CN106250924B - A kind of newly-increased category detection method based on multi-instance learning - Google Patents
A kind of newly-increased category detection method based on multi-instance learning Download PDFInfo
- Publication number
- CN106250924B CN106250924B CN201610600041.3A CN201610600041A CN106250924B CN 106250924 B CN106250924 B CN 106250924B CN 201610600041 A CN201610600041 A CN 201610600041A CN 106250924 B CN106250924 B CN 106250924B
- Authority
- CN
- China
- Prior art keywords
- packet
- classification
- protein
- super
- distance
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention discloses a kind of newly-increased category detection method based on multi-instance learning, extracts the crucial example in more examples " packet " first with crucial example detection algorithm more mature in multi-instance learning;Later, for each known class, its corresponding crucial example is combined into one " the super packet of classification ", while not being identified as crucial exemplary all composition examples one " the super packet of member ";Then, the distance between Bao Yuchao packet can be determined by subsequent metric learning.In the practical stage, for the packet of known class, its concept classification is determined apart from the super packet of nearest classification according to it;And the packet for increasing classification newly, since there is no the super packet of its corresponding classification, the super Bao Ze nearest apart from it should be the super packet of member, and newly-increased classification can be so determined that it is according to this situation.
Description
Technical field
The present invention relates to machine learning, application technology, in particular to multi-instance learning, newly-increased classification monitoring technology, measurement
Study, be it is a kind of both automatic concept classification prediction/classification can be carried out to existing classification, while newly-increased classification can be detected
Robust multi-instance learning algorithm.
Background technique
Study is considered as most promising machine learning approach from sample.If using the ambiguousness of training sample as
The criteria for classifying, then the research in the field at present is substantially established under three kinds of learning frameworks, i.e. supervised learning, unsupervised learning and by force
Chemistry is practised.
Supervised learning by learning to the training example marked with concept, with as correctly as possible to training set except
Exemplary concept label predicted.Here all training samples are all markd, therefore its ambiguousness is minimum.Non- prison
Educational inspector practises to be learnt by the training example marked to no concept, to find the structure hidden in data.Here all instructions
Practicing sample is all not have markd, therefore its ambiguousness highest.Intensified learning is by marking no concept but postponing with one
Award or effectiveness (the concept label that can be considered delay) associated trained example are learnt, to obtain certain from state to action
Mapping.Here all training samples are all markd, but unlike supervised learning, and label is delay, therefore
The ambiguousness of intensified learning is between supervised learning and unsupervised learning.
Middle and later periods the 1990s, researchers propose more examples in the research to pharmaceutical activity forecasting problem
The concept of study.In such study, training set is made of several " packets " for marking with concept, comprising several in each packet
The example for not having concept to mark.If at least one example is subordinate to Mr. Yu's concept classification in a packet, which is labeled as should
Classification;If a packet is not belonging to certain concept classification, any one example in the packet is not subordinate to the category.By to training
The study of packet, it is desirable to which learning system as correctly as possible predicts the concept label of the packet except training set.
Compared with supervised learning, the training example in multi-instance learning is no concept label, this in supervised learning
All training examples have concept label different;Compared with unsupervised learning, training package is that have concept label in multi-instance learning
, this does not have any concept label different yet in training sample from unsupervised learning;And compared with intensified learning, more examples
Again without the concept of timeliness delay in habit.Importantly, a sample is exactly one and shows in previous various learning frameworks
Example, i.e., sample and example are one-to-one relationships;And in multi-instance learning, a sample (wrapping) contains multiple examples,
I.e. sample and example are one-to-many corresponding relationships.Therefore, the ambiguousness of training sample and supervised learning in multi-instance learning, non-
Supervised learning, the ambiguousness of intensified learning are all different, this allows for previous learning method and is difficult to well solve problems.
Since multi-instance learning has unique property and broad application prospect, belong to a blind area of previous machine learning research,
It therefore is considered as a kind of new learning framework.
The application scenarios of existing multi-instance learning algorithm are classification type and fixed number of static environment, rather than concept
The variable Open Dynamic environment of classification.Such as, when constructing the image classification system based on multi-instance learning, in the number of training stage
It only include " elephant ", " fox " and " bird " three kinds of image concept classifications according to concentrating.And the system practical stage then most probably occurs newly
Image category, such as Tiger.At this point, existing multi-instance learning algorithm will simply can only belong to the sample of newly-increased classification (such as
The picture of tiger) it is mistakenly divided into certain a kind of (such as " fox ") of known class, can thus it make system in Open Dynamic
It fails in environment.Therefore, multi-instance learning needs a kind of Robust learning algorithm that can be detected to newly-increased classification.
