CN108462708B - HDP-HMM-based behavior sequence detection method - Google Patents
HDP-HMM-based behavior sequence detection method Download PDFInfo
- Publication number
- CN108462708B CN108462708B CN201810218284.XA CN201810218284A CN108462708B CN 108462708 B CN108462708 B CN 108462708B CN 201810218284 A CN201810218284 A CN 201810218284A CN 108462708 B CN108462708 B CN 108462708B
- Authority
- CN
- China
- Prior art keywords
- hdp
- data
- hidden state
- hmm
- model
- 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
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L63/00—Network architectures or network communication protocols for network security
- H04L63/14—Network architectures or network communication protocols for network security for detecting or protecting against malicious traffic
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/29—Graphical models, e.g. Bayesian networks
- G06F18/295—Markov models or related models, e.g. semi-Markov models; Markov random fields; Networks embedding Markov models
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/01—Protocols
- H04L67/02—Protocols based on web technology, e.g. hypertext transfer protocol [HTTP]
Abstract
The invention discloses a modeling and detecting method of an action sequence based on an HDP-HMM (high-level hierarchical hidden Markov model). Log transition probability of a state track is used as a judgment index, the method is simpler, two models are established to jointly determine a detection result from a positive angle and a negative angle, and the problem that judgment index parameter selection and calculation are too complex in the traditional detection method is solved. Compared with the prior art, the method provided by the invention has the advantages that the defects of HMM in model definition and parameter estimation are overcome, meanwhile, a better detection effect is achieved, and experiments show that the average detection rate of the modeling and detection method provided by the invention is 95.3%.
Description
Technical Field
The invention relates to the field of behavior pattern analysis, in particular to a modeling and detecting method of a behavior sequence based on an HDP-HMM.
Background
HMMs are widely used in many fields and have very good performance, but they have some disadvantages in themselves. HMMs are very limited in either the definition of model structure or the standard estimation method of model parameters, and there are many imperfect solutions to many real-world problems. For example, the maximum likelihood estimation method does not fully consider the complexity of the model, which is likely to cause overfitting of the parameters. In addition, the model structure of the conventional HMM must be determined in advance, that is, the observed value and hidden state of the model need to be determined in advance, which makes the model lack adaptability.
With respect to these disadvantages of HMM, Beal et al apply the theory of Hierarchical Dirichlet Process (HDP) to Hidden Markov models in the literature, and thus propose an HMM extension-based nonparametric bayesian Model, namely, Hierarchical Dirichlet Process-Hidden Markov Model (HDP-HMM), also called Infinite Hidden Markov Model (iHMM). Compared with the traditional HMM, the HDP-HMM is a data-driven learning algorithm, has the advantages of no need of predicting the structure of data, capability of automatically iterating state numbers and the like, and can reflect the real characteristics of the data better. The HDP-HMM is a nonparametric Bayes model and can estimate the state number in time sequence data, so that the HDP-HMM has a very good application prospect in a plurality of fields such as visual scene recognition, anomaly detection and the like.
Disclosure of Invention
The invention aims to solve the defects in the prior art and provide a modeling and detection method based on an HDP-HMM behavior sequence.
In order to achieve the purpose, the invention is implemented according to the following technical scheme:
a modeling method of an action sequence based on an HDP-HMM models the action sequence and uses a Beam Sampling algorithm to automatically iterate out model parameters, and comprises the following specific steps:
s1, establishing an HDP-HMM model, and selecting object sequence data of an HTTP request of a user to a server as observation quantity to describe the behavior of the user;
s2, preprocessing the data, and dividing the object sequence data in the step S1 into training data and testing data;
s3, representing a set of all HTTP request sequences in the training data as y ═ y { (y)1,...yT}；
S4, assigning a preliminary estimated prior value S to the model parameterpThen, training parameters of the HDP-HMM model by adopting a Beam Sampling method: the algorithm introduces an auxiliary variable u through judgment
And dividing a limited number of hidden states, then using dynamic programming to iteratively calculate the conditional probability of state transition, sampling all hidden state tracks, assuming that other variables are known, and iteratively sampling an auxiliary variable u, a hidden state s, a transition matrix pi, a shared parameter beta and other parameters.
