SlideShare a Scribd company logo
1 of 35
Download to read offline
Final Report:
Construction through Deep RNN
Representation Learning
of Knowledge Bases
KoreaUniversity,
DepartmentofComputerScience&Radio
CommunicationEngineering
MASSIVEDATAMANAGEMENT
Professor JaewooKang
1
2015010661
2015011155
2016010646
MinhwanYu
YonghwaChoi
BumsooKim
MASSIVE DATA MANAGEMENT Presentation format
Contents
01.Introduction
02.PTransE
1-1.Whatdowewant?
1-3.TransH
2-1.Whatisdifferent?
2-2.RelationPathRepresentation
2-3.RecurrentNeuralNetwork
2
03.Ouralgorithm
3-1.Objectfunction
3-2.Activationfunction
3-3.LongShortTermMemory
04.Evaluation
4-1.Results
4-2.Comparison&Analysis
1-4.TransR
1-5.PTransE
1-2.TransE
MASSIVE DATA MANAGEMENT Presentation format
Introduction
1-2.TransE
1-3.TransH
3
1-4.TransR
1-5.PTransE
1-1.Whatdowewant?
MASSIVE DATA MANAGEMENT Presentation format
1. Introduction
4
1-1. What do we want?
Large-Scale
KnowledgeBases
Freebase
DBpedia
Yago
Realworld triples
KnowledgeBase(KB)
MASSIVE DATA MANAGEMENT Presentation format
1. Introduction
5
1-1. What do we want?
Large-Scale
KnowledgeBases
Freebase
Realworld triples
∴KBsareveryincomplete
Misrelation
1.Norelationfound
2.Wrongrelationfound
DBpedia
Yago
MASSIVE DATA MANAGEMENT Presentation format
1. Introduction
6
1-1. What do we want?
How doweaddress mis-relations?
MASSIVE DATA MANAGEMENT Presentation format
1. Introduction
7
1-1. What do we want?
Q.Howdoweaddressmisrelation?
A1. Addexternalsourcesfor
completion
Manualandintuitional
Requirestoomuchlabor
Inefficient,time-consuming
PerformancedropsasKBgetslarger
MASSIVE DATA MANAGEMENT Presentation format
1. Introduction
8
1-1. What do we want?
Q.Howdoweaddressmisrelation?
A1. Addexternalsourcesfor
completion
A2. Referencingothertuples
forcompletion
Powerful&Efficient
Largelyexpandable
MorepowerfulasKBgetslarger
MASSIVE DATA MANAGEMENT Presentation format
1. Introduction
9
1-1. What do we want?
Wetriedout…
TransE
TransH
TransR
PTransE
CompletionofKBbyaddressingmisrelationsthrough
co-entity,co-relationreference
MASSIVE DATA MANAGEMENT Presentation format
1. Introduction
10
1-2. TransE
relation#1Boy Girl
relation#1Queen
TransE
Representingdirectrelationswith
differencesbetweenentityvectors
Simple&Effective
reference:TransE(Bordesetal.NIPS2013)
MASSIVE DATA MANAGEMENT Presentation format
1. Introduction
11
1-2. TransE
TransE
Simple&Effective
relation#1Boy Girl
relation#1Queen King
Representingdirectrelationswith
differencesbetweenentityvectors
reference:TransE(Bordesetal.NIPS2013)
MASSIVE DATA MANAGEMENT Presentation format
1. Introduction
12
1-2. TransE
However…
TransEcan’trepresentmorethan
onerelationshipbetweenentities.
Inrealworld,weconstructmany
relationshipswithmanysubjects.
MASSIVE DATA MANAGEMENT Presentation format
1. Introduction
13
1-3. TransH
TransH
AbletorepresentM-to-Mrelations
Representingprojectedrelationswith
differencesbetweenentityvectors
Entitiesandrelationshavedifferentcharacteristics
However,theyarerepresentedinthesamespace
reference:TransH(Wangetal.AAAI2014)
MASSIVE DATA MANAGEMENT Presentation format
1. Introduction
14
1-4. TransR
Abletorepresentdifferentcharacteristics
betweenentitiesandrelations
Representingprojectedentitiesand
mappingthemintorelationspace‘r’
TransR
However,thesewereallnotgoodenoughfor
detectingandaddressingmisrelations!!
reference:TransR(AAAI2015)
MASSIVE DATA MANAGEMENT Presentation format
1. Introduction
15
1-5. PTransE
PTransE
Representingrelationsthroughcompositionof
relationsbetweenentityvectors
Widelyexpandable&powerfulmethod
h
t
e1
r1
r2
r=r1⋅r2
r1:father r1:father
reference:emnlpprocessing
MASSIVE DATA MANAGEMENT Presentation format
1. Introduction
16
1-5. PTransE
PTransE
Representingrelationsthroughcompositionof
relationsbetweenentityvectors
Widelyexpandable&powerfulmethod
h
t
e1
r1
r2
r=r1⋅r2
r:grandfather
reference:emnlpprocessing
MASSIVE DATA MANAGEMENT Presentation format
PTransE
2-2.RelationPathRepresentation
2-3.RecurrentNeuralNetwork
17
2-1.Whatisdifferent?
MASSIVE DATA MANAGEMENT Presentation format
2. PTransE
18
2-1. What is different?
PTransE
LearnsonlyonesinglematrixEachrelationneedsitsown
projectionmap.
TransR
50,000relationsneeds50,000
enormousprojectionmaps!
Learnsn-steprelationalpaths
Learnsrelationsthatevendon’t
haveenoughtrainingdataIfarelationdoesn’thaveenough
trainingdata,itwillsufferfromlow
performance.
Learnsrelationsonlyinthetraining
dataset.
