar layer of the layer to be predicted if it exists then estimates whether it calculates time and assessment estimation using similar layer
It is acceptable:
Firstly, further judge the identical layer that whether there is the layer to be predicted in history log,
It is then multiplexed the identical layer if it exists and calculates the time as its calculating time;
If it does not exist,
It then has to take the second best, the calculating time of the time estimation layer to be predicted is calculated using its similar layer;
Wherein, before estimation, whether it is subjected to according to the maximum of the layer to be predicted, the assessment estimation of minimum of computation time:
The max calculation time of the layer to be evaluated, minimum of computation time are initialized as 0 and infinity respectively;
The partial ordering relation addressed according to front, from similar layer,
It selects to calculate similar layer of the time greater than the layer to be evaluated on partial ordering relation, these similar layers calculating times is pressed
Size sequence, selects max calculation time of the minimum value therein as the layer to be evaluated;
It selects to calculate similar layer of the time less than the layer to be evaluated on partial ordering relation, these similar layers calculating times is pressed
Size sequence, selects minimum of computation time of the maximum value therein as the layer to be evaluated;
If the difference of its max calculation time, minimum of computation time are less than threshold value Ts(Ts=γ tmin, wherein tminFor this wait for it is pre-
It surveys the minimum of computation time of layer, and γ value 20% in the present embodiment), then the estimation is acceptable;
It averages to above-mentioned max calculation time, minimum of computation time, the calculating time as the layer to be predicted.
Otherwise, the estimation is unacceptable (or can not estimate);
Then measure the calculating time of this layer.
The foregoing is merely a specific embodiment of the invention, but scope of protection of the present invention is not limited thereto.
Claims (10)
1. a kind of method using similar layer in layer attribute evaluation deep learning model, which is characterized in that this method comprises:
The layer attribute of each layer in model is obtained, the layer attribute includes layer input, output scale;
It is the layer of same type if it is inputted, output scale is identical to any of them two or more layers;
To the layer of any two or more than two same types, if their whole corresponding determinant attributes are all identical,
They are similar layer;
Wherein, the determinant attribute refers to that there are the layer attributes of non-linear effects to the time is calculated in whole layer attributes;It is described
Non-key attribute refer to that there are the layer attributes of linear effect the time to calculating in whole layer attributes.
2. the method according to claim 1 using similar layer in layer attribute evaluation deep learning model, which is characterized in that
To any two or more than two similar layers, if their whole corresponding non-key attributes are also all identical,
Then they are identical layer.
3. a kind of method of similarity degree between assessment deep learning model middle layer, which is characterized in that this method comprises:
To any similar layer set assessed in method described in claim 1, according to layer all or part of in set
Between partial ordering relation, determine the similarity degree of these layers:
To arbitrary two layers or more than two layers in the similar layer set, if there is following partial order passes for they
System:
The positive non-key attribute that influences layer calculate time any on its, if the non-key attribute makes its calculating when being increased monotonically
Time is increased monotonically,
And the non-key attribute that adversely affects layer calculate time any to its, if the non-key attribute monotone decreasing hour makes its meter
Evaluation time is increased monotonically;
The similarity degree of each layer is then assessed according to the partial ordering relation.
4. a kind of method for calculating time estimation model middle layer and calculating the time using similar layer, which is characterized in that this method packet
It includes:
To any layer to be evaluated,
When according to the calculating of calculating this layer of time estimation for the similar layer assessed with either method claimed in claims 1-2
Between;Wherein, the similar layer calculates the time, is obtained by query history log;
If calculating the time without the similar layer in history log, this layer can not be estimated.
5. the method according to claim 4 that calculate time estimation model middle layer and calculate the time using similar layer, special
Sign is that the similar layer calculates the time, is obtained according to measurement.
6. the method according to claim 4 that calculate time estimation model middle layer and calculate the time using similar layer, special
Sign is,
If it exists identical layer calculate the time, then be multiplexed the identical layer calculate the time, the calculating time as the layer to be evaluated;
Otherwise, assess to obtain the similarity degree of the similar layer and layer to be evaluated in history log in method as claimed in claim 3,
Suitable similar layer is therefrom selected accordingly, and the calculating time of the time estimation layer to be evaluated is calculated according to it.
7. the method according to claim 6 that calculate time estimation model middle layer and calculate the time using similar layer, special
Sign is that the suitably similar layer of the selection calculates time estimation layer to be estimated and calculates the time, comprising:
Partial ordering relation according to claim 3, from the similar layer in history log,
It selects to calculate similar layer of the time greater than the layer to be evaluated on partial ordering relation, these similar layers is calculated into times by size
Sequence, selects max calculation time of the minimum value therein as the layer to be evaluated;
It selects to calculate similar layer of the time less than the layer to be evaluated on partial ordering relation, these similar layers is calculated into times by size
Sequence, selects minimum of computation time of the maximum value therein as the layer to be evaluated;
According to the max calculation time, the calculating time of the minimum of computation time estimation layer to be evaluated.