Specifically, the step S4 is as follows:
s41, sampling an auxiliary variable u:
s42, sampling the hidden state track S: since the s-state trajectories are only satisfied for all t-states when the auxiliary variable u is given
The probability of time is non-zero and because of the nature of the truncated stick construction, the state trajectory is a finite number, so the state is sampled using dynamic programming, which is used to sample the state trajectory s using the following equation:
s43, sampling parameters pi, phi and beta: first, n is definedijFor the number of transitions of state i to state j in one step in state trace s, i, j ∈ { 1., K }, where K is the known number of states in state trace s.
In addition, the invention also provides a detection method of the behavior sequence based on the HDP-HMM, which comprises the following steps:
firstly, preprocessing object sequence data of an HTTP request to a server to construct a training data set and a test data set;
step two, determining a judgment index: because the user behavior pattern has the properties of multiple complex, regular and accidental, the normal behaviors of most users have the same or similar operations, and because the state track is a Markov chain, the transition probability matrix of the adjacent state in the state track is determined to be pi, so that the logarithmic mean transition probability of the determined state track s is defined to be
Then, taking | eta | as a behavior mode judgment index;
step three, predefining an observation sequence length threshold value omega0When the length of the observation value sequence reaches the threshold value omega0Calculating corresponding index parameters;
step four, establishing two models HDP-HMM1 and HDP-HMM2 for the data objects in the step one, wherein the HDP-HMM1 is used for describing the normal behavior mode of a legal user and is sensitive to the abnormal behavior mode; the HDP-HMM2 is used for describing abnormal behavior patterns of an intruder or a legal user, is sensitive to the normal behavior patterns, and inputs normal training data to train HDP-HMM1 model parameters and inputs abnormal training data to train HDP-HMM2 model parameters during detection;
step five, respectively sampling the input detection data of the two models to obtain the optimal state track from the trained models, and obtaining the state track eta from the state transition probability of the HDP-HMM1 model1The state trajectory eta is obtained from the state transition probability of the HDP-HMM2 model2Then respectively taking the absolute value and comparing with the decision threshold value1And2comparing, when the | eta is satisfied1|＞1And | η2|≤1If not, the output is abnormal, otherwise, the output is normal.
Compared with the prior art, the invention has the beneficial effects that:
1. the behavior sequence modeling and detecting method based on the HDP-HMM, which is provided by the invention, uses the logarithmic transition probability of the state track as a judgment index, is simpler, and establishes two models to jointly determine a detection result from a positive angle and a negative angle, thereby solving the problem that the selection and calculation of judgment index parameters are too complex in the conventional detecting method.
2. And a great deal of energy is not needed to be spent on deep analysis of the data before modeling, and model parameters of the data set are estimated.
3. The defects of the HMM in model definition and parameter estimation are overcome, a better detection effect is achieved, and experiments show that the average detection rate of the modeling and detection method is 95.3%.
Drawings
FIG. 1 is a diagram of the modeling process of the HDP-HMM-based behavior sequence of the present invention.
FIG. 2 is a flow chart of the HDP-HMM-based behavior sequence detection method of the present invention.
FIG. 3 shows an embodiment of the present inventionEta of two HDP-HMM models in the examples1| and | η2The value of | is plotted.
Detailed Description
The present invention will be further described with reference to specific examples, which are illustrative of the invention and are not to be construed as limiting the invention.