Learnsrelationsthatarenotinthe
trainingdatasetthroughrelation
paths.(Zero-shotKBinference)
MASSIVE DATA MANAGEMENT Presentation format
2. PTransE
19
2-2. Relation Path Representation
MicrosoftisbasedinSeattle.
WhichcountryisMicrosoftlocatedin?
Microsoft Seattle
IsBasedInCountryIn
???
RelationNOTFOUND!(Misrelation)
MASSIVE DATA MANAGEMENT Presentation format
2. PTransE
20
2-2. Relation Path Representation
‘Microsoft’isbasedin‘Seattle’.
‘Seattle’islocatedinstate‘Washington’
Microsoft Seattle Washington USA
IsBasedIn
‘Washington’islocatedincountry‘USA’
StateIn CountryIn
MASSIVE DATA MANAGEMENT Presentation format
2. PTransE
21
2-2. Relation Path Representation
‘Microsoft’isbasedin‘Seattle’.
‘Seattle’islocatedinstate‘Washington’
Microsoft USA
‘Washington’islocatedincountry‘USA’
CountryLocatedIn
‘Microsoft’islocatedin‘USA’.
MASSIVE DATA MANAGEMENT Presentation format
2. PTransE
22
2-3. Recurrent Neural Network
Microsoft Seattle Washington
IsBasedIn StateIn CountryIn
USA
CountryLocatedIn
RNN
RNN
Learningamatrix 𝑾𝑾𝒓𝒓 for
constructingcomposition
vector
StateLocatedIn
p= 𝒇𝒇( 𝑾𝑾𝒓𝒓 𝒄𝒄𝟑𝟑;𝒄𝒄𝟒𝟒 ) = 𝒕𝒕𝒕𝒕𝒕𝒕𝒕𝒕( 𝑾𝑾𝒓𝒓 𝒄𝒄𝟑𝟑;𝒄𝒄𝟒𝟒 )
𝐜𝐜𝟑𝟑 = 𝒇𝒇( 𝑾𝑾𝒓𝒓 𝒄𝒄𝟏𝟏;𝒄𝒄𝟐𝟐 )
𝒄𝒄𝟏𝟏 𝒄𝒄𝟐𝟐
𝒄𝒄𝟒𝟒
reference:CompositionalVectorSpaceModelsforKnowledgeBaseInference
MASSIVE DATA MANAGEMENT Presentation format
Our Algorithm
3-2.Activationfunction
3-3.LongShortTermMemory
23
3-1.Objectivefunction
MASSIVE DATA MANAGEMENT Presentation format
3. Our Algorithm
24
3-1. Objective function
Me Father Grandfather
FatherOf FatherOf
RNN
Compositionvector
𝐩𝐩 = 𝒇𝒇( 𝑾𝑾𝒓𝒓 𝒄𝒄𝟏𝟏;𝒄𝒄𝟐𝟐 )
𝒄𝒄𝟏𝟏 𝒄𝒄𝟐𝟐
Me Grandfather
GrandfatherOf
Update 𝑾𝑾𝒓𝒓
MASSIVE DATA MANAGEMENT Presentation format
3. Our Algorithm
25
3-1. Objective function
Compositionvector
𝐩𝐩 = 𝒇𝒇( 𝑾𝑾𝒓𝒓 𝒄𝒄𝟏𝟏;𝒄𝒄𝟐𝟐 )GrandfatherOf
n-steprelationvector
𝒍𝒍𝒍𝒍𝒍𝒍𝒍𝒍
Updating 𝑾𝑾𝒓𝒓 …
𝒍𝒍𝒍𝒍𝒍𝒍𝒍𝒍BackPropagation
ForwardPropagation 𝒑𝒑𝒑𝒑𝒑𝒑𝒑𝒑
−𝒕𝒕𝒕𝒕𝒕𝒕𝒕𝒕𝒕𝒕𝒕𝒕𝑼𝑼𝑼𝑼𝑼𝑼𝑼𝑼𝑼𝑼𝑼𝑼
𝐖𝐖𝐫𝐫
MASSIVE DATA MANAGEMENT Presentation format
3. Our Algorithm
26
3-2. Activation function
However,NeuralNetworkisbasicallyalinearoperation
𝐩𝐩 = 𝒇𝒇( 𝑾𝑾𝒓𝒓 𝒄𝒄𝟏𝟏;𝒄𝒄𝟐𝟐 )
Applyinganon-linearaftereach
operationsinordertolearnnon-
lineardecisionboundary
tanhistheconventionaldefault
non-linearactivationfunctionfor
RNNmodels.
MASSIVE DATA MANAGEMENT Presentation format
3. Our Algorithm
27
3-2. Activation function
Then,what’swrongwithtanh?
Whenbackpropagatinggradients
inordertoupdatethematrix,
thesegradientstendtoconvergeto
0andeliminatesthegradientflow.
Therefore,nogradientflows
backward,andtheparameterstays
unmodified.
GradientVanishingProblem
MASSIVE DATA MANAGEMENT Presentation format
3. Our Algorithm
28
3-3. Long Short Term Memory
AddressesGradientVanishingproblems
HandleLongtermdependencies
ForgetGate
MASSIVE DATA MANAGEMENT Presentation format
Evaluation
4-2.Comparison&Analysis
29
4-1.Results
MASSIVE DATA MANAGEMENT Presentation format
4. Evaluation
30
4-1. Results
WeusedFB15K,andFB40Kdatasets
FB15K FB40K
# Relations 1,345 1,336
# Entities 14,951 39,528
# Training set 483,142 370,648
# Validation set 50,000 67,948
# Test set 59,071 96,678
MASSIVE DATA MANAGEMENT Presentation format
4. Evaluation
31
4-2. Comparisons & Analysis
Hit 10 (RAW) Hit 10 (FILTER)
TransE 34.9% 47.1%
TransH 45.7% 64.4%
TransR 43.8% 65.5%
Hit 10 (RAW) Hit 10 (FILTER)
PTransE(RNN)
(Original model)
50.6% 82.2%
PTransE(LSTM)
(Our model)
53.1% 86.6%
PTransE,RNNcomposition
Baseline
MASSIVE DATA MANAGEMENT Presentation format
4. Evaluation
32
4-2. Comparisons & Analysis
Hit 10 (RAW) Hit 10 (FILTER)
TransE 34.9% 47.1%
TransH 45.7% 64.4%
TransR 43.8% 65.5%
Hit 10 (RAW) Hit 10 (FILTER)
PTransE(RNN)
(Original model)
50.6% 82.2%
PTransE(LSTM)
(Our model)
53.1% 86.6%
PTransE,RNNcomposition
(+9.3%)
Baseline
MASSIVE DATA MANAGEMENT Presentation format
4. Evaluation
33
4-2. Comparisons & Analysis
Hit 10 (RAW) Hit 10 (FILTER)
TransE 34.9% 47.1%
TransH 45.7% 64.4%
TransR 43.8% 65.5%
Hit 10 (RAW) Hit 10 (FILTER)
PTransE(RNN)
(Original model)
50.6% 82.2%
PTransE(LSTM)
(Our model)
53.1% 86.6%
PTransE,RNNcomposition
(+21.1%)
Baseline
MASSIVE DATA MANAGEMENT Presentation format
4. Evaluation
34
4-2. Comparisons & Analysis
Hit 10 (RAW) Hit 10 (FILTER)
PTransE(ADD)
(Original model)
51.8% 83.4%
PTransE(ADD)
(Our model)
52.1% 84.1%
PTransE(MUL)
(Original model)
47.4% 77.7%
PTransE(MUL)
(Our model)
47.1% 77.2%
PTransE(RNN)
(Original model)
50.6% 82.2%
PTransE(LSTM)
(Our model)
53.1% 86.6%
PTransE,Compositioncomparison
Thank you for your attention!
35