Summary of the invention
Goal of the invention: current multi-instance learning algorithm can only to existing classification sample carry out concept classification prediction/point
Class, under the scene for having newly-increased classification to occur, existing algorithm newly-increased classification sample is simple and mistake can only be divided into known class
One of not.In view of the above-mentioned problems, present invention firstly provides and solve the newly-increased classification under multi-instance learning scene detection
Its form, is turned to the frame of a metric learning by task, and proposes the corresponding new class detection learning algorithm of more examples.It is specific next
It says, extracts the crucial example in more examples " packet " first with crucial example detection algorithm more mature in multi-instance learning
Out.These so-called " crucial examples " just refer to that those in " packet " can determine the example of the concept label of corresponding packet.Later,
For each known class, its corresponding crucial example is combined into one " the super packet of classification ", while not being identified as key and showing
All composition examples one " the super packet of member " of example.Then, the distance between Bao Yuchao packet can pass through subsequent metric learning
To determine.Wherein, the super packet of classification is used for the classification of concept classification, and the super Bao Ze of member is used to increase newly the detection of classification.In training rank
Section, for the sample (packet) of known class, its some crucial example is wrapped its corresponding classification is super, in addition some
Example is in the super packet of member.But in order to obtain better discriminant classification ability, it is known that the packet of class is answered at a distance from the super packet of corresponding classification
To take second place at a distance from minimum, with the super packet of member, and answered at a distance from the super packet of other classifications larger.In the system practical stage, for
Know the packet of classification, its concept classification can be determined apart from the super packet of nearest classification according to it;And the packet for increasing classification newly, by
In there is no the super packet of its corresponding classification, super Bao Ze nearest apart from it should be the super packet of member, so can according to this situation come
Determine that it is newly-increased classification.
Technical solution: a kind of newly-increased category detection method based on multi-instance learning, including multi-instance learning disaggregated model
Training step and disaggregated model prediction steps;
More example classification model training steps specifically:
Step 1.1, on existing more sample datas, using existing crucial example detection algorithm from each more example packets
XiIt is middle to extract corresponding crucial example;
Step 1.2, crucial composition examples " the super packet of the classification " S packet for being under the jurisdiction of identical concept classification extractedc(c
=1 ..., C).In addition, not being detected as crucial exemplary all composition examples " the super packet of member " S0;
Step 1.3, distance of the definition packet to super packet are as follows: the weighted sum of each example in super packet to the packet shortest distance,
In, the shortest distance of example to packet is example exemplary minimum euclidean distance into packet, and each exemplary weight is by spending in super packet
Amount study is determined according to training data training.The super packet is also possible to the super packet of member either the super packet of classification.
Step 1.4, according to packet to the distance of super packet, utilization measure learns to learn the weight.In order to guarantee the system
Have enough discriminant classification abilities and new class detectability, need to meet following condition in metric learning: (1) packet corresponds to class to it
The distance that Chao do not wrap is minimum;(2) distance of packet to the super packet of member is second small;(3) it wraps to the super distance wrapped of other classifications before being greater than
The two.On this basis, the training of model is carried out.
More example classification model prediction steps specifically:
Step 2.1, measurement obtained in utilization measure learning process calculates separately more example packets in test set to classification
The distance of super packet and the super packet of member;
Step 2.2, according to the concept classification of the more example packets of calculated distance discriminating test collection.
It is described from each more example packet XiIt is middle to extract corresponding crucial example, specifically: according to multi-instance learning it is assumed that
If wrapping XiIt is under the jurisdiction of in concept classification c, then XiIn at least one example be subordinate to c;If wrapping XiIt is not affiliated in concept classification c, then
XiIn any one example be all not affiliated with c.After using existing crucial example extraction algorithm, X is wrappediIn example i.e. be divided into
Crucial example is (to indicate packet XiConcept classification ownership) and non-key example.
