CN112396184B - Relation mining method and system based on graph structure data - Google Patents
Relation mining method and system based on graph structure data Download PDFInfo
- Publication number
- CN112396184B CN112396184B CN202011380368.7A CN202011380368A CN112396184B CN 112396184 B CN112396184 B CN 112396184B CN 202011380368 A CN202011380368 A CN 202011380368A CN 112396184 B CN112396184 B CN 112396184B
- Authority
- CN
- China
- Prior art keywords
- path
- graph structure
- attention weight
- structure data
- relation
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/02—Knowledge representation; Symbolic representation
- G06N5/022—Knowledge engineering; Knowledge acquisition
- G06N5/025—Extracting rules from data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/04—Inference or reasoning models
- G06N5/046—Forward inferencing; Production systems
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Abstract
The application discloses a relation mining method and a system based on graph structure data, wherein the method comprises the following steps: s1, acquiring and analyzing an image to obtain image structure data; s2, processing graph structure data based on a task layer, and performing relationship reasoning on the graph structure to obtain subtasks; s3, completing interaction with the environment according to the subtasks to obtain corresponding rewards; s4, feeding the corresponding rewards back to the task layer; s5, circulating the steps S2-S4 until the subtask obtaining the maximum rewards is completed. The system comprises: the system comprises an object and relation detection module, a task layer module, an action layer module, a feedback module and a circulation module. The present application directly models and exploits relationships between objects to provide interpreted performance when the same performance is achieved. The method and the system for relation mining based on the graph structure data can be widely applied to the field of reinforcement learning.
Description
Technical Field
The application belongs to the field of reinforcement learning, and particularly relates to a relation mining method and system based on graph structure data.
Background
Most of the existing deep reinforcement learning uses unstructured data such as images and the like, a strategy from an observation state to actions can be mechanically learned by means of nonlinear fitting of a neural network, however, the non-interpretability of the intermediate reasoning process makes the application of the method difficult in the fields with high safety requirements such as automatic driving and the like, and in addition, the implicit relation between objects learned by the neural network cannot be intuitively understood by human beings.
Disclosure of Invention
In order to solve the technical problems, the application aims to provide a relation mining method and a relation mining system based on graph structure data, which utilize structural expression to model the relation of objects, use deep learning to carry out relation propagation and reasoning in the structural relation, and learn strategies from states to actions in complex task scenes.
The first technical scheme adopted by the application is as follows: a relation mining method based on graph structure data comprises the following steps:
s1, acquiring and analyzing an image to obtain image structure data;
s2, processing graph structure data based on a task layer, and performing relationship reasoning on the graph structure to obtain subtasks;
s3, completing interaction with the environment according to the subtasks to obtain corresponding rewards;
s4, feeding the corresponding rewards back to the task layer;
s5, circulating the steps S2-S4 until the subtask obtaining the maximum rewards is completed.
Further, the step of obtaining and analyzing the image to obtain the image structure data specifically includes:
s101, acquiring an image and detecting the image to obtain a detected object and corresponding attributes of the object;
s102, representing the detected object as a point of a graph structure;
s103, embedding priori knowledge into the detected object and the corresponding attribute of the object to generate an edge of the graph structure;
s104, combining the points and the edges of the graph structure to obtain graph structure data.
Further, the step of processing the graph structure data based on the task layer and performing relationship reasoning on the graph structure to obtain subtasks specifically includes:
s201, obtaining a relationship side attention weight according to the graph structure data;
s202, obtaining a path attention weight according to graph structure data;
s203, carrying out weighted aggregation on the relation paths according to the relation side attention weight and the path attention weight to obtain path aggregation information;
s204, generating an action strategy according to the path aggregation information to obtain a subtask.
Further, the step of obtaining the attention weight of the relation edge according to the graph structure data specifically includes:
s2011, expanding the graph structure data to obtain a two-dimensional matrix;
s2012, inputting the two-dimensional matrix into a transducer framework to obtain a relationship side attention weight and a hidden vector in the first step;
s2013, re-inputting the hidden vector and the two-dimensional matrix into a transducer framework to obtain the attention weight and the corresponding hidden vector of the relationship side in the next step;
s2014, cycling the step S2013 until the repetition number reaches the maximum relation path length;
s2015, outputting the relationship side attention weight and the corresponding hidden vector of each step.