8. a kind of whether acceptable method of estimation that assessment calculates layer the time, which is characterized in that
For any layer to be evaluated described in claim 4-7,
Whether it is subjected to according to its max calculation time, the assessment estimation of minimum of computation time;Wherein,
If the difference of its max calculation time, minimum of computation time are less than the threshold value of setting, the estimation is acceptable;
Otherwise, the estimation is unacceptable;
Wherein, it is based on partial ordering relation as claimed in claim 3, the max calculation time is closed according to partial order in history log
Fasten the similar layer assignment for calculating the time greater than the layer to be evaluated;The minimum of computation time is according to partial ordering relation in history log
The upper calculating time is less than the similar layer assignment of the layer to be evaluated.
9. the whether acceptable method of estimation that assessment according to claim 8 calculates layer the time, which is characterized in that
To the maximum, minimum of computation time assignment, assessment, comprising:
The max calculation time of the layer to be evaluated, minimum of computation time are initialized as 0 and infinity respectively;
Partial ordering relation according to claim 3, from the similar layer in history log,
It selects to calculate similar layer of the time greater than the layer to be evaluated on partial ordering relation, these similar layers is calculated into times by size
Sequence, selects max calculation time of the minimum value therein as the layer to be evaluated;
It selects to calculate similar layer of the time less than the layer to be evaluated on partial ordering relation, these similar layers is calculated into times by size
Sequence, selects minimum of computation time of the maximum value therein as the layer to be evaluated;
Whether it is subjected to according to the max calculation time, the assessment estimation of minimum of computation time,
If the difference of its max calculation time, minimum of computation time are less than the threshold value of setting, the estimation is acceptable;
Otherwise, the estimation is unacceptable.
10. a kind of method for calculating the time using similar layer predetermined deep learning model middle layer, which is characterized in that this method packet
It includes:
To any layer to be predicted,
Whether method assessment is subjected to the layer estimation to be predicted either according to claim 8-9,
If so, estimating the calculating time of the layer to be predicted with method either described in claim 4-7;
If not or the layer to be predicted can not be estimated, then measure the calculating time of this layer.
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 true CN109711535A (en) | 2019-05-03 |
CN109711535B 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 |
---|
YAN ZHUANG等: "Hike: A Hybrid Human-Machine Method for Entity Alignment in Large-Scale Knowledge Bases", 《CIKM’17》 * |
Also Published As
Publication number | Publication date |
---|---|
CN109711535B (en) | 2021-05-11 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110659744B (en) | Training event prediction model, and method and device for evaluating operation event | |
Henseler et al. | Goodness-of-fit indices for partial least squares path modeling | |
Duchon et al. | Forecasting volatility with the multifractal random walk model | |
CN110246037B (en) | Transaction characteristic prediction method, device, server and readable storage medium | |
Behboudian et al. | A long-term prediction of domestic water demand using preprocessing in artificial neural network | |
Chen et al. | Computation and application of the paired combinatorial logit stochastic user equilibrium problem | |
CN110705688B (en) | Neural network system, method and device for performing risk assessment on operation event | |
Wang et al. | An AHP-weighted aggregated data quality indicator (AWADQI) approach for estimating embodied energy of building materials | |
CN104217091A (en) | Website page view prediction method based on historical tendency weights | |
Luz et al. | Minimax-robust filtering problem for stochastic sequences with stationary increments and cointegrated sequences | |
Poloni et al. | A note on forecasting demand using the multivariate exponential smoothing framework | |
Liu et al. | Some asymptotic results of Gaussian random fields with varying mean functions and the associated processes | |
Kotillová | Very short-term load forecasting using exponential smoothing and ARIMA models | |
Breidt et al. | Nonparametric variance estimation under fine stratification: an alternative to collapsed strata | |
Inoue et al. | Software reliability growth modeling frameworks with change of testing-environment | |
Perera et al. | A goodness-of-fit test for a class of autoregressive conditional duration models | |
CN109711535A (en) | A method of the time is calculated using similar layer predetermined depth learning model middle layer | |
CN109670593A (en) | A method of assessment and predetermined depth learning model middle layer calculate the time | |
Sbrana et al. | Aggregation of exponential smoothing processes with an application to portfolio risk evaluation | |
Kang et al. | Unbiased sensitivity estimation of one-dimensional diffusion processes | |
Rohan | A time varying GARCH (p, q) model and related statistical inference | |
Di Graziano et al. | A new approach to the modeling and pricing of correlation credit derivatives | |
Susila et al. | Performance Analysis of the Triple Exponential Smoothing Method During the Covid19 Pandemic on Tourist Visit Data | |
Ghysels | Mixed frequency models | |
Bareche et al. | Statistical techniques for a numerical evaluation of the proximity of G/G/1 and G/M/1 queueing systems |
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 |
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. |
|
CB02 | Change of applicant information | ||
GR01 | Patent grant | ||
GR01 | Patent grant |