As shown in fig. 1, in the modeling method for an HDP-HMM-based behavior sequence according to this embodiment, an HDP-HMM model is used to model the behavior sequence, and a Beam Sampling algorithm is used to automatically iterate model parameters, which includes the following specific steps:
s1, establishing an HDP-HMM model, and selecting object sequence data of an HTTP request of a user to a server as observation quantity to describe the behavior of the user;
s2, preprocessing the data, and dividing the object sequence data in the step S1 into training data and testing data;
s3, representing a set of all HTTP request sequences in the training data as y ═ y { (y)1,...yT}；
S4, assigning a preliminary estimated prior value S to the model parameterpThen, training parameters of the HDP-HMM model by adopting a Beam Sampling method: the algorithm introduces an auxiliary variable u through judgment
Dividing a limited number of hidden states, then calculating the conditional probability of state transition by using dynamic programming iteration, sampling all hidden state tracks, assuming that other variables are known, and iteratively sampling an auxiliary variable u, a hidden state s, a transition matrix pi, a shared parameter beta and other parameters;
s41, sampling an auxiliary variable u:
s42, sampling the hidden state track S: since the s-state trajectories are only satisfied for all t-states when the auxiliary variable u is given
The probability of time is non-zero and because of the nature of the truncated stick construction, the state trajectory is a finite number, so the state is sampled using dynamic programming, which is used to sample the state trajectory s using the following equation:
s43, sampling parameters pi, phi and beta: first, n is definedijFor the number of transitions of state i to state j in one step in state trace s, i, j ∈ { 1., K }, where K is the known number of states in state trace s.
In addition, as shown in fig. 2, the embodiment further provides a detection method based on an HDP-HMM behavior sequence, which includes the following steps:
firstly, preprocessing object sequence data of an HTTP request to a server to construct a training data set and a test data set;
step two, determining a judgment index: because the user behavior pattern has the properties of multiple complex, regular and accidental, the normal behaviors of most users have the same or similar operations, and because the state track is a Markov chain, the transition probability matrix of the adjacent state in the state track is determined to be pi, so that the logarithmic mean transition probability of the determined state track s is defined to be
Then, taking | eta | as a behavior mode judgment index;
step three, predefining an observation sequence length threshold value omega0When the length of the observation value sequence reaches the threshold value omega0Calculating corresponding index parameters;
step four, establishing two models HDP-HMM1 and HDP-HMM2 for the data objects in the step one, wherein the HDP-HMM1 is used for describing the normal behavior mode of a legal user and is sensitive to the abnormal behavior mode; the HDP-HMM2 is used for describing abnormal behavior patterns of an intruder or a legal user, is sensitive to the normal behavior patterns, and inputs normal training data to train HDP-HMM1 model parameters and inputs abnormal training data to train HDP-HMM2 model parameters during detection;
step five, respectively sampling the input detection data of the two models to obtain the optimal state track from the trained models, and obtaining the state track eta from the state transition probability of the HDP-HMM1 model1The state trajectory eta is obtained from the state transition probability of the HDP-HMM2 model2Then respectively taking the absolute value and comparing with the decision threshold value1And2comparing, when the | eta is satisfied1|＞1And | η2|≤1If not, the output is abnormal, otherwise, the output is normal.
Simulation experiment
Experimental data
The dataset used in the experiments herein was from the NASA-HTTP (ftp:// ita. ee.lb. gov/trace/NASA _ access _ log _ Aug15.g z) dataset. All HTTP request data from the user on NASA kennedy space base server, florida, on days 8, 4, to 31 in 2015 is contained in the data set. Part of data is randomly selected and processed to obtain two data sets, including a positive sample set, namely, 5500 normal request data and a negative sample set, namely, 2300 abnormal request data, which are respectively marked as DS1 and DS2 and are time sequence data.
Experimental setup
DataSet1 and DataSet2 were roughly divided into 70% training data and 30% testing data, respectively. The HDP-HMM1 and HDP-HMM2 were trained using normal and abnormal training data sets, respectively. The model test is divided into two parts, one part uses the model to respectively detect normal and abnormal test data sets, and the other part tests a mixed data set of the normal and abnormal data sets. In the process, the models HDP-HMM1 and HDP-HMM2 respectively detect the same group of data and respectively output eta1And η2Then, the absolute value is respectively taken and compared with the decision threshold value1And2comparing, when the | eta is satisfied1|＞1And | η2|≤1If not, the output is abnormal, otherwise, the output is normal.