More Related Content

What's hot

Presentation on "Mizan: A System for Dynamic Load Balancing in Large-scale Gr...
Presentation on "Mizan: A System for Dynamic Load Balancing in Large-scale Gr...Presentation on "Mizan: A System for Dynamic Load Balancing in Large-scale Gr...
Presentation on "Mizan: A System for Dynamic Load Balancing in Large-scale Gr...Zuhair khayyat
 
Reactive/Proactive Connectivity Management in a Tactical Service-Oriented Inf...
Reactive/Proactive Connectivity Management in a Tactical Service-Oriented Inf...Reactive/Proactive Connectivity Management in a Tactical Service-Oriented Inf...
Reactive/Proactive Connectivity Management in a Tactical Service-Oriented Inf...Roberto Rigolin F. Lopes
 
Segmentation of Images by using Fuzzy k-means clustering with ACO
Segmentation of Images by using Fuzzy k-means clustering with ACOSegmentation of Images by using Fuzzy k-means clustering with ACO
Segmentation of Images by using Fuzzy k-means clustering with ACOIJTET Journal
 
Improvement of chaotic secure communication scheme based on steganographic me...
Improvement of chaotic secure communication scheme based on steganographic me...Improvement of chaotic secure communication scheme based on steganographic me...
Improvement of chaotic secure communication scheme based on steganographic me...Massoud Khodadadzadeh
 
Cerebellar Model Articulation Controller
Cerebellar Model Articulation ControllerCerebellar Model Articulation Controller
Cerebellar Model Articulation ControllerZahra Sadeghi
 
Electricity Demand Forecasting Using Fuzzy-Neural Network
Electricity Demand Forecasting Using Fuzzy-Neural NetworkElectricity Demand Forecasting Using Fuzzy-Neural Network
Electricity Demand Forecasting Using Fuzzy-Neural NetworkNaren Chandra Kattla
 
Caravan insurance data mining prediction models
Caravan insurance data mining prediction modelsCaravan insurance data mining prediction models
Caravan insurance data mining prediction modelsMuthu Kumaar Thangavelu
 
Reliability Prediction using the Fussel Algorithm
Reliability Prediction using the Fussel AlgorithmReliability Prediction using the Fussel Algorithm
Reliability Prediction using the Fussel AlgorithmIRJET Journal
 
11 construction productivity and cost estimation using artificial
11 construction productivity and cost estimation using artificial 11 construction productivity and cost estimation using artificial
11 construction productivity and cost estimation using artificial Vivan17
 
Electricity Demand Forecasting Using ANN
Electricity Demand Forecasting Using ANNElectricity Demand Forecasting Using ANN
Electricity Demand Forecasting Using ANNNaren Chandra Kattla
 
Traffic sign classification
Traffic sign classificationTraffic sign classification
Traffic sign classificationBill Kromydas
 
IRJET- LS Chaotic based Image Encryption System Via Permutation Models
IRJET- LS Chaotic based Image Encryption System Via Permutation ModelsIRJET- LS Chaotic based Image Encryption System Via Permutation Models
IRJET- LS Chaotic based Image Encryption System Via Permutation ModelsIRJET Journal
 

What's hot (13)

Presentation on "Mizan: A System for Dynamic Load Balancing in Large-scale Gr...
Presentation on "Mizan: A System for Dynamic Load Balancing in Large-scale Gr...Presentation on "Mizan: A System for Dynamic Load Balancing in Large-scale Gr...
Presentation on "Mizan: A System for Dynamic Load Balancing in Large-scale Gr...
 
Mathematics 08-00326
Mathematics 08-00326Mathematics 08-00326
Mathematics 08-00326
 
Reactive/Proactive Connectivity Management in a Tactical Service-Oriented Inf...
Reactive/Proactive Connectivity Management in a Tactical Service-Oriented Inf...Reactive/Proactive Connectivity Management in a Tactical Service-Oriented Inf...
Reactive/Proactive Connectivity Management in a Tactical Service-Oriented Inf...
 
Segmentation of Images by using Fuzzy k-means clustering with ACO
Segmentation of Images by using Fuzzy k-means clustering with ACOSegmentation of Images by using Fuzzy k-means clustering with ACO
Segmentation of Images by using Fuzzy k-means clustering with ACO
 
Improvement of chaotic secure communication scheme based on steganographic me...
Improvement of chaotic secure communication scheme based on steganographic me...Improvement of chaotic secure communication scheme based on steganographic me...
Improvement of chaotic secure communication scheme based on steganographic me...
 
Cerebellar Model Articulation Controller
Cerebellar Model Articulation ControllerCerebellar Model Articulation Controller
Cerebellar Model Articulation Controller
 
Electricity Demand Forecasting Using Fuzzy-Neural Network
Electricity Demand Forecasting Using Fuzzy-Neural NetworkElectricity Demand Forecasting Using Fuzzy-Neural Network
Electricity Demand Forecasting Using Fuzzy-Neural Network
 
Caravan insurance data mining prediction models
Caravan insurance data mining prediction modelsCaravan insurance data mining prediction models
Caravan insurance data mining prediction models
 
Reliability Prediction using the Fussel Algorithm
Reliability Prediction using the Fussel AlgorithmReliability Prediction using the Fussel Algorithm
Reliability Prediction using the Fussel Algorithm
 
11 construction productivity and cost estimation using artificial
11 construction productivity and cost estimation using artificial 11 construction productivity and cost estimation using artificial
11 construction productivity and cost estimation using artificial
 
Electricity Demand Forecasting Using ANN
Electricity Demand Forecasting Using ANNElectricity Demand Forecasting Using ANN
Electricity Demand Forecasting Using ANN
 
Traffic sign classification
Traffic sign classificationTraffic sign classification
Traffic sign classification
 
IRJET- LS Chaotic based Image Encryption System Via Permutation Models
IRJET- LS Chaotic based Image Encryption System Via Permutation ModelsIRJET- LS Chaotic based Image Encryption System Via Permutation Models
IRJET- LS Chaotic based Image Encryption System Via Permutation Models
 

Similar to Representation learning

BACKPROPAGATION LEARNING ALGORITHM BASED ON LEVENBERG MARQUARDT ALGORITHM
BACKPROPAGATION LEARNING ALGORITHM BASED ON LEVENBERG MARQUARDT ALGORITHMBACKPROPAGATION LEARNING ALGORITHM BASED ON LEVENBERG MARQUARDT ALGORITHM
BACKPROPAGATION LEARNING ALGORITHM BASED ON LEVENBERG MARQUARDT ALGORITHMcscpconf
 
A COMPARATIVE STUDY OF BACKPROPAGATION ALGORITHMS IN FINANCIAL PREDICTION
A COMPARATIVE STUDY OF BACKPROPAGATION ALGORITHMS IN FINANCIAL PREDICTIONA COMPARATIVE STUDY OF BACKPROPAGATION ALGORITHMS IN FINANCIAL PREDICTION
A COMPARATIVE STUDY OF BACKPROPAGATION ALGORITHMS IN FINANCIAL PREDICTIONIJCSEA Journal
 
Neural Network Based Individual Classification System
Neural Network Based Individual Classification SystemNeural Network Based Individual Classification System
Neural Network Based Individual Classification SystemIRJET Journal
 
Neural Network Implementation Control Mobile Robot
Neural Network Implementation Control Mobile RobotNeural Network Implementation Control Mobile Robot
Neural Network Implementation Control Mobile RobotIRJET Journal
 
05 backpropagation automatic_differentiation
05 backpropagation automatic_differentiation05 backpropagation automatic_differentiation
05 backpropagation automatic_differentiationAndres Mendez-Vazquez
 
Electricity Price Forecasting Using ELM-Tree Approach
Electricity Price Forecasting Using ELM-Tree ApproachElectricity Price Forecasting Using ELM-Tree Approach
Electricity Price Forecasting Using ELM-Tree ApproachIRJET Journal
 
Introduction to deep learning
Introduction to deep learningIntroduction to deep learning
Introduction to deep learningJunaid Bhat
 
An Optimized Parallel Algorithm for Longest Common Subsequence Using Openmp –...
An Optimized Parallel Algorithm for Longest Common Subsequence Using Openmp –...An Optimized Parallel Algorithm for Longest Common Subsequence Using Openmp –...
An Optimized Parallel Algorithm for Longest Common Subsequence Using Openmp –...IRJET Journal
 