The super packet of the crucial composition examples classification that the packet for being under the jurisdiction of identical concept classification is extracted, specifically:
After extracting its crucial example to the packet for belonging to concept classification c, these one set of crucial composition examples are denoted asWhereinFor i-th of crucial example, VcFor the exemplary number of key for being under the jurisdiction of classification c, ScIt is as general
Read the super packet of the corresponding classification of classification c.Similarly, the super package definition of all non-key composition examples is the super packet of member, is denoted asWhereinFor the example of the super packet of composition member, as i-th of non-key example in the super packet of member.
The packet is specially to the super distance wrapped Wherein VkFor super packet SkIn example numbers, C indicate classification number,ForIn packet XiIn arest neighbors,Table
Show super packet SkIn v-th of example, MkIt is the parameter that is obtained by metric learning (i.e. from XiTo SkMeasurement).
The metric learning learns weight, is that will wrap to the distance of super packet the objective function for writing a belt restraining, leads to
The solution of traditional alternative optimization is crossed to solve to obtain required parameter.
It is described in model prediction step, utilize classifier calculated packet obtained in training process to the distance of super packet, tool
Body are as follows: the measurement M that will learn in model training stepkIt brings into The distance that packet arrives each super packet can be obtained.
The concept classification according to the more example packets of calculated distance discriminating test collection, specifically: if packet arrives certain class
S Chao not wrappedcDistance it is nearest, then the packet is under the jurisdiction of concept classification c;If packet wraps S to member is super0Distance it is nearest, then the packet is under the jurisdiction of
Newly-increased concept classification.
The utility model has the advantages that compared with prior art, it is provided by the present invention to be determined using the super packet of construction and metric learning
Know the more example packets of classification while detecting newly-increased class method for distinguishing, it is more to have can make full use of flag data training in implementation process
Classifier of learning from example can classify to the more sample datas of known class suitable for the scene for having newly-increased classification to occur, while can be with
Detect newly-increased classification.In addition, the method for the invention can still obtain excellent performance under without the scene that classification occurs is increased newly.
Detailed description of the invention
Fig. 1 is more example classification model training work flow diagrams of the embodiment of the present invention;
Fig. 2 is more example classification model prediction work flow diagrams of the embodiment of the present invention.
Specific embodiment
Combined with specific embodiments below, the present invention is furture elucidated, it should be understood that these embodiments are merely to illustrate the present invention
Rather than limit the scope of the invention, after the present invention has been read, those skilled in the art are to various equivalences of the invention
The modification of form falls within the application range as defined in the appended claims.
As shown in Figs. 1-2, the newly-increased category detection method based on multi-instance learning, including the classification of protein multi-instance learning
Model training step and protein classification model prediction step;
As shown in Figure 1, more example classification model training steps specifically:
Step 1.1, on the more sample datas of existing protein, using existing crucial example detection algorithm from each egg
The more example packet X of white matteriIt is middle to extract corresponding key protein matter example;" key protein matter example " just refers to that those in " packet " can be with
Determine the example of the concept label of corresponding packet.
From the more example packet X of each proteiniIt is middle to extract corresponding key protein matter example, specifically: according to more examples
Practise it is assumed that if packet XiIt is under the jurisdiction of in protein classification c, then XiIn at least one example be subordinate to c;If wrapping XiIt is not affiliated with egg
In white matter classification c, then XiIn any one example be all not affiliated with c.After using existing crucial example extraction algorithm, X is wrappediIn
Protein example i.e. be divided into key protein matter example (come indicate packet XiProtein classification ownership) and non-key protein
Example.
Step 1.2, the key protein matter composition examples " protein packet for being under the jurisdiction of same protein classification extracted
The super packet of classification " Sc(c=1 ..., C).In addition, not being detected as the exemplary all composition examples of key protein matter, " protein member is super
Packet " S0;
The super packet of the key protein matter composition examples protein classification that the packet for being under the jurisdiction of same protein classification is extracted,
Specifically: after extracting its key protein matter example to the packet for belonging to protein classification c, by these key protein matter example sets
Gather at one, is denoted asWhereinFor i-th of key protein matter example, VcFor the pass for being under the jurisdiction of classification c
The exemplary number of key protein, ScThe super packet of the corresponding protein classification of as protein classification c.Similarly, all non-key eggs
The super package definition of white matter composition examples is the super packet of protein member, is denoted as
Step 1.3, distance of the definition packet to super packet are as follows: the weighted sum of each example in super packet to the packet shortest distance,
In, the shortest distance of example to packet is example exemplary minimum euclidean distance into packet, and each exemplary weight is by spending in super packet
Amount study is determined according to training data training.