Further, the step of weighting and aggregating the relationship paths according to the relationship side attention weight and the path attention weight to obtain path aggregation information specifically includes:
s2031, selecting a current reasoning step, and weighting all relationship side information according to the relationship side attention weight of the current reasoning step to obtain the relationship side information of the current reasoning step;
s2032, selecting the length of the current path, and obtaining path information of the current length according to the relationship side information of each reasoning step of the current path;
s2033, carrying out weighted average on all path information according to the path attention weight to obtain path aggregation information.
Further, the generating the action policy according to the path aggregation information specifically includes deriving a state-action function value of the action atom according to the path aggregation information.
The second technical scheme adopted by the application is as follows: a graph structure data based relational mining system comprising the following modules:
the object and relation detection module is used for acquiring and analyzing the image to obtain image structure data;
the task layer module is used for processing the graph structure data based on the task layer and performing relationship reasoning on the graph structure to obtain subtasks;
the action layer module is used for completing interaction with the environment according to the subtasks and obtaining corresponding rewards;
the feedback module is used for feeding back the corresponding rewards to the task layer;
and the loop module is used for executing the loop instruction.
Further, the task layer module further comprises the following submodules:
a relationship side attention submodule for obtaining the attention weight of the relationship side according to the graph structure data;
the path attention sub-module is used for obtaining path attention weight according to the graph structure data;
the reasoning sub-module is used for carrying out weighted aggregation on the relation paths according to the attention weights of the relation edges and the path attention weights to obtain path aggregation information;
and the strategy sub-module is used for generating an action strategy according to the path aggregation information to obtain a sub-task.
The method and the system have the beneficial effects that: the application converts the high-dimensional picture data into the picture structure data, then utilizes the deep learning model to perform relation reasoning on the picture structure data to obtain a series of relation paths, provides a certain interpretability for the algorithm, can directly model and utilize the relation between objects, and can provide interpretation performance when the same performance is achieved.
Drawings
FIG. 1 is a flow chart of steps of a method for relational mining based on graph structure data in accordance with an embodiment of the present application;
FIG. 2 is a block diagram of a relational mining system based on graph structure data in accordance with an embodiment of the present application;
FIG. 3 is a schematic diagram of the image resolution to obtain the image structure data according to an embodiment of the present application;
FIG. 4 is a diagram illustrating the overall architecture and data flow of an embodiment of the present application.
Detailed Description
The application will now be described in further detail with reference to the drawings and to specific examples. The step numbers in the following embodiments are set for convenience of illustration only, and the order between the steps is not limited in any way, and the execution order of the steps in the embodiments may be adaptively adjusted according to the understanding of those skilled in the art.
As shown in fig. 1, the present application provides a graph structure data-based relation mining method, which includes the following steps:
s1, acquiring and analyzing an image to obtain image structure data;
s2, processing graph structure data based on a task layer, and performing relationship reasoning on the graph structure to obtain subtasks;
s3, completing interaction with the environment according to the subtasks to obtain corresponding rewards;
s4, feeding the corresponding rewards back to the task layer;
s5, circulating the steps S2-S4 until the subtask obtaining the maximum rewards is completed.
Further as a preferred embodiment of the method, the step of acquiring and analyzing the image to obtain the graph structure data specifically includes:
s101, acquiring an image and detecting the image to obtain a detected object and corresponding attributes of the object;
s102, representing the detected object as a point of a graph structure;
in particular, meaningful objects are abstracted out of unstructured state data, representing the objects as points in a graph structure, the input of which is unstructured data such as a picture.
S103, embedding priori knowledge into the detected object and the corresponding attribute of the object to generate an edge of the graph structure;
specifically, human prior knowledge descriptions such as up, down, left, right, distance, and the like are embedded into detected objects and attributes thereof such as coordinates, sizes, and the like, and prior knowledge is used for constructing correlations between every two objects, so that edges of the graph structure are generated.
S104, combining the points and the edges of the graph structure to obtain graph structure data.
Specifically, referring to fig. 3, based on human prior knowledge, we first determine objects to be extracted from the image, such as a door, a key, a stair, a rope, a conveyor belt, a person, a skeleton head, etc., which are points in the graph structure, and then embed human prior knowledge, such as on the left, on the top, far and near, accessibility, etc., to construct relationships between the objects, which are edges of the graph structure. The graph data is very visual, can be simply understood by people, and can mine the relation between objects.