A Connectionist Approach to Dynamic Resource Management for Virtualised Netwo...
A Connectionist Approach to Dynamic Resource Management for Virtualised Netwo...A Connectionist Approach to Dynamic Resource Management for Virtualised Netwo...
A Connectionist Approach to Dynamic Resource Management for Virtualised Netwo...Rashid Mijumbi
 
Studies On Traffic Management Models for Wireless Communication Network
Studies On Traffic Management Models for Wireless Communication NetworkStudies On Traffic Management Models for Wireless Communication Network
Studies On Traffic Management Models for Wireless Communication NetworkNeetaSingh38
 
220206 transformer interpretability beyond attention visualization
220206 transformer interpretability beyond attention visualization220206 transformer interpretability beyond attention visualization
220206 transformer interpretability beyond attention visualizationtaeseon ryu
 
Manifold Blurring Mean Shift algorithms for manifold denoising, report, 2012
Manifold Blurring Mean Shift algorithms for manifold denoising, report, 2012Manifold Blurring Mean Shift algorithms for manifold denoising, report, 2012
Manifold Blurring Mean Shift algorithms for manifold denoising, report, 2012Florent Renucci
 
Implementation of recurrent neural network for the forecasting of USD buy ra...
Implementation of recurrent neural network for the forecasting  of USD buy ra...Implementation of recurrent neural network for the forecasting  of USD buy ra...
Implementation of recurrent neural network for the forecasting of USD buy ra...IJECEIAES
 
Analysis of Various Single Frame Super Resolution Techniques for better PSNR
Analysis of Various Single Frame Super Resolution Techniques for better PSNRAnalysis of Various Single Frame Super Resolution Techniques for better PSNR
Analysis of Various Single Frame Super Resolution Techniques for better PSNRIRJET Journal
 
Paper review: Learned Optimizers that Scale and Generalize.
Paper review: Learned Optimizers that Scale and Generalize.Paper review: Learned Optimizers that Scale and Generalize.
Paper review: Learned Optimizers that Scale and Generalize.Wuhyun Rico Shin
 
IRJET- RESULT:Wavelet Transform along with SPIHT Algorithm Used for Image Com...
IRJET- RESULT:Wavelet Transform along with SPIHT Algorithm Used for Image Com...IRJET- RESULT:Wavelet Transform along with SPIHT Algorithm Used for Image Com...
IRJET- RESULT:Wavelet Transform along with SPIHT Algorithm Used for Image Com...IRJET Journal
 
Neural Network-Based Actuator Fault Diagnosis for a Non-Linear Multi-Tank System
Neural Network-Based Actuator Fault Diagnosis for a Non-Linear Multi-Tank SystemNeural Network-Based Actuator Fault Diagnosis for a Non-Linear Multi-Tank System
Neural Network-Based Actuator Fault Diagnosis for a Non-Linear Multi-Tank SystemISA Interchange
 

Similar to Representation learning (20)

24csit38.pdf
24csit38.pdf24csit38.pdf
24csit38.pdf
 
BACKPROPAGATION LEARNING ALGORITHM BASED ON LEVENBERG MARQUARDT ALGORITHM
BACKPROPAGATION LEARNING ALGORITHM BASED ON LEVENBERG MARQUARDT ALGORITHMBACKPROPAGATION LEARNING ALGORITHM BASED ON LEVENBERG MARQUARDT ALGORITHM
BACKPROPAGATION LEARNING ALGORITHM BASED ON LEVENBERG MARQUARDT ALGORITHM
 
A COMPARATIVE STUDY OF BACKPROPAGATION ALGORITHMS IN FINANCIAL PREDICTION
A COMPARATIVE STUDY OF BACKPROPAGATION ALGORITHMS IN FINANCIAL PREDICTIONA COMPARATIVE STUDY OF BACKPROPAGATION ALGORITHMS IN FINANCIAL PREDICTION
A COMPARATIVE STUDY OF BACKPROPAGATION ALGORITHMS IN FINANCIAL PREDICTION
 
Neural Network Based Individual Classification System
Neural Network Based Individual Classification SystemNeural Network Based Individual Classification System
Neural Network Based Individual Classification System
 
Neural Network Implementation Control Mobile Robot
Neural Network Implementation Control Mobile RobotNeural Network Implementation Control Mobile Robot
Neural Network Implementation Control Mobile Robot
 
05 backpropagation automatic_differentiation
05 backpropagation automatic_differentiation05 backpropagation automatic_differentiation
05 backpropagation automatic_differentiation
 
Electricity Price Forecasting Using ELM-Tree Approach
Electricity Price Forecasting Using ELM-Tree ApproachElectricity Price Forecasting Using ELM-Tree Approach
Electricity Price Forecasting Using ELM-Tree Approach
 
Introduction to deep learning
Introduction to deep learningIntroduction to deep learning
Introduction to deep learning
 
An Optimized Parallel Algorithm for Longest Common Subsequence Using Openmp –...
An Optimized Parallel Algorithm for Longest Common Subsequence Using Openmp –...An Optimized Parallel Algorithm for Longest Common Subsequence Using Openmp –...
An Optimized Parallel Algorithm for Longest Common Subsequence Using Openmp –...
 
A Connectionist Approach to Dynamic Resource Management for Virtualised Netwo...
A Connectionist Approach to Dynamic Resource Management for Virtualised Netwo...A Connectionist Approach to Dynamic Resource Management for Virtualised Netwo...
A Connectionist Approach to Dynamic Resource Management for Virtualised Netwo...
 
Studies On Traffic Management Models for Wireless Communication Network
Studies On Traffic Management Models for Wireless Communication NetworkStudies On Traffic Management Models for Wireless Communication Network
Studies On Traffic Management Models for Wireless Communication Network
 
220206 transformer interpretability beyond attention visualization
220206 transformer interpretability beyond attention visualization220206 transformer interpretability beyond attention visualization
220206 transformer interpretability beyond attention visualization
 
Manifold Blurring Mean Shift algorithms for manifold denoising, report, 2012
Manifold Blurring Mean Shift algorithms for manifold denoising, report, 2012Manifold Blurring Mean Shift algorithms for manifold denoising, report, 2012
Manifold Blurring Mean Shift algorithms for manifold denoising, report, 2012
 
Implementation of recurrent neural network for the forecasting of USD buy ra...
Implementation of recurrent neural network for the forecasting  of USD buy ra...Implementation of recurrent neural network for the forecasting  of USD buy ra...
Implementation of recurrent neural network for the forecasting of USD buy ra...
 
Analysis of Various Single Frame Super Resolution Techniques for better PSNR
Analysis of Various Single Frame Super Resolution Techniques for better PSNRAnalysis of Various Single Frame Super Resolution Techniques for better PSNR
Analysis of Various Single Frame Super Resolution Techniques for better PSNR
 
Paper review: Learned Optimizers that Scale and Generalize.
Paper review: Learned Optimizers that Scale and Generalize.Paper review: Learned Optimizers that Scale and Generalize.
Paper review: Learned Optimizers that Scale and Generalize.
 