The packet is specially to the super distance wrapped Wherein VkFor super packet SkIn example numbers,ForIn packet XiIn arest neighbors, MkTo be obtained by metric learning
Parameter (i.e. from XiTo SkMeasurement), C indicate classification number,Indicate super packet SkIn v-th of example.
Metric learning learns weight, is that will wrap to the distance of super packet the objective function for writing a belt restraining, passes through biography
The solution of alternative optimization of uniting solves to obtain required parameter.
Step 1.4, according to packet to the distance of super packet, utilization measure learns to learn the weight.In order to guarantee the system
Have enough discriminant classification abilities and new class detectability, need to meet following condition in metric learning: (1) packet is corresponding to its
The distance of the super packet of protein classification is minimum;(2) distance of packet to the super packet of protein member is second small;(3) packet arrives other oroteins
The distance of the super packet of classification is greater than the above two.On this basis, the training of model is carried out.
As shown in Fig. 2, more example classification model prediction steps specifically:
Step 2.1, the more example packets of protein in test set are calculated separately to egg using measurement obtained in training process
The distance of the super packet of white matter classification and the super packet of protein member;The measurement M that will learn in model training stepkIt brings intoThe distance that packet arrives each super packet can be obtained.
Step 2.2, according to the protein classification of the more example packets of calculated distance discriminating test collection: if packet arrives certain albumen
The super packet S of matter classificationcDistance it is nearest, then the packet is under the jurisdiction of protein classification c;If packet wraps S to protein member is super0Distance it is nearest,
Then the packet is under the jurisdiction of newly-increased protein classification.
Claims (7)
1. a kind of newly-increased category detection method based on multi-instance learning, it is characterised in that: including protein multi-instance learning point
Class model training step and protein classification model prediction step;
More example classification model training steps specifically:
Step 1.1, on the more sample datas of existing protein, using existing crucial example detection algorithm from each protein
More example packetsIt is middle to extract corresponding key protein matter example;
Step 1.2, key protein matter composition examples " the protein classification packet for being under the jurisdiction of same protein classification extracted
Super packet ";In addition, not being detected as that key protein matter is exemplary all to be shown
Example composition " the super packet of protein member ";
Step 1.3, distance of the definition packet to super packet are as follows: the weighted sum of each example in super packet to the packet shortest distance, wherein show
The shortest distance of example to packet is example exemplary minimum euclidean distance into packet, and each exemplary weight is by metric learning in super packet
It is determined according to training data training;
Step 1.4, according to packet to the distance of super packet, utilization measure learns to learn the weight;It needs to meet in metric learning
Following condition: (1) distance that packet corresponds to the super packet of protein classification to it is minimum;(2) distance of packet to the super packet of protein member is the
Two;(3) distance of packet to the super packet of other oroteins classification is greater than the above two;On this basis, the training of model is carried out;
More example classification model prediction steps specifically:
Step 2.1, the more example packets of protein in test set are calculated separately to protein using measurement obtained in training process
The distance of the super packet of classification and the super packet of protein member;
Step 2.2, according to the protein classification of the more example packets of calculated distance discriminating test collection.
2. the newly-increased category detection method based on multi-instance learning as described in claim 1, it is characterised in that: described from each
The more example packets of proteinIt is middle to extract corresponding key protein matter example, specifically: according to multi-instance learning it is assumed that if packet
It is under the jurisdiction of protein classificationIn, thenIn at least one example be subordinate to;If packetIt is not affiliated with protein classificationIn, thenIn any one example be all not affiliated with;After using existing crucial example extraction algorithm, packetIn example i.e. be divided into
Key protein matter example and non-key protein example.
3. the newly-increased category detection method based on multi-instance learning as described in claim 1, it is characterised in that: described to be subordinate to
Surpass packet in the key protein matter composition examples protein classification that the packet of same protein classification is extracted, specifically: to same
Belong to protein classificationPacket extract its crucial example after, these one set of crucial composition examples are denoted as, whereinIt isA key protein matter example,To be under the jurisdiction of classificationKey protein matter show
The number of example,As protein classificationThe super packet of corresponding protein classification;Similarly, all non-key protein example sets
At super package definition be the super packet of protein member, be denoted as,For the protein for being under the jurisdiction of the super packet of protein member
Exemplary number.