Decision threshold values for two models1，2The method is important for the detection performance of the model, and the text firstly carries out detection on the | eta1| and | η2The value of | is counted to obtain1And2and then, comparing by adopting a plurality of groups of experiments to finally determine the optimal judgment threshold value.
Results and analysis of the experiments
In the anomaly detection, a false positive rate (FP rate) and a false negative rate (TN rate) are two important indexes for measuring the performance of the detection system. The two indexes and the two decision threshold values in the model of the embodiment1And2in this regard, the relationship between the parameters needs to be further analyzed to obtain a better decision threshold, and thus better system performance.
The parameters of the HDP-HMM1 and the HDP-HMM2 are trained respectively, and the number of the finally converged states in the training data, namely the number of the normal behavior patterns is 32. The number of states in the anomaly data eventually converges to 9. The state transition trace diagram can intuitively show that the number of states of the abnormal behavior data is less, and the state transition trace is more regular.
The selection of the decision threshold has a great influence on the performance of the system. To obtain the preliminary decision threshold, we use training data as the model input pair | η1| and | η2The value of | is counted, as shown in fig. 3, it can be seen in fig. 3 that there are a few outliers, that is, there is a possibility of a few abnormal behaviors in DataSet 1. Similarly, there may be a small amount of normal behavior in DataSet 2. Determining the value of the decision threshold to exclude these few outliers to ensure the detection rate, and determining the decision threshold value from the graph1And2taken around 5.30 and 3.25, respectively.
And (3) counting the number TN of false alarms and the number FP of missed detections of the model by taking the test data as model input, and calculating the detection rate according to the following formula, wherein T and F are respectively positive and negative sample numbers.
Dividing the test data into three groups including normal test data of 1000 data, abnormal test data of 600 data and test data of 900 normal and 100 abnormal data, respectively detecting, and comparing1，2And comparing different values. According to experimental data, when1And2the detection effect is better when the detection is respectively near 5.25 and 3.30, and the average detection rate of multiple detections and calculations is 95.3%, so that the better detection effect is achieved.
The technical solution of the present invention is not limited to the limitations of the above specific embodiments, and all technical modifications made according to the technical solution of the present invention fall within the protection scope of the present invention.
Claims (2)
1. A detection method of an HDP-HMM-based behavior sequence comprises the following steps:
firstly, preprocessing object sequence data of an HTTP request to a server to construct a training data set and a test data set;
step two, determining a judgment index: because the user behavior pattern has the properties of multiple complex, regular and accidental, the normal behaviors of most users have the same or similar operations, and because the hidden state track is a Markov chain, the transition probability matrix of the adjacent state in the hidden state track is determined to be pi, so the logarithmic mean transition probability of the determined hidden state track s is defined to be
Then, taking | eta | as a behavior mode judgment index;
step three, predefining an observation sequence length threshold value omega0When the length of the observation value sequence reaches the threshold value omega0Calculating corresponding index parameters;
step four, establishing two models HDP-HMM1 and HDP-HMM2 for the data objects in the step one, wherein the HDP-HMM1 is used for describing the normal behavior mode of a legal user and is sensitive to the abnormal behavior mode; the HDP-HMM2 is used for describing abnormal behavior patterns of an intruder or a legal user, is sensitive to the normal behavior patterns, and inputs normal training data to train HDP-HMM1 model parameters and inputs abnormal training data to train HDP-HMM2 model parameters during detection;