In addition, the algorithm is applied to a task layer, and can conduct targeted relation reasoning according to the relation between objects in the graph structure, so that key tasks in the graph structure can be mined. As shown in the environment of fig. 3, the small person needs to acquire the key first, then reaches the right lower corner stair and finally passes through the door at the right upper corner to finish the clearance. The application can train the task layer by collecting rewards from the process of continuous trial and error, and finally excavates the key tasks which pass through. Meanwhile, according to the weight of the attention module, the key task can be selected according to the visual relation reasoning path answer, and compared with a reinforcement learning algorithm of a general black box, the method has better interpretability.
Further as a preferred embodiment of the method, the step of processing the graph structure data based on the task layer and performing relationship reasoning on the graph structure to obtain subtasks specifically includes:
s201, obtaining a relationship side attention weight according to the graph structure data;
s202, obtaining a path attention weight according to graph structure data;
s203, carrying out weighted aggregation on the relation paths according to the relation side attention weight and the path attention weight to obtain path aggregation information;
s204, generating an action strategy according to the path aggregation information to obtain a subtask.
Further as a preferred embodiment of the present application, the step of obtaining the relationship side attention weight according to the graph structure data specifically includes:
s2011, expanding the graph structure data M to obtain a two-dimensional matrix M';
s2012, inputting the two-dimensional matrix M' into a transducer frame to obtain the relationship side attention weight of the first stepHidden vector v 1 ；
S2013, hidden vector v i And re-inputting the two-dimensional matrix M' into a transducer framework to obtain the relationship side attention weight of the i+1 stepAnd corresponding hidden vector v i+1 ；
S2014, cycling the step S2013 until the repetition number reaches the maximum relation path length;
specifically, T is repeated, where T is the maximum relationship path length.
S2015, outputting the relationship side attention weight S of each step i And corresponding hidden vector v i 。
Further as a preferred embodiment of the present application, the step of obtaining the path attention weight according to the graph structure data specifically includes:
s2021, splicing the corresponding hidden vectors of each step into an integrated vector;
specifically, the hidden vector v 1 ，v 2 ，……，v T And splicing the vector v.
S2022, inputting the integrated vector v and the two-dimensional matrix M' into a transducer frame to obtain a path attention weight S φ 。
Further, as a preferred embodiment of the present application, the step of performing weighted aggregation on the relationship paths according to the relationship side attention weight and the path attention weight to obtain path aggregation information specifically includes:
s2031, selecting a current reasoning step and weighting all relationship side information according to the relationship side attention weight of the current reasoning step to obtain the relationship side information of the current reasoning step;
specifically, at the current reasoning step t, attention weights are followed according to the kth relationship of the current stepWeighting all relation side information: />Wherein M is k A relationship matrix representing a kth relationship.
S2032, selecting the length of the current path, and obtaining path information of the current length according to the relationship side information of each reasoning step of the current path;
specifically, the current path length is selected, the path information of the current length is obtained from the relation side information of each inference step of the current path, wherein the relation side information is represented by a matrix, the matrix of each inference step is continuously multiplied to obtain the path information of the current path length, and the relation side information of the previous t 'steps is aggregated when the current path length is t':
s2033, carrying out weighted average on all path information according to the path attention weight to obtain path aggregation information.
Specifically, according to the path attention weight, information of paths with different lengths is aggregated:where T is the maximum relationship path length.
Further as a preferred embodiment of the method, the generating the action policy according to the path aggregation information is specifically deriving a state-action function value of the action atom according to the path aggregation information.
Specifically, using action atoms to represent a single action, the input is the inference module result:
the output is the Q value of the action atom. Here, for each action relationship, we use a fully connected neural network to simulate, output:
in the above formula, action_a (x, x') represents an atom acting as a, Q represents a state-action value of taking the current action atom in the current state, S represents the current state, v x Representation of the sole heat of the corresponding entityCoding using a fully connected network MLP a A relationship matrix representing the action predicate a.