IRJET- RESULT:Wavelet Transform along with SPIHT Algorithm Used for Image Com...
IRJET- RESULT:Wavelet Transform along with SPIHT Algorithm Used for Image Com...IRJET- RESULT:Wavelet Transform along with SPIHT Algorithm Used for Image Com...
IRJET- RESULT:Wavelet Transform along with SPIHT Algorithm Used for Image Com...
 
I05745368
I05745368I05745368
I05745368
 
Neural Network-Based Actuator Fault Diagnosis for a Non-Linear Multi-Tank System
Neural Network-Based Actuator Fault Diagnosis for a Non-Linear Multi-Tank SystemNeural Network-Based Actuator Fault Diagnosis for a Non-Linear Multi-Tank System
Neural Network-Based Actuator Fault Diagnosis for a Non-Linear Multi-Tank System
 
Jf3515881595
Jf3515881595Jf3515881595
Jf3515881595
 

More from Brian Kim

20190708 bumsookim yolact
20190708 bumsookim yolact20190708 bumsookim yolact
20190708 bumsookim yolactBrian Kim
 
20190718 bumsookim 2_attention
20190718 bumsookim 2_attention20190718 bumsookim 2_attention
20190718 bumsookim 2_attentionBrian Kim
 
Spectral cnn
Spectral cnnSpectral cnn
Spectral cnnBrian Kim
 
Compressing neural language models by sparse word representation
Compressing neural language models by sparse word representationCompressing neural language models by sparse word representation
Compressing neural language models by sparse word representationBrian Kim
 

More from Brian Kim (7)

FreeAnchor
FreeAnchorFreeAnchor
FreeAnchor
 
20190708 bumsookim yolact
20190708 bumsookim yolact20190708 bumsookim yolact
20190708 bumsookim yolact
 
20190718 bumsookim 2_attention
20190718 bumsookim 2_attention20190718 bumsookim 2_attention
20190718 bumsookim 2_attention
 
Spectral cnn
Spectral cnnSpectral cnn
Spectral cnn
 
Compressing neural language models by sparse word representation
Compressing neural language models by sparse word representationCompressing neural language models by sparse word representation
Compressing neural language models by sparse word representation
 
Dcgan
DcganDcgan
Dcgan
 
Google net
Google netGoogle net
Google net
 

Recently uploaded

Just Call Vip call girls kakinada Escorts ☎️9352988975 Two shot with one girl...
Just Call Vip call girls kakinada Escorts ☎️9352988975 Two shot with one girl...Just Call Vip call girls kakinada Escorts ☎️9352988975 Two shot with one girl...
Just Call Vip call girls kakinada Escorts ☎️9352988975 Two shot with one girl...gajnagarg
 
Call Girls Jalahalli Just Call 👗 7737669865 👗 Top Class Call Girl Service Ban...
Call Girls Jalahalli Just Call 👗 7737669865 👗 Top Class Call Girl Service Ban...Call Girls Jalahalli Just Call 👗 7737669865 👗 Top Class Call Girl Service Ban...
Call Girls Jalahalli Just Call 👗 7737669865 👗 Top Class Call Girl Service Ban...amitlee9823
 
Call Girls In Doddaballapur Road ☎ 7737669865 🥵 Book Your One night Stand
Call Girls In Doddaballapur Road ☎ 7737669865 🥵 Book Your One night StandCall Girls In Doddaballapur Road ☎ 7737669865 🥵 Book Your One night Stand
Call Girls In Doddaballapur Road ☎ 7737669865 🥵 Book Your One night Standamitlee9823
 
Call Girls Begur Just Call 👗 7737669865 👗 Top Class Call Girl Service Bangalore
Call Girls Begur Just Call 👗 7737669865 👗 Top Class Call Girl Service BangaloreCall Girls Begur Just Call 👗 7737669865 👗 Top Class Call Girl Service Bangalore
Call Girls Begur Just Call 👗 7737669865 👗 Top Class Call Girl Service Bangaloreamitlee9823
 
Call Girls In Bellandur ☎ 7737669865 🥵 Book Your One night Stand
Call Girls In Bellandur ☎ 7737669865 🥵 Book Your One night StandCall Girls In Bellandur ☎ 7737669865 🥵 Book Your One night Stand
Call Girls In Bellandur ☎ 7737669865 🥵 Book Your One night Standamitlee9823
 
➥🔝 7737669865 🔝▻ mahisagar Call-girls in Women Seeking Men 🔝mahisagar🔝 Esc...
➥🔝 7737669865 🔝▻ mahisagar Call-girls in Women Seeking Men  🔝mahisagar🔝   Esc...➥🔝 7737669865 🔝▻ mahisagar Call-girls in Women Seeking Men  🔝mahisagar🔝   Esc...
➥🔝 7737669865 🔝▻ mahisagar Call-girls in Women Seeking Men 🔝mahisagar🔝 Esc...amitlee9823
 
Thane Call Girls 7091864438 Call Girls in Thane Escort service book now -
Thane Call Girls 7091864438 Call Girls in Thane Escort service book now -Thane Call Girls 7091864438 Call Girls in Thane Escort service book now -
Thane Call Girls 7091864438 Call Girls in Thane Escort service book now -Pooja Nehwal
 
Call Girls Hsr Layout Just Call 👗 7737669865 👗 Top Class Call Girl Service Ba...
Call Girls Hsr Layout Just Call 👗 7737669865 👗 Top Class Call Girl Service Ba...Call Girls Hsr Layout Just Call 👗 7737669865 👗 Top Class Call Girl Service Ba...
Call Girls Hsr Layout Just Call 👗 7737669865 👗 Top Class Call Girl Service Ba...amitlee9823
 
Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...
Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...
Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...amitlee9823
 
Just Call Vip call girls Mysore Escorts ☎️9352988975 Two shot with one girl (...
Just Call Vip call girls Mysore Escorts ☎️9352988975 Two shot with one girl (...Just Call Vip call girls Mysore Escorts ☎️9352988975 Two shot with one girl (...
Just Call Vip call girls Mysore Escorts ☎️9352988975 Two shot with one girl (...gajnagarg
 
Aspirational Block Program Block Syaldey District - Almora
Aspirational Block Program Block Syaldey District - AlmoraAspirational Block Program Block Syaldey District - Almora
Aspirational Block Program Block Syaldey District - AlmoraGovindSinghDasila
 
Call Girls In Shivaji Nagar ☎ 7737669865 🥵 Book Your One night Stand
Call Girls In Shivaji Nagar ☎ 7737669865 🥵 Book Your One night StandCall Girls In Shivaji Nagar ☎ 7737669865 🥵 Book Your One night Stand
Call Girls In Shivaji Nagar ☎ 7737669865 🥵 Book Your One night Standamitlee9823
 
SAC 25 Final National, Regional & Local Angel Group Investing Insights 2024 0...
SAC 25 Final National, Regional & Local Angel Group Investing Insights 2024 0...SAC 25 Final National, Regional & Local Angel Group Investing Insights 2024 0...
SAC 25 Final National, Regional & Local Angel Group Investing Insights 2024 0...Elaine Werffeli
 
VIP Model Call Girls Hinjewadi ( Pune ) Call ON 8005736733 Starting From 5K t...
VIP Model Call Girls Hinjewadi ( Pune ) Call ON 8005736733 Starting From 5K t...VIP Model Call Girls Hinjewadi ( Pune ) Call ON 8005736733 Starting From 5K t...
VIP Model Call Girls Hinjewadi ( Pune ) Call ON 8005736733 Starting From 5K t...SUHANI PANDEY
 
➥🔝 7737669865 🔝▻ Bangalore Call-girls in Women Seeking Men 🔝Bangalore🔝 Esc...
➥🔝 7737669865 🔝▻ Bangalore Call-girls in Women Seeking Men  🔝Bangalore🔝   Esc...➥🔝 7737669865 🔝▻ Bangalore Call-girls in Women Seeking Men  🔝Bangalore🔝   Esc...
➥🔝 7737669865 🔝▻ Bangalore Call-girls in Women Seeking Men 🔝Bangalore🔝 Esc...amitlee9823
 
Call Girls In Hsr Layout ☎ 7737669865 🥵 Book Your One night Stand
Call Girls In Hsr Layout ☎ 7737669865 🥵 Book Your One night StandCall Girls In Hsr Layout ☎ 7737669865 🥵 Book Your One night Stand
Call Girls In Hsr Layout ☎ 7737669865 🥵 Book Your One night Standamitlee9823
 

Recently uploaded (20)

Anomaly detection and data imputation within time series
Anomaly detection and data imputation within time seriesAnomaly detection and data imputation within time series
Anomaly detection and data imputation within time series
 
Just Call Vip call girls kakinada Escorts ☎️9352988975 Two shot with one girl...
Just Call Vip call girls kakinada Escorts ☎️9352988975 Two shot with one girl...Just Call Vip call girls kakinada Escorts ☎️9352988975 Two shot with one girl...
Just Call Vip call girls kakinada Escorts ☎️9352988975 Two shot with one girl...
 
Call Girls Jalahalli Just Call 👗 7737669865 👗 Top Class Call Girl Service Ban...
Call Girls Jalahalli Just Call 👗 7737669865 👗 Top Class Call Girl Service Ban...Call Girls Jalahalli Just Call 👗 7737669865 👗 Top Class Call Girl Service Ban...
Call Girls Jalahalli Just Call 👗 7737669865 👗 Top Class Call Girl Service Ban...
 
Call Girls In Doddaballapur Road ☎ 7737669865 🥵 Book Your One night Stand
Call Girls In Doddaballapur Road ☎ 7737669865 🥵 Book Your One night StandCall Girls In Doddaballapur Road ☎ 7737669865 🥵 Book Your One night Stand
Call Girls In Doddaballapur Road ☎ 7737669865 🥵 Book Your One night Stand
 
Call Girls Begur Just Call 👗 7737669865 👗 Top Class Call Girl Service Bangalore
Call Girls Begur Just Call 👗 7737669865 👗 Top Class Call Girl Service BangaloreCall Girls Begur Just Call 👗 7737669865 👗 Top Class Call Girl Service Bangalore
Call Girls Begur Just Call 👗 7737669865 👗 Top Class Call Girl Service Bangalore
 
Call Girls In Bellandur ☎ 7737669865 🥵 Book Your One night Stand
Call Girls In Bellandur ☎ 7737669865 🥵 Book Your One night StandCall Girls In Bellandur ☎ 7737669865 🥵 Book Your One night Stand
Call Girls In Bellandur ☎ 7737669865 🥵 Book Your One night Stand
 
(NEHA) Call Girls Katra Call Now 8617697112 Katra Escorts 24x7
(NEHA) Call Girls Katra Call Now 8617697112 Katra Escorts 24x7(NEHA) Call Girls Katra Call Now 8617697112 Katra Escorts 24x7
(NEHA) Call Girls Katra Call Now 8617697112 Katra Escorts 24x7
 
➥🔝 7737669865 🔝▻ mahisagar Call-girls in Women Seeking Men 🔝mahisagar🔝 Esc...
➥🔝 7737669865 🔝▻ mahisagar Call-girls in Women Seeking Men  🔝mahisagar🔝   Esc...➥🔝 7737669865 🔝▻ mahisagar Call-girls in Women Seeking Men  🔝mahisagar🔝   Esc...
➥🔝 7737669865 🔝▻ mahisagar Call-girls in Women Seeking Men 🔝mahisagar🔝 Esc...
 
Thane Call Girls 7091864438 Call Girls in Thane Escort service book now -
Thane Call Girls 7091864438 Call Girls in Thane Escort service book now -Thane Call Girls 7091864438 Call Girls in Thane Escort service book now -
Thane Call Girls 7091864438 Call Girls in Thane Escort service book now -
 
Call Girls Hsr Layout Just Call 👗 7737669865 👗 Top Class Call Girl Service Ba...
Call Girls Hsr Layout Just Call 👗 7737669865 👗 Top Class Call Girl Service Ba...Call Girls Hsr Layout Just Call 👗 7737669865 👗 Top Class Call Girl Service Ba...
Call Girls Hsr Layout Just Call 👗 7737669865 👗 Top Class Call Girl Service Ba...
 
Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...
Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...
Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...
 
Just Call Vip call girls Mysore Escorts ☎️9352988975 Two shot with one girl (...
Just Call Vip call girls Mysore Escorts ☎️9352988975 Two shot with one girl (...Just Call Vip call girls Mysore Escorts ☎️9352988975 Two shot with one girl (...
Just Call Vip call girls Mysore Escorts ☎️9352988975 Two shot with one girl (...
 
Aspirational Block Program Block Syaldey District - Almora
Aspirational Block Program Block Syaldey District - AlmoraAspirational Block Program Block Syaldey District - Almora
Aspirational Block Program Block Syaldey District - Almora
 
Call Girls In Shalimar Bagh ( Delhi) 9953330565 Escorts Service
Call Girls In Shalimar Bagh ( Delhi) 9953330565 Escorts ServiceCall Girls In Shalimar Bagh ( Delhi) 9953330565 Escorts Service
Call Girls In Shalimar Bagh ( Delhi) 9953330565 Escorts Service
 
Call Girls In Shivaji Nagar ☎ 7737669865 🥵 Book Your One night Stand
Call Girls In Shivaji Nagar ☎ 7737669865 🥵 Book Your One night StandCall Girls In Shivaji Nagar ☎ 7737669865 🥵 Book Your One night Stand
Call Girls In Shivaji Nagar ☎ 7737669865 🥵 Book Your One night Stand
 
SAC 25 Final National, Regional & Local Angel Group Investing Insights 2024 0...
SAC 25 Final National, Regional & Local Angel Group Investing Insights 2024 0...SAC 25 Final National, Regional & Local Angel Group Investing Insights 2024 0...
SAC 25 Final National, Regional & Local Angel Group Investing Insights 2024 0...
 
VIP Model Call Girls Hinjewadi ( Pune ) Call ON 8005736733 Starting From 5K t...
VIP Model Call Girls Hinjewadi ( Pune ) Call ON 8005736733 Starting From 5K t...VIP Model Call Girls Hinjewadi ( Pune ) Call ON 8005736733 Starting From 5K t...
VIP Model Call Girls Hinjewadi ( Pune ) Call ON 8005736733 Starting From 5K t...
 
➥🔝 7737669865 🔝▻ Bangalore Call-girls in Women Seeking Men 🔝Bangalore🔝 Esc...
➥🔝 7737669865 🔝▻ Bangalore Call-girls in Women Seeking Men  🔝Bangalore🔝   Esc...➥🔝 7737669865 🔝▻ Bangalore Call-girls in Women Seeking Men  🔝Bangalore🔝   Esc...
➥🔝 7737669865 🔝▻ Bangalore Call-girls in Women Seeking Men 🔝Bangalore🔝 Esc...
 
Abortion pills in Doha Qatar (+966572737505 ! Get Cytotec
Abortion pills in Doha Qatar (+966572737505 ! Get CytotecAbortion pills in Doha Qatar (+966572737505 ! Get Cytotec
Abortion pills in Doha Qatar (+966572737505 ! Get Cytotec
 
Call Girls In Hsr Layout ☎ 7737669865 🥵 Book Your One night Stand
Call Girls In Hsr Layout ☎ 7737669865 🥵 Book Your One night StandCall Girls In Hsr Layout ☎ 7737669865 🥵 Book Your One night Stand
Call Girls In Hsr Layout ☎ 7737669865 🥵 Book Your One night Stand
 

Representation learning

  • 1. Final Report: Construction through Deep RNN Representation Learning of Knowledge Bases KoreaUniversity, DepartmentofComputerScience&Radio CommunicationEngineering MASSIVEDATAMANAGEMENT Professor JaewooKang 1 2015010661 2015011155 2016010646 MinhwanYu YonghwaChoi BumsooKim
  • 2. MASSIVE DATA MANAGEMENT Presentation format Contents 01.Introduction 02.PTransE 1-1.Whatdowewant? 1-3.TransH 2-1.Whatisdifferent? 2-2.RelationPathRepresentation 2-3.RecurrentNeuralNetwork 2 03.Ouralgorithm 3-1.Objectfunction 3-2.Activationfunction 3-3.LongShortTermMemory 04.Evaluation 4-1.Results 4-2.Comparison&Analysis 1-4.TransR 1-5.PTransE 1-2.TransE
  • 3. MASSIVE DATA MANAGEMENT Presentation format Introduction 1-2.TransE 1-3.TransH 3 1-4.TransR 1-5.PTransE 1-1.Whatdowewant?
  • 4. MASSIVE DATA MANAGEMENT Presentation format 1. Introduction 4 1-1. What do we want? Large-Scale KnowledgeBases Freebase DBpedia Yago Realworld triples KnowledgeBase(KB)
  • 5. MASSIVE DATA MANAGEMENT Presentation format 1. Introduction 5 1-1. What do we want? Large-Scale KnowledgeBases Freebase Realworld triples ∴KBsareveryincomplete Misrelation 1.Norelationfound 2.Wrongrelationfound DBpedia Yago
  • 6. MASSIVE DATA MANAGEMENT Presentation format 1. Introduction 6 1-1. What do we want? How doweaddress mis-relations?
  • 7. MASSIVE DATA MANAGEMENT Presentation format 1. Introduction 7 1-1. What do we want? Q.Howdoweaddressmisrelation? A1. Addexternalsourcesfor completion Manualandintuitional Requirestoomuchlabor Inefficient,time-consuming PerformancedropsasKBgetslarger
  • 8. MASSIVE DATA MANAGEMENT Presentation format 1. Introduction 8 1-1. What do we want? Q.Howdoweaddressmisrelation? A1. Addexternalsourcesfor completion A2. Referencingothertuples forcompletion Powerful&Efficient Largelyexpandable MorepowerfulasKBgetslarger
  • 9. MASSIVE DATA MANAGEMENT Presentation format 1. Introduction 9 1-1. What do we want? Wetriedout… TransE TransH TransR PTransE CompletionofKBbyaddressingmisrelationsthrough co-entity,co-relationreference
  • 10. MASSIVE DATA MANAGEMENT Presentation format 1. Introduction 10 1-2. TransE relation#1Boy Girl relation#1Queen TransE Representingdirectrelationswith differencesbetweenentityvectors Simple&Effective reference:TransE(Bordesetal.NIPS2013)
  • 11. MASSIVE DATA MANAGEMENT Presentation format 1. Introduction 11 1-2. TransE TransE Simple&Effective relation#1Boy Girl relation#1Queen King Representingdirectrelationswith differencesbetweenentityvectors reference:TransE(Bordesetal.NIPS2013)
  • 12. MASSIVE DATA MANAGEMENT Presentation format 1. Introduction 12 1-2. TransE However… TransEcan’trepresentmorethan onerelationshipbetweenentities. Inrealworld,weconstructmany relationshipswithmanysubjects.
  • 13. MASSIVE DATA MANAGEMENT Presentation format 1. Introduction 13 1-3. TransH TransH AbletorepresentM-to-Mrelations Representingprojectedrelationswith differencesbetweenentityvectors Entitiesandrelationshavedifferentcharacteristics However,theyarerepresentedinthesamespace reference:TransH(Wangetal.AAAI2014)
  • 14. MASSIVE DATA MANAGEMENT Presentation format 1. Introduction 14 1-4. TransR Abletorepresentdifferentcharacteristics betweenentitiesandrelations Representingprojectedentitiesand mappingthemintorelationspace‘r’ TransR However,thesewereallnotgoodenoughfor detectingandaddressingmisrelations!! reference:TransR(AAAI2015)
  • 15. MASSIVE DATA MANAGEMENT Presentation format 1. Introduction 15 1-5. PTransE PTransE Representingrelationsthroughcompositionof relationsbetweenentityvectors Widelyexpandable&powerfulmethod h t e1 r1 r2 r=r1⋅r2 r1:father r1:father reference:emnlpprocessing
  • 16. MASSIVE DATA MANAGEMENT Presentation format 1. Introduction 16 1-5. PTransE PTransE Representingrelationsthroughcompositionof relationsbetweenentityvectors Widelyexpandable&powerfulmethod h t e1 r1 r2 r=r1⋅r2 r:grandfather reference:emnlpprocessing
  • 17. MASSIVE DATA MANAGEMENT Presentation format PTransE 2-2.RelationPathRepresentation 2-3.RecurrentNeuralNetwork 17 2-1.Whatisdifferent?
  • 18. MASSIVE DATA MANAGEMENT Presentation format 2. PTransE 18 2-1. What is different? PTransE LearnsonlyonesinglematrixEachrelationneedsitsown projectionmap. TransR 50,000relationsneeds50,000 enormousprojectionmaps! Learnsn-steprelationalpaths Learnsrelationsthatevendon’t haveenoughtrainingdataIfarelationdoesn’thaveenough trainingdata,itwillsufferfromlow performance. Learnsrelationsonlyinthetraining dataset. Learnsrelationsthatarenotinthe trainingdatasetthroughrelation paths.(Zero-shotKBinference)
  • 19. MASSIVE DATA MANAGEMENT Presentation format 2. PTransE 19 2-2. Relation Path Representation MicrosoftisbasedinSeattle. WhichcountryisMicrosoftlocatedin? Microsoft Seattle IsBasedInCountryIn ??? RelationNOTFOUND!(Misrelation)
  • 20. MASSIVE DATA MANAGEMENT Presentation format 2. PTransE 20 2-2. Relation Path Representation ‘Microsoft’isbasedin‘Seattle’. ‘Seattle’islocatedinstate‘Washington’ Microsoft Seattle Washington USA IsBasedIn ‘Washington’islocatedincountry‘USA’ StateIn CountryIn
  • 21. MASSIVE DATA MANAGEMENT Presentation format 2. PTransE 21 2-2. Relation Path Representation ‘Microsoft’isbasedin‘Seattle’. ‘Seattle’islocatedinstate‘Washington’ Microsoft USA ‘Washington’islocatedincountry‘USA’ CountryLocatedIn ‘Microsoft’islocatedin‘USA’.
  • 22. MASSIVE DATA MANAGEMENT Presentation format 2. PTransE 22 2-3. Recurrent Neural Network Microsoft Seattle Washington IsBasedIn StateIn CountryIn USA CountryLocatedIn RNN RNN Learningamatrix 𝑾𝑾𝒓𝒓 for constructingcomposition vector StateLocatedIn p= 𝒇𝒇( 𝑾𝑾𝒓𝒓 𝒄𝒄𝟑𝟑;𝒄𝒄𝟒𝟒 ) = 𝒕𝒕𝒕𝒕𝒕𝒕𝒕𝒕( 𝑾𝑾𝒓𝒓 𝒄𝒄𝟑𝟑;𝒄𝒄𝟒𝟒 ) 𝐜𝐜𝟑𝟑 = 𝒇𝒇( 𝑾𝑾𝒓𝒓 𝒄𝒄𝟏𝟏;𝒄𝒄𝟐𝟐 ) 𝒄𝒄𝟏𝟏 𝒄𝒄𝟐𝟐 𝒄𝒄𝟒𝟒 reference:CompositionalVectorSpaceModelsforKnowledgeBaseInference
  • 23. MASSIVE DATA MANAGEMENT Presentation format Our Algorithm 3-2.Activationfunction 3-3.LongShortTermMemory 23 3-1.Objectivefunction
  • 24. MASSIVE DATA MANAGEMENT Presentation format 3. Our Algorithm 24 3-1. Objective function Me Father Grandfather FatherOf FatherOf RNN Compositionvector 𝐩𝐩 = 𝒇𝒇( 𝑾𝑾𝒓𝒓 𝒄𝒄𝟏𝟏;𝒄𝒄𝟐𝟐 ) 𝒄𝒄𝟏𝟏 𝒄𝒄𝟐𝟐 Me Grandfather GrandfatherOf Update 𝑾𝑾𝒓𝒓
  • 25. MASSIVE DATA MANAGEMENT Presentation format 3. Our Algorithm 25 3-1. Objective function Compositionvector 𝐩𝐩 = 𝒇𝒇( 𝑾𝑾𝒓𝒓 𝒄𝒄𝟏𝟏;𝒄𝒄𝟐𝟐 )GrandfatherOf n-steprelationvector 𝒍𝒍𝒍𝒍𝒍𝒍𝒍𝒍 Updating 𝑾𝑾𝒓𝒓 … 𝒍𝒍𝒍𝒍𝒍𝒍𝒍𝒍BackPropagation ForwardPropagation 𝒑𝒑𝒑𝒑𝒑𝒑𝒑𝒑 −𝒕𝒕𝒕𝒕𝒕𝒕𝒕𝒕𝒕𝒕𝒕𝒕𝑼𝑼𝑼𝑼𝑼𝑼𝑼𝑼𝑼𝑼𝑼𝑼 𝐖𝐖𝐫𝐫
  • 26. MASSIVE DATA MANAGEMENT Presentation format 3. Our Algorithm 26 3-2. Activation function However,NeuralNetworkisbasicallyalinearoperation 𝐩𝐩 = 𝒇𝒇( 𝑾𝑾𝒓𝒓 𝒄𝒄𝟏𝟏;𝒄𝒄𝟐𝟐 ) Applyinganon-linearaftereach operationsinordertolearnnon- lineardecisionboundary tanhistheconventionaldefault non-linearactivationfunctionfor RNNmodels.
  • 27. MASSIVE DATA MANAGEMENT Presentation format 3. Our Algorithm 27 3-2. Activation function Then,what’swrongwithtanh? Whenbackpropagatinggradients inordertoupdatethematrix, thesegradientstendtoconvergeto 0andeliminatesthegradientflow. Therefore,nogradientflows backward,andtheparameterstays unmodified. GradientVanishingProblem
  • 28. MASSIVE DATA MANAGEMENT Presentation format 3. Our Algorithm 28 3-3. Long Short Term Memory AddressesGradientVanishingproblems HandleLongtermdependencies ForgetGate
  • 29. MASSIVE DATA MANAGEMENT Presentation format Evaluation 4-2.Comparison&Analysis 29 4-1.Results
  • 30. MASSIVE DATA MANAGEMENT Presentation format 4. Evaluation 30 4-1. Results WeusedFB15K,andFB40Kdatasets FB15K FB40K # Relations 1,345 1,336 # Entities 14,951 39,528 # Training set 483,142 370,648 # Validation set 50,000 67,948 # Test set 59,071 96,678
  • 31. MASSIVE DATA MANAGEMENT Presentation format 4. Evaluation 31 4-2. Comparisons & Analysis Hit 10 (RAW) Hit 10 (FILTER) TransE 34.9% 47.1% TransH 45.7% 64.4% TransR 43.8% 65.5% Hit 10 (RAW) Hit 10 (FILTER) PTransE(RNN) (Original model) 50.6% 82.2% PTransE(LSTM) (Our model) 53.1% 86.6% PTransE,RNNcomposition Baseline
  • 32. MASSIVE DATA MANAGEMENT Presentation format 4. Evaluation 32 4-2. Comparisons & Analysis Hit 10 (RAW) Hit 10 (FILTER) TransE 34.9% 47.1% TransH 45.7% 64.4% TransR 43.8% 65.5% Hit 10 (RAW) Hit 10 (FILTER) PTransE(RNN) (Original model) 50.6% 82.2% PTransE(LSTM) (Our model) 53.1% 86.6% PTransE,RNNcomposition (+9.3%) Baseline
  • 33. MASSIVE DATA MANAGEMENT Presentation format 4. Evaluation 33 4-2. Comparisons & Analysis Hit 10 (RAW) Hit 10 (FILTER) TransE 34.9% 47.1% TransH 45.7% 64.4% TransR 43.8% 65.5% Hit 10 (RAW) Hit 10 (FILTER) PTransE(RNN) (Original model) 50.6% 82.2% PTransE(LSTM) (Our model) 53.1% 86.6% PTransE,RNNcomposition (+21.1%) Baseline
  • 34. MASSIVE DATA MANAGEMENT Presentation format 4. Evaluation 34 4-2. Comparisons & Analysis Hit 10 (RAW) Hit 10 (FILTER) PTransE(ADD) (Original model) 51.8% 83.4% PTransE(ADD) (Our model) 52.1% 84.1% PTransE(MUL) (Original model) 47.4% 77.7% PTransE(MUL) (Our model) 47.1% 77.2% PTransE(RNN) (Original model) 50.6% 82.2% PTransE(LSTM) (Our model) 53.1% 86.6% PTransE,Compositioncomparison
  • 35. Thank you for your attention! 35