step five, respectively sampling the input detection data of the two models to obtain the optimal hidden state track from the trained models, and obtaining the logarithmic mean transition probability eta of the hidden state track from the state transition probability of the HDP-HMM1 model1Obtaining the logarithm average transition probability eta of the hidden state track from the state transition probability of the HDP-HMM2 model2Then, the absolute value is respectively taken and compared with the decision threshold value1And2comparing, when the | eta is satisfied1|＞1And | η2|≤2If the output is abnormal, otherwise, the output is normal;
the HDP-HMM1 and HDP-HMM2 were created by the following steps:
s1, establishing an HDP-HMM model, and selecting object sequence data of an HTTP request of a user to a server as observation quantity to describe the behavior of the user;
s2, preprocessing the data, and dividing the object sequence data in the step S1 into training data and testing data;
s3, representing a set of all HTTP request sequences in the training data as y ═ y { (y)1,...yTT is a natural number assigned value;
s4, assigning a preliminary estimated prior value S to the model parameterpThen, training parameters of the HDP-HMM model by adopting a Beam Sampling method: the algorithm introduces an auxiliary variable u through judgment
Dividing a limited number of hidden states, then using dynamic programming to iteratively calculate the conditional probability of state transition, sampling all hidden state tracks, and iteratively sampling an auxiliary variable u, a hidden state track s, a transition probability matrix pi, a shared parameter beta and other parameters on the assumption that other variables are known。
2. The HDP-HMM-based behavior sequence detection method of claim 1, wherein the step S4 comprises the following steps:
s41, sampling an auxiliary variable u:
s42, sampling the hidden state track S: since the hidden state trajectories s are only satisfied for all t when the auxiliary variable u is given
The probability of time is non-zero, and because of the nature of the truncated stick structure, the hidden state trajectory is a finite number, so the hidden state trajectory is sampled using dynamic programming, applying the following equation, the hidden state trajectory s is sampled using dynamic programming:
s43, sampling the parameters pi, beta and other parameters.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810218284.XA CN108462708B (en) | 2018-03-16 | 2018-03-16 | HDP-HMM-based behavior sequence detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810218284.XA CN108462708B (en) | 2018-03-16 | 2018-03-16 | HDP-HMM-based behavior sequence detection method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108462708A CN108462708A (en) | 2018-08-28 |
| CN108462708B true CN108462708B (en) | 2020-12-08 |
Family
ID=63236746
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810218284.XA Active CN108462708B (en) | 2018-03-16 | 2018-03-16 | HDP-HMM-based behavior sequence detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108462708B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111353352B (en) * | 2018-12-24 | 2023-05-09 | 杭州海康威视数字技术股份有限公司 | Abnormal behavior detection method and device |
| CN111222319B (en) * | 2019-11-14 | 2021-09-14 | 电子科技大学 | Document information extraction method based on HDP model |
| CN111209942B (en) * | 2019-12-27 | 2023-12-19 | 广东省智能制造研究所 | Multi-mode sensing abnormality monitoring method for foot robot |
| CN111614695B (en) * | 2020-05-29 | 2022-04-22 | 华侨大学 | Network intrusion detection method and device of generalized inverse Dirichlet mixed HMM model |
| CN113642682B (en) * | 2021-10-14 | 2021-12-31 | 北京理工大学 | Trajectory primitive extraction and analysis method and system under multi-vehicle interaction environment |
| CN117079669A (en) * | 2023-10-17 | 2023-11-17 | 博上(山东)网络科技有限公司 | Feature vector extraction method for LSB audio steganography with low embedding rate |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8010341B2 (en) * | 2007-09-13 | 2011-08-30 | Microsoft Corporation | Adding prototype information into probabilistic models |