4. the newly-increased category detection method based on multi-instance learning as described in claim 1, it is characterised in that: the packet arrives super
The distance of packet is specially,For super packet
In example numbers,When super packet, whenForIt is wrappingIn arest neighbors,To pass through measurement
Learn obtained parameter.
5. the newly-increased category detection method based on multi-instance learning as described in claim 1, it is characterised in that: the tolerance
Practising is that will wrap to the distance of super packet the objective function for writing a belt restraining to learn weight, passes through the solution of traditional alternative optimization
Method solves to obtain required parameter.
6. the newly-increased category detection method based on multi-instance learning as described in claim 1, it is characterised in that: described in model
In prediction steps, using classifier calculated packet obtained in training process to the distance of super packet, specifically: it will be walked in model training
The measurement learnt in rapidIt brings into, can be obtained and wrap to each
The distance of super packet,,For super packetIn example numbers,When super packet, whenWhen, surpass packetThat is the super packet of protein classification,ForIt is wrappingIn
Arest neighbors,For the parameter obtained by metric learning.
7. the newly-increased category detection method based on multi-instance learning as described in claim 1, it is characterised in that: described according to meter
The concept classification of the more example packets of obtained range estimation test set, specifically: if packet is wrapped to certain classification protein is superAway from
From nearest, then the packet is under the jurisdiction of protein classification;If packet is wrapped to member is superDistance it is nearest, then the packet is under the jurisdiction of newly-increased egg
White matter classification.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610600041.3A CN106250924B (en) | 2016-07-27 | 2016-07-27 | A kind of newly-increased category detection method based on multi-instance learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610600041.3A CN106250924B (en) | 2016-07-27 | 2016-07-27 | A kind of newly-increased category detection method based on multi-instance learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106250924A CN106250924A (en) | 2016-12-21 |
CN106250924B true CN106250924B (en) | 2019-07-16 |
Family
ID=57604236
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610600041.3A Active CN106250924B (en) | 2016-07-27 | 2016-07-27 | A kind of newly-increased category detection method based on multi-instance learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106250924B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109192314B (en) * | 2018-08-27 | 2021-08-31 | 山东大学 | Cervical vertebra health scoring device based on multi-example learning and application thereof |
CN110377636A (en) * | 2019-07-09 | 2019-10-25 | 南京工业大学 | A kind of Outliers Detection algorithm based on more example models |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP1480158A1 (en) * | 2003-05-21 | 2004-11-24 | Microsoft Corporation | Systems and methods for adaptive handwriting recognition |
CN102509084A (en) * | 2011-11-18 | 2012-06-20 | 中国科学院自动化研究所 | Multi-examples-learning-based method for identifying horror video scene |
CN103325125A (en) * | 2013-07-03 | 2013-09-25 | 北京工业大学 | Moving target tracking method based on improved multi-example learning algorithm |
CN104091038A (en) * | 2013-04-01 | 2014-10-08 | 太原理工大学 | Method for weighting multiple example studying features based on master space classifying criterion |
CN104361059A (en) * | 2014-11-03 | 2015-02-18 | 中国科学院自动化研究所 | Harmful information identification and web page classification method based on multi-instance learning |
CN104778457A (en) * | 2015-04-18 | 2015-07-15 | 吉林大学 | Video face identification algorithm on basis of multi-instance learning |
CN105045907A (en) * | 2015-08-10 | 2015-11-11 | 北京工业大学 | Method for constructing visual attention-label-user interest tree for personalized social image recommendation |
CN105117429A (en) * | 2015-08-05 | 2015-12-02 | 广东工业大学 | Scenario image annotation method based on active learning and multi-label multi-instance learning |
-
2016
- 2016-07-27 CN CN201610600041.3A patent/CN106250924B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP1480158A1 (en) * | 2003-05-21 | 2004-11-24 | Microsoft Corporation | Systems and methods for adaptive handwriting recognition |
CN102509084A (en) * | 2011-11-18 | 2012-06-20 | 中国科学院自动化研究所 | Multi-examples-learning-based method for identifying horror video scene |
CN104091038A (en) * | 2013-04-01 | 2014-10-08 | 太原理工大学 | Method for weighting multiple example studying features based on master space classifying criterion |
CN103325125A (en) * | 2013-07-03 | 2013-09-25 | 北京工业大学 | Moving target tracking method based on improved multi-example learning algorithm |
CN104361059A (en) * | 2014-11-03 | 2015-02-18 | 中国科学院自动化研究所 | Harmful information identification and web page classification method based on multi-instance learning |
CN104778457A (en) * | 2015-04-18 | 2015-07-15 | 吉林大学 | Video face identification algorithm on basis of multi-instance learning |
CN105117429A (en) * | 2015-08-05 | 2015-12-02 | 广东工业大学 | Scenario image annotation method based on active learning and multi-label multi-instance learning |
CN105045907A (en) * | 2015-08-10 | 2015-11-11 | 北京工业大学 | Method for constructing visual attention-label-user interest tree for personalized social image recommendation |
Non-Patent Citations (6)
Title |
---|
Key Instance Detection in Multi-Instance Learning;Guoqing Liu;《JMLR: Workshop and Conference Proceedings》;20121231;全文 |
Learning with Augmented Multi-Instance View;Yue Zhu;《JMLR: Workshop and Conference Proceedings》;20141231;全文 |
Scalable Multi-instance Learning;Xiu-Shen Wei;《2014 IEEE International Conference on Data Mining》;20141231;全文 |
基于强模态辅助的弱模态图像概念分类及检测;邹晓川;《南京大学学报》;20140331;第50卷(第2期);全文 |
多示例学习在多ROI乳腺肿瘤分类中的应用;胡聪;《智能计算机与应用》;20110630;第1卷(第1期);全文 |
构建新包空间的多示例学习方法;温超;《西安交通大学学报》;20110831;第45卷(第8期);全文 |
Also Published As
Publication number | Publication date |
---|---|
CN106250924A (en) | 2016-12-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104330721B (en) | IC Hardware Trojan detecting method and system | |
CN104200237B (en) | One kind being based on the High-Speed Automatic multi-object tracking method of coring correlation filtering | |
WO2022037130A1 (en) | Network traffic anomaly detection method and apparatus, and electronic apparatus and storage medium | |
CN106294590B (en) | A kind of social networks junk user filter method based on semi-supervised learning | |
US20150185270A1 (en) | Method for recognizing transformer partial discharge pattern based on singular value decomposition algorithm | |
CN112580664A (en) | Small target detection method based on SSD (solid State disk) network | |
CN106897738A (en) | A kind of pedestrian detection method based on semi-supervised learning | |
CN106570853A (en) | Shape and color integration insulator identification and defect detection method | |
WO2020001311A1 (en) | Method for detecting interference, apparatus, device, and storage medium | |
CN110084151A (en) | Video abnormal behaviour method of discrimination based on non-local network's deep learning | |
Schwaiger et al. | Is uncertainty quantification in deep learning sufficient for out-of-distribution detection? | |
CN106250924B (en) | A kind of newly-increased category detection method based on multi-instance learning | |
CN109298993A (en) | A kind of method, apparatus and computer readable storage medium detecting failure | |
CN107194418A (en) | A kind of Aphids in Rice Field detection method based on confrontation feature learning | |
CN106919164B (en) | Water conservancy unit failure analysis methods based on storehouse autocoder | |
CN106157688A (en) | The parking space detection method with big data and system is learnt based on the degree of depth | |
CN101251896B (en) | Object detecting system and method based on multiple classifiers | |
CN111397902A (en) | Rolling bearing fault diagnosis method based on feature alignment convolutional neural network | |
CN111506598A (en) | Fault discrimination method, system and device based on small sample self-learning fault migration | |
CN106330949B (en) | One kind being based on markovian intrusion detection method | |
CN109344886A (en) | Occlusion number plate distinguishing method based on convolutional neural network | |
CN111127448A (en) | Model and method for detecting air spring fault based on isolated forest | |
CN110363228A (en) | Noise label correcting method | |
CN109543743A (en) | A kind of refrigeration unit multiple sensor faults diagnosis method based on reconstruction prediction residual | |
CN109444604A (en) | A kind of DC/DC converter method for diagnosing faults based on convolutional neural networks |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
CB02 | Change of applicant information | ||
CB02 | Change of applicant information |
Address after: 210008 Hankou Road, Drum Tower District, Nanjing, Jiangsu Province, No. 22 Applicant after: Nanjing University Address before: 210046 Xianlin Avenue 163, Qixia District, Nanjing City, Jiangsu Province Applicant before: Nanjing University |
|
GR01 | Patent grant | ||
GR01 | Patent grant |