Referring to FIG. 4, the present application is based on the premise that objects and relationship detectors can abstract objects well and model relationships between objects. The environment first inputs the unstructured data, image, into the object and relationship detector, then the detector inputs the structured data of the graph into the task layer, which will assign tasks to be performed by the action layer, which interacts with the environment through the controller and collects rewards, which are then fed back to the task layer. The task layer will train the attention module, the reasoning module and the strategy module therein with rewards.
The application provides a novel deep reinforcement learning framework, which is used for abstracting a complex game environment of objects in real life, constructing graph structure data describing the relation between the objects by embedding human priori knowledge, and providing a certain interpretability by digging important key tasks and visualizing relation paths through an attention module, an inference module and a strategy module.
As shown in fig. 2, a graph structure data-based relationship mining system includes the following modules:
the object and relation detection module is used for acquiring and analyzing the image to obtain image structure data;
specifically, the object and relationship detection module realizes the expression from unstructured data to the structured state of the graph based on a supervised learning paradigm, and human priori knowledge can be embedded in the process. Therefore, at each time step, unstructured data is extracted by using the module, so that the structured expression of each step state can be obtained, and input is provided for the subsequent module.
The task layer module is used for processing the graph structure data based on the task layer and performing relationship reasoning on the graph structure to obtain subtasks;
the action layer module is used for completing interaction with the environment according to the subtasks and obtaining corresponding rewards;
the feedback module is used for feeding back the corresponding rewards to the task layer;
and the loop module is used for executing the loop instruction.
Further as a preferred embodiment of the system, the task layer module further comprises the following submodules:
a relationship side attention submodule for obtaining the attention weight of the relationship side according to the graph structure data;
specifically, the relationship side attention module has its input of the graph structure data M output by the detection module and its output of the relationship side attention weight
The path attention sub-module is used for obtaining path attention weight according to the graph structure data;
in particular, the path attention module, the input of which is the hidden vector output v of the relationship side attention module 1 ，v 2 ，……，v T And a two-dimensional matrix M' whose output is the path attention weight s φ . Wherein hidden vectors embed hidden information for paths of different lengths.
The reasoning sub-module is used for carrying out weighted aggregation on the relation paths according to the attention weights of the relation edges and the path attention weights to obtain path aggregation information;
in particular, there are many relationship paths in the graph structure, and the reasoning module performs weighted aggregation on the relationship paths according to the attention weight generated by the attention module, so as to embed the information of the path of the whole graph structure into the matrix. And in each reasoning step, weighting all the relation side information, then multiplying the relation side information with the path information obtained in the previous step by a matrix to obtain the path information with the current length, and then carrying out weighted average on all the path information according to the path attention weight to obtain a reasoning result.
And the strategy sub-module is used for generating an action strategy according to the path aggregation information to obtain a sub-task.
Specifically, reinforcement learning eventually needs to learn a strategy capable of maximizing the sum of external environmental rewards, so that the action module is used for deriving a state-action function value of an action atom according to path aggregation information obtained by the reasoning module, and outputting an action strategy.
The content in the system embodiment is applicable to the method embodiment, the functions specifically realized by the method embodiment are the same as those of the system embodiment, and the achieved beneficial effects are the same as those of the system embodiment.
While the preferred embodiment of the present application has been described in detail, the application is not limited to the embodiment, and various equivalent modifications and substitutions can be made by those skilled in the art without departing from the spirit of the application, and these equivalent modifications and substitutions are intended to be included in the scope of the present application as defined in the appended claims.
Claims (4)
1. The relation mining method based on the graph structure data is characterized by comprising the following steps of:
s1, acquiring and analyzing an image to obtain image structure data;
s2, processing graph structure data based on a task layer, and performing relationship reasoning on the graph structure to obtain subtasks;
s3, completing interaction with the environment according to the subtasks to obtain corresponding rewards;
s4, feeding the corresponding rewards back to the task layer;
s5, circulating the steps S2-S4 until the subtask obtaining the maximum rewards is completed;
the step of processing the graph structure data based on the task layer and performing relation reasoning on the graph structure to obtain subtasks specifically comprises the following steps:
s201, obtaining a relationship side attention weight according to the graph structure data;
s202, obtaining a path attention weight according to graph structure data;
s203, carrying out weighted aggregation on the relation paths according to the relation side attention weight and the path attention weight to obtain path aggregation information;
s204, generating an action strategy according to the path aggregation information to obtain a subtask;
the step of obtaining the relationship side attention weight according to the graph structure data specifically comprises the following steps:
s2011, expanding the graph structure data to obtain a two-dimensional matrix;
s2012, inputting the two-dimensional matrix into a transducer framework to obtain a relationship side attention weight and a hidden vector in the first step;
s2013, re-inputting the hidden vector and the two-dimensional matrix into a transducer framework to obtain the attention weight and the corresponding hidden vector of the relationship side in the next step;
s2014, cycling the step S2013 until the repetition number reaches the maximum relation path length;
s2015, outputting the relationship side attention weight and the corresponding hidden vector of each step;
the step of obtaining the path attention weight according to the graph structure data specifically comprises the following steps:
s2021, splicing the corresponding hidden vectors of each step into an integrated vector;
s2022, inputting the integration vector and the two-dimensional matrix into a transducer framework to obtain a path attention weight;
the step of carrying out weighted aggregation on the relation paths according to the relation side attention weight and the path attention weight to obtain path aggregation information specifically comprises the following steps:
s2031, selecting a current reasoning step and weighting all relationship side information according to the relationship side attention weight of the current reasoning step to obtain the relationship side information of the current reasoning step;
s2032, selecting the length of the current path, and obtaining path information of the current length according to the relationship side information of each reasoning step of the current path;
s2033, carrying out weighted average on all path information according to the path attention weight to obtain path aggregation information.
2. The method for mining relationships based on graph structure data according to claim 1, wherein the step of acquiring and analyzing the image to obtain the graph structure data specifically comprises the steps of:
s101, acquiring an image and detecting the image to obtain a detected object and corresponding attributes of the object;
s102, representing the detected object as a point of a graph structure;
s103, embedding priori knowledge into the detected object and the corresponding attribute of the object to generate an edge of the graph structure;
s104, combining the points and the edges of the graph structure to obtain graph structure data.
3. The method of claim 1, wherein the generating the action policy according to the path aggregation information is specifically deriving a state-action function value of the action atom according to the path aggregation information.
4. A graph structure data-based relational mining system, comprising the following modules:
the object and relation detection module is used for acquiring and analyzing the image to obtain image structure data;
the task layer module is used for processing the graph structure data based on the task layer and performing relationship reasoning on the graph structure to obtain subtasks;
the action layer module is used for completing interaction with the environment according to the subtasks and obtaining corresponding rewards;
the feedback module is used for feeding back the corresponding rewards to the task layer;
the circulation module is used for executing a circulation instruction;
the step of processing the graph structure data based on the task layer and performing relation reasoning on the graph structure to obtain subtasks specifically comprises the following steps:
s201, obtaining a relationship side attention weight according to the graph structure data;
s202, obtaining a path attention weight according to graph structure data;
s203, carrying out weighted aggregation on the relation paths according to the relation side attention weight and the path attention weight to obtain path aggregation information;
s204, generating an action strategy according to the path aggregation information to obtain a subtask;
the step of obtaining the relationship side attention weight according to the graph structure data specifically comprises the following steps:
s2011, expanding the graph structure data to obtain a two-dimensional matrix;
s2012, inputting the two-dimensional matrix into a transducer framework to obtain a relationship side attention weight and a hidden vector in the first step;
s2013, re-inputting the hidden vector and the two-dimensional matrix into a transducer framework to obtain the attention weight and the corresponding hidden vector of the relationship side in the next step;
s2014, cycling the step S2013 until the repetition number reaches the maximum relation path length;
s2015, outputting the relationship side attention weight and the corresponding hidden vector of each step;
the step of obtaining the path attention weight according to the graph structure data specifically comprises the following steps:
s2021, splicing the corresponding hidden vectors of each step into an integrated vector;
s2022, inputting the integration vector and the two-dimensional matrix into a transducer framework to obtain a path attention weight;
the step of carrying out weighted aggregation on the relation paths according to the relation side attention weight and the path attention weight to obtain path aggregation information specifically comprises the following steps:
s2031, selecting a current reasoning step and weighting all relationship side information according to the relationship side attention weight of the current reasoning step to obtain the relationship side information of the current reasoning step;
s2032, selecting the length of the current path, and obtaining path information of the current length according to the relationship side information of each reasoning step of the current path;
s2033, carrying out weighted average on all path information according to the path attention weight to obtain path aggregation information.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011380368.7A CN112396184B (en) | 2020-12-01 | 2020-12-01 | Relation mining method and system based on graph structure data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011380368.7A CN112396184B (en) | 2020-12-01 | 2020-12-01 | Relation mining method and system based on graph structure data |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112396184A CN112396184A (en) | 2021-02-23 |
| CN112396184B true CN112396184B (en) | 2023-09-05 |
Family
ID=74604875
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011380368.7A Active CN112396184B (en) | 2020-12-01 | 2020-12-01 | Relation mining method and system based on graph structure data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112396184B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2014146463A1 (en) * | 2013-03-19 | 2014-09-25 | 中国科学院自动化研究所 | Behaviour recognition method based on hidden structure reasoning |
| WO2020056984A1 (en) * | 2018-09-19 | 2020-03-26 | 平安科技（深圳）有限公司 | Shortest path query method, system, computer device and storage medium |
| CN111581343A (en) * | 2020-04-24 | 2020-08-25 | 北京航空航天大学 | Reinforced learning knowledge graph reasoning method and device based on graph convolution neural network |
-
2020
- 2020-12-01 CN CN202011380368.7A patent/CN112396184B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2014146463A1 (en) * | 2013-03-19 | 2014-09-25 | 中国科学院自动化研究所 | Behaviour recognition method based on hidden structure reasoning |
| WO2020056984A1 (en) * | 2018-09-19 | 2020-03-26 | 平安科技（深圳）有限公司 | Shortest path query method, system, computer device and storage medium |
| CN111581343A (en) * | 2020-04-24 | 2020-08-25 | 北京航空航天大学 | Reinforced learning knowledge graph reasoning method and device based on graph convolution neural network |
Non-Patent Citations (1)
| Title |
|---|
| 基于语法推导的溯源依赖关系路径模式挖掘算法;裴继升;叶晓俊;;清华大学学报(自然科学版)(06);第3-10页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112396184A (en) | 2021-02-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Saxena et al. | Generative adversarial networks (GANs) challenges, solutions, and future directions | |
| Nápoles et al. | FCM expert: software tool for scenario analysis and pattern classification based on fuzzy cognitive maps | |
| Bacanin et al. | Multivariate energy forecasting via metaheuristic tuned long-short term memory and gated recurrent unit neural networks | |
| CN111292195A (en) | Risk account identification method and device | |
| Sindhwani et al. | Performance analysis of deep neural networks using computer vision | |
| CN112529166A (en) | Fusion neuron model, neural network structure, training and reasoning method, storage medium and device | |
| Shindo et al. | Neuro-symbolic forward reasoning | |
| CN111199216A (en) | Motion prediction method and system for human skeleton | |
| Jankowski et al. | Interactive complex granules | |
| Joshi et al. | A survey of fractional calculus applications in artificial neural networks | |
| Pan et al. | Frequency-based search-control in dyna | |
| Klemmer et al. | Spate-gan: Improved generative modeling of dynamic spatio-temporal patterns with an autoregressive embedding loss | |
| Shindo et al. | α ILP: thinking visual scenes as differentiable logic programs | |
| Yoshimura et al. | Toward understanding acceleration-based activity recognition neural networks with activation maximization | |
| Zhu et al. | Multitask neural tensor factorization for road traffic speed-volume correlation pattern learning and joint imputation | |
| CN112396184B (en) | Relation mining method and system based on graph structure data | |
| Pal | Generative adversarial network-based cross-project fault prediction | |
| Ugur et al. | A web‐based tool for teaching neural network concepts | |
| Kolman et al. | Knowledge extraction from neural networks using the all-permutations fuzzy rule base: the LED display recognition problem | |
| Kumar | APTx: better activation function than MISH, SWISH, and ReLU's variants used in deep learning | |
| Ikram | A benchmark for evaluating Deep Learning based Image Analytics | |
| Khan et al. | A highly optimized multi-stage teacher-learner inspired particle swarm optimizer system | |
| Ansari et al. | Introduction To Machine Learning Techniques | |
| Nápoles et al. | Fuzzy cognitive maps: A business intelligence discussion | |
| Pu et al. | EAKT: Embedding Cognitive Framework with Attention for Interpretable Knowledge Tracing |
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 |