| CN103530603A (en) * | 2013-09-24 | 2014-01-22 | 杭州电子科技大学 | Video abnormality detection method based on causal loop diagram model |
| CN103678346A (en) * | 2012-09-07 | 2014-03-26 | 阿里巴巴集团控股有限公司 | Man-machine recognition method and system |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101557506B (en) * | 2009-05-19 | 2011-02-09 | 浙江工业大学 | Intelligent detecting device for violent behavior in elevator car based on computer vision |
| CN103473555B (en) * | 2013-08-26 | 2016-09-21 | 中国科学院自动化研究所 | Horrible video scene recognition method based on various visual angles multi-instance learning |
-
2018
- 2018-03-16 CN CN201810218284.XA patent/CN108462708B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8010341B2 (en) * | 2007-09-13 | 2011-08-30 | Microsoft Corporation | Adding prototype information into probabilistic models |
| CN103678346A (en) * | 2012-09-07 | 2014-03-26 | 阿里巴巴集团控股有限公司 | Man-machine recognition method and system |
| CN103530603A (en) * | 2013-09-24 | 2014-01-22 | 杭州电子科技大学 | Video abnormality detection method based on causal loop diagram model |
Non-Patent Citations (3)
| Title |
|---|
| 基于多示例学习的异常行为检测方法研究;崔永艳;《中国优秀硕士学位论文全文数据库（电子期刊），信息科技辑》;20111115(第11期);正文第2章、第4章 * |
| 基于环路LDA-HMM模型的视频异常事件检测;岳猛;《中国优秀硕士学位论文全文数据库（电子期刊），信息科技辑》;20140915(第9期);正文第3章 * |
| 崔永艳.基于多示例学习的异常行为检测方法研究.《中国优秀硕士学位论文全文数据库（电子期刊），信息科技辑》.2011,(第11期),正文第2章、第4章. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108462708A (en) | 2018-08-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108462708B (en) | HDP-HMM-based behavior sequence detection method | |
| CN110632572A (en) | Radar radiation source individual identification method and device based on unintentional phase modulation characteristics | |
| CN110502636B (en) | Joint modeling and mining method and system for subjective and objective test questions | |
| CN108664690A (en) | Long-life electron device reliability lifetime estimation method under more stress based on depth belief network | |
| CN108536784B (en) | Comment information sentiment analysis method and device, computer storage medium and server | |
| Le Son et al. | Remaining useful life estimation on the non-homogenous gamma with noise deterioration based on gibbs filtering: A case study | |
| Bastani et al. | Unsupervised trajectory pattern classification using hierarchical Dirichlet Process Mixture hidden Markov model | |
| Ulapane et al. | Hyper-parameter initialization for squared exponential kernel-based gaussian process regression | |
| CN114218998A (en) | Power system abnormal behavior analysis method based on hidden Markov model | |
| CN112738092A (en) | Log data enhancement method, classification detection method and system | |
| Choudhary et al. | On the runtime-efficacy trade-off of anomaly detection techniques for real-time streaming data | |
| JP6172317B2 (en) | Method and apparatus for mixed model selection | |
| CN111340233B (en) | Training method and device of machine learning model, and sample processing method and device | |
| CN114298299A (en) | Model training method, device, equipment and storage medium based on course learning | |
| Alasalmi et al. | Classification uncertainty of multiple imputed data | |
| CN112381213A (en) | Industrial equipment residual life prediction method based on bidirectional long-term and short-term memory network | |
| Subeno et al. | Optimisation towards Latent Dirichlet Allocation: Its Topic Number and Collapsed Gibbs Sampling Inference Process. | |
| CN116502161A (en) | Anomaly detection method based on dynamic hypergraph neural network | |
| Twomey et al. | An application of hierarchical Gaussian processes to the detection of anomalies in star light curves | |
| Visser et al. | Parameter recovery, bias and standard errors in the linear ballistic accumulator model | |
| CN111027680B (en) | Monitoring quantity uncertainty prediction method and system based on variational self-encoder | |
| Sumin et al. | Information model of trainee characteristics with definition of stochastic behavior of dynamic system | |
| Simen | Reaction Time Analysis for Interval Timing Research | |
| Subramanian et al. | Lorenz System State Stability Identification using Neural Networks | |
| Asanjan et al. | Quantum-compatible variational segmentation for image-to-image wildfire detection using satellite data |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |