View
4
Download
0
Category
Preview:
Citation preview
1/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Aprendizado de Máquina e Avaliação deDesempenho:
uma dupla dinâmica engajada da teoria àprática
Edmundo de Souza e Silva1
Universidade Federal do Rio de Janeiro1Programa de Engenharia de Sistemas e Computação, COPPE
2015
E. de Souza e Silva Semana PESC
2/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Performance Evaluation and Machine Learnig: TheDynamic Duo
E. de Souza e Silva Semana PESC
3/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Preliminaries
What is Computer System Modeling and Analysis?What is Machine Learning?Can we take advantage of both areas?What problems we address?Is this useful in practice?
E. de Souza e Silva Semana PESC
3/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Preliminaries
What is Computer System Modeling and Analysis?What is Machine Learning?Can we take advantage of both areas?What problems we address?Is this useful in practice?
E. de Souza e Silva Semana PESC
3/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Preliminaries
What is Computer System Modeling and Analysis?What is Machine Learning?Can we take advantage of both areas?What problems we address?Is this useful in practice?
E. de Souza e Silva Semana PESC
3/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Preliminaries
What is Computer System Modeling and Analysis?What is Machine Learning?Can we take advantage of both areas?What problems we address?Is this useful in practice?
E. de Souza e Silva Semana PESC
3/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Preliminaries
What is Computer System Modeling and Analysis?What is Machine Learning?Can we take advantage of both areas?What problems we address?Is this useful in practice?
E. de Souza e Silva Semana PESC
4/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Performance Evaluation
E. de Souza e Silva Semana PESC
5/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Performance Evaluation
Modeling and analysis is an essential ingredient of thedesign process of most systemsDevising new systems: generally needs analysis of itsperformance
What are the advantages of the new architecture?which scheduling policies to use?what speed to operate servers?On what conditions can the system efficiently operate
We want to predict behavior ofan algorithma protocola new computer architecturecustomers accessing some system...
We want to perform tradeoff analysis
E. de Souza e Silva Semana PESC
5/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Performance Evaluation
Modeling and analysis is an essential ingredient of thedesign process of most systemsDevising new systems: generally needs analysis of itsperformance
What are the advantages of the new architecture?which scheduling policies to use?what speed to operate servers?On what conditions can the system efficiently operate
We want to predict behavior ofan algorithma protocola new computer architecturecustomers accessing some system...
We want to perform tradeoff analysis
E. de Souza e Silva Semana PESC
5/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Performance Evaluation
Modeling and analysis is an essential ingredient of thedesign process of most systemsDevising new systems: generally needs analysis of itsperformance
What are the advantages of the new architecture?which scheduling policies to use?what speed to operate servers?On what conditions can the system efficiently operate
We want to predict behavior ofan algorithma protocola new computer architecturecustomers accessing some system...
We want to perform tradeoff analysis
E. de Souza e Silva Semana PESC
5/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Performance Evaluation
Modeling and analysis is an essential ingredient of thedesign process of most systemsDevising new systems: generally needs analysis of itsperformance
What are the advantages of the new architecture?which scheduling policies to use?what speed to operate servers?On what conditions can the system efficiently operate
We want to predict behavior ofan algorithma protocola new computer architecturecustomers accessing some system...
We want to perform tradeoff analysis
E. de Souza e Silva Semana PESC
6/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Examples
including research work ofour group
E. de Souza e Silva Semana PESC
7/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Disk clusters
Assume that the MTBF of a single disk is 300.000 hours.Probabilidade that a failure occurs in one disk unit duringone hour = 3.33310−6. (in one month: 0.0023942)Question: if you have 1000 disks, what is the probabilitythat one disk fails in one month?Answer: 0.90902Question: if you have 100.000 disks, what is the probabilitythat one disk fails in one hour? (common in disc clusters)Answer: 0.99966 −→ you WILL have a disk failedsomewhere in your cluster!!!
E. de Souza e Silva Semana PESC
7/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Disk clusters
Assume that the MTBF of a single disk is 300.000 hours.Probabilidade that a failure occurs in one disk unit duringone hour = 3.33310−6. (in one month: 0.0023942)Question: if you have 1000 disks, what is the probabilitythat one disk fails in one month?Answer: 0.90902Question: if you have 100.000 disks, what is the probabilitythat one disk fails in one hour? (common in disc clusters)Answer: 0.99966 −→ you WILL have a disk failedsomewhere in your cluster!!!
E. de Souza e Silva Semana PESC
7/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Disk clusters
Assume that the MTBF of a single disk is 300.000 hours.Probabilidade that a failure occurs in one disk unit duringone hour = 3.33310−6. (in one month: 0.0023942)Question: if you have 1000 disks, what is the probabilitythat one disk fails in one month?Answer: 0.90902Question: if you have 100.000 disks, what is the probabilitythat one disk fails in one hour? (common in disc clusters)Answer: 0.99966 −→ you WILL have a disk failedsomewhere in your cluster!!!
E. de Souza e Silva Semana PESC
7/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Disk clusters
Assume that the MTBF of a single disk is 300.000 hours.Probabilidade that a failure occurs in one disk unit duringone hour = 3.33310−6. (in one month: 0.0023942)Question: if you have 1000 disks, what is the probabilitythat one disk fails in one month?Answer: 0.90902Question: if you have 100.000 disks, what is the probabilitythat one disk fails in one hour? (common in disc clusters)Answer: 0.99966 −→ you WILL have a disk failedsomewhere in your cluster!!!
E. de Souza e Silva Semana PESC
7/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Disk clusters
Assume that the MTBF of a single disk is 300.000 hours.Probabilidade that a failure occurs in one disk unit duringone hour = 3.33310−6. (in one month: 0.0023942)Question: if you have 1000 disks, what is the probabilitythat one disk fails in one month?Answer: 0.90902Question: if you have 100.000 disks, what is the probabilitythat one disk fails in one hour? (common in disc clusters)Answer: 0.99966 −→ you WILL have a disk failedsomewhere in your cluster!!!
E. de Souza e Silva Semana PESC
8/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Disk clusters in Data Centers
Data centers consume a lot of powerQuestion: How can we reduce power consumption?Note: not all data in a large data center is accessedsimultaneously→ disks are not used all the time.Question: can we reduce power consumption by puttingsome disks to sleep?Answer: YES, but there are tradeoffs to investigate.What are the tradeoffs?
E. de Souza e Silva Semana PESC
8/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Disk clusters in Data Centers
Data centers consume a lot of powerQuestion: How can we reduce power consumption?Note: not all data in a large data center is accessedsimultaneously→ disks are not used all the time.Question: can we reduce power consumption by puttingsome disks to sleep?Answer: YES, but there are tradeoffs to investigate.What are the tradeoffs?
E. de Souza e Silva Semana PESC
8/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Disk clusters in Data Centers
Data centers consume a lot of powerQuestion: How can we reduce power consumption?Note: not all data in a large data center is accessedsimultaneously→ disks are not used all the time.Question: can we reduce power consumption by puttingsome disks to sleep?Answer: YES, but there are tradeoffs to investigate.What are the tradeoffs?
E. de Souza e Silva Semana PESC
8/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Disk clusters in Data Centers
Data centers consume a lot of powerQuestion: How can we reduce power consumption?Note: not all data in a large data center is accessedsimultaneously→ disks are not used all the time.Question: can we reduce power consumption by puttingsome disks to sleep?Answer: YES, but there are tradeoffs to investigate.What are the tradeoffs?
E. de Souza e Silva Semana PESC
8/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Disk clusters in Data Centers
Data centers consume a lot of powerQuestion: How can we reduce power consumption?Note: not all data in a large data center is accessedsimultaneously→ disks are not used all the time.Question: can we reduce power consumption by puttingsome disks to sleep?Answer: YES, but there are tradeoffs to investigate.What are the tradeoffs?
E. de Souza e Silva Semana PESC
8/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Disk clusters in Data Centers
Data centers consume a lot of powerQuestion: How can we reduce power consumption?Note: not all data in a large data center is accessedsimultaneously→ disks are not used all the time.Question: can we reduce power consumption by puttingsome disks to sleep?Answer: YES, but there are tradeoffs to investigate.What are the tradeoffs?
E. de Souza e Silva Semana PESC
9/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Another example
2 movie theaters competing with each other. They showthe same movieCustomers that arrive to see movie choose one of thetheaters with equal probability (theaters are identical)Question: after some time:( ) both theathers will receive approx. same amount ofcustomers (and make approx. same amount of money( ) one theater will get much more customers (make muchmore money) than the otherQuestion: can we calculate the probability that one theatermakes more than D dollars than the other?Answer: YES.
E. de Souza e Silva Semana PESC
9/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Another example
2 movie theaters competing with each other. They showthe same movieCustomers that arrive to see movie choose one of thetheaters with equal probability (theaters are identical)Question: after some time:( ) both theathers will receive approx. same amount ofcustomers (and make approx. same amount of money( ) one theater will get much more customers (make muchmore money) than the otherQuestion: can we calculate the probability that one theatermakes more than D dollars than the other?Answer: YES.
E. de Souza e Silva Semana PESC
9/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Another example
2 movie theaters competing with each other. They showthe same movieCustomers that arrive to see movie choose one of thetheaters with equal probability (theaters are identical)Question: after some time:( ) both theathers will receive approx. same amount ofcustomers (and make approx. same amount of money( ) one theater will get much more customers (make muchmore money) than the otherQuestion: can we calculate the probability that one theatermakes more than D dollars than the other?Answer: YES.
E. de Souza e Silva Semana PESC
9/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Another example
2 movie theaters competing with each other. They showthe same movieCustomers that arrive to see movie choose one of thetheaters with equal probability (theaters are identical)Question: after some time:( ) both theathers will receive approx. same amount ofcustomers (and make approx. same amount of money( ) one theater will get much more customers (make muchmore money) than the otherQuestion: can we calculate the probability that one theatermakes more than D dollars than the other?Answer: YES.
E. de Souza e Silva Semana PESC
9/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Another example
2 movie theaters competing with each other. They showthe same movieCustomers that arrive to see movie choose one of thetheaters with equal probability (theaters are identical)Question: after some time:( ) both theathers will receive approx. same amount ofcustomers (and make approx. same amount of money( ) one theater will get much more customers (make muchmore money) than the otherQuestion: can we calculate the probability that one theatermakes more than D dollars than the other?Answer: YES.
E. de Souza e Silva Semana PESC
9/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Another example
2 movie theaters competing with each other. They showthe same movieCustomers that arrive to see movie choose one of thetheaters with equal probability (theaters are identical)Question: after some time:( ) both theathers will receive approx. same amount ofcustomers (and make approx. same amount of money( ) one theater will get much more customers (make muchmore money) than the otherQuestion: can we calculate the probability that one theatermakes more than D dollars than the other?Answer: YES.After a long time there is a 80% chance one of the theatersgot more than 20.000 customers than the other!
E. de Souza e Silva Semana PESC
10/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Modeling Cycle
the real system the model
λ1 + δλ2/δy = z
mentalabstraction
the future
measurementexperimentation
prediction
modelrefinement
the objectives
estimation
E. de Souza e Silva Semana PESC
11/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Machine Learning
E. de Souza e Silva Semana PESC
12/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Big Data
Large amount of data produced and consumed everydaysocial networksonline video streamingmicroblogginggenome informationmeasurements
How to obtain insights from data?What can we learn from the data?
E. de Souza e Silva Semana PESC
13/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Machine Learning
What is
Murphy:Set is methods that can automatically detectpatterns in data
E. de Souza e Silva Semana PESC
14/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Big Data
Large amount of data produced and consumed everydaysocial networksonline video streamingmicroblogginggenome informationmeasurements
How to obtain insights from data?
data insight?
E. de Souza e Silva Semana PESC
15/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Machine Learning
What isMurphy:Set is methods that can automatically detect patterns in data
Uncovered patterns→:
predict future dataperform decision makingplanning
E. de Souza e Silva Semana PESC
15/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Machine Learning
What isMurphy:Set is methods that can automatically detect patterns in data
Uncovered patterns→:
predict future dataperform decision makingplanning
E. de Souza e Silva Semana PESC
16/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Obtaining insights from traces
Machine learning = automatic pattern recognitionPerformance evaluation = model building and analysismachine learning tools can help to solve performanceevaluation problems (and vice versa)
traces recommendations
models(performance evaluation)
machine learning
E. de Souza e Silva Semana PESC
17/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Time Series
Given a time series, how to parameterize model to predictfuture values?
inferring customer behaviormodeling network channel lossesmodeling trafficgenerating workload...
Note: we have traces of time series of one or morevariables.Is there a structure behind the data?
E. de Souza e Silva Semana PESC
17/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Time Series
Given a time series, how to parameterize model to predictfuture values?
inferring customer behaviormodeling network channel lossesmodeling trafficgenerating workload...
Note: we have traces of time series of one or morevariables.Is there a structure behind the data?
E. de Souza e Silva Semana PESC
17/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Time Series
Given a time series, how to parameterize model to predictfuture values?
inferring customer behaviormodeling network channel lossesmodeling trafficgenerating workload...
Note: we have traces of time series of one or morevariables.Is there a structure behind the data?
E. de Souza e Silva Semana PESC
17/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Time Series
Given a time series, how to parameterize model to predictfuture values?
inferring customer behaviormodeling network channel lossesmodeling trafficgenerating workload...
Note: we have traces of time series of one or morevariables.Is there a structure behind the data?
E. de Souza e Silva Semana PESC
17/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Time Series
Given a time series, how to parameterize model to predictfuture values?
inferring customer behaviormodeling network channel lossesmodeling trafficgenerating workload...
Note: we have traces of time series of one or morevariables.Is there a structure behind the data?
E. de Souza e Silva Semana PESC
17/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Time Series
Given a time series, how to parameterize model to predictfuture values?
inferring customer behaviormodeling network channel lossesmodeling trafficgenerating workload...
Note: we have traces of time series of one or morevariables.Is there a structure behind the data?
E. de Souza e Silva Semana PESC
17/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Time Series
Given a time series, how to parameterize model to predictfuture values?
inferring customer behaviormodeling network channel lossesmodeling trafficgenerating workload...
Note: we have traces of time series of one or morevariables.Is there a structure behind the data?
E. de Souza e Silva Semana PESC
17/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Time Series
Given a time series, how to parameterize model to predictfuture values?
inferring customer behaviormodeling network channel lossesmodeling trafficgenerating workload...
Note: we have traces of time series of one or morevariables.Is there a structure behind the data?
E. de Souza e Silva Semana PESC
18/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Summary
Recall from Performance EvaluationMachine Learning
the real system the model
λ1 + δλ2/δy = z
mentalabstraction
the future
measurementexperimentation
prediction
modelrefinement
the objectives
estimation
E. de Souza e Silva Semana PESC
18/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Summary
Recall from Performance EvaluationMachine Learning
the real system the model
λ1 + δλ2/δy = z
mentalabstraction
the future
lots of dataBIGDATA
prediction
the objectives
estimation
E. de Souza e Silva Semana PESC
19/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Projects of our Group
E. de Souza e Silva Semana PESC
20/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Distance Learning Initiative
P&D - servicevideoaula@RNP
E. de Souza e Silva Semana PESC
21/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Distance Learning Initiative
More than 800 videolectures (approximately 40-90 minuteseach)Technology completely developed at the universityCEDERJ Computer Science course started in 2005It has been a service of the RNP since 2011:Videoaula@RNPDesigns started as a research project (CNPq - FAPERJprojects) and later made into a product (supported byRNP) and transfered to RNP.
E. de Souza e Silva Semana PESC
21/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Distance Learning Initiative
More than 800 videolectures (approximately 40-90 minuteseach)Technology completely developed at the universityCEDERJ Computer Science course started in 2005It has been a service of the RNP since 2011:Videoaula@RNPDesigns started as a research project (CNPq - FAPERJprojects) and later made into a product (supported byRNP) and transfered to RNP.
E. de Souza e Silva Semana PESC
21/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Distance Learning Initiative
More than 800 videolectures (approximately 40-90 minuteseach)Technology completely developed at the universityCEDERJ Computer Science course started in 2005It has been a service of the RNP since 2011:Videoaula@RNPDesigns started as a research project (CNPq - FAPERJprojects) and later made into a product (supported byRNP) and transfered to RNP.
E. de Souza e Silva Semana PESC
21/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Distance Learning Initiative
More than 800 videolectures (approximately 40-90 minuteseach)Technology completely developed at the universityCEDERJ Computer Science course started in 2005It has been a service of the RNP since 2011:Videoaula@RNPDesigns started as a research project (CNPq - FAPERJprojects) and later made into a product (supported byRNP) and transfered to RNP.
E. de Souza e Silva Semana PESC
22/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
The Service videoaula@RNPExample
E. de Souza e Silva Semana PESC
23/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
The Service videoaula@RNPExample
E. de Souza e Silva Semana PESC
24/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Videoaula@RNP ServiceUsage
Daily access: almost 7,000 accesses in one day (Feb)Reached more than 110,000 accesses in one month
E. de Souza e Silva Semana PESC
25/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Project supported by GOOGLE
Project
An Intelligent Recommendation System based on VideoLectures for Distance Education
E. de Souza e Silva Semana PESC
26/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Project supported by Google
Empresa investe em estudos académicos do...... ao comportamento de alunos de videoaulas
E. de Souza e Silva Semana PESC
26/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Project supported by Google
E. de Souza e Silva Semana PESC
27/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Project supported by Google
Reportagem Folha do estado de São Paulo, 10 de Junhode 2015http://www1.folha.uol.com.br/tec/2015/06/
1633848-google-dara-bolsas-de-mestrado-e-
doutorado-em-computacao-no-brasil.shtml
E. de Souza e Silva Semana PESC
28/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Project: Recommendation System forVideolecturesOur Objectives
Develop an Intelligent Recommendation System based onVideoLectures for Distance EducationResearch Goals:
To adapt the videolecture material according to individualuser’s needs.Automatically make suggestions to each student inrealtime:
additional written material on the subjectadditional explanation from short videosadditional exercises
Give important feedback to faculty involved on each class
E. de Souza e Silva Semana PESC
28/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Project: Recommendation System forVideolecturesOur Objectives
Develop an Intelligent Recommendation System based onVideoLectures for Distance EducationResearch Goals:
To adapt the videolecture material according to individualuser’s needs.Automatically make suggestions to each student inrealtime:
additional written material on the subjectadditional explanation from short videosadditional exercises
Give important feedback to faculty involved on each class
E. de Souza e Silva Semana PESC
28/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Project: Recommendation System forVideolecturesOur Objectives
Develop an Intelligent Recommendation System based onVideoLectures for Distance EducationResearch Goals:
To adapt the videolecture material according to individualuser’s needs.Automatically make suggestions to each student inrealtime:
additional written material on the subjectadditional explanation from short videosadditional exercises
Give important feedback to faculty involved on each class
E. de Souza e Silva Semana PESC
28/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Project: Recommendation System forVideolecturesOur Objectives
Develop an Intelligent Recommendation System based onVideoLectures for Distance EducationResearch Goals:
To adapt the videolecture material according to individualuser’s needs.Automatically make suggestions to each student inrealtime:
additional written material on the subjectadditional explanation from short videosadditional exercises
Give important feedback to faculty involved on each class
E. de Souza e Silva Semana PESC
28/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Project: Recommendation System forVideolecturesOur Objectives
Develop an Intelligent Recommendation System based onVideoLectures for Distance EducationResearch Goals:
To adapt the videolecture material according to individualuser’s needs.Automatically make suggestions to each student inrealtime:
additional written material on the subjectadditional explanation from short videosadditional exercises
Give important feedback to faculty involved on each class
E. de Souza e Silva Semana PESC
28/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Project: Recommendation System forVideolecturesOur Objectives
Develop an Intelligent Recommendation System based onVideoLectures for Distance EducationResearch Goals:
To adapt the videolecture material according to individualuser’s needs.Automatically make suggestions to each student inrealtime:
additional written material on the subjectadditional explanation from short videosadditional exercises
Give important feedback to faculty involved on each class
E. de Souza e Silva Semana PESC
29/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
The System
Facial Features
webcam
EEG (NeuroSky/Emotiv)
Filter
videolecture
Facial expressionclassifier
Sync
hron
izer
Loggenerator
RECOMMENDER
reco
mm
enda
tion
s
Affectiva EDA/GSR
Filter
IR webcam
Pupil Diameter
E. de Souza e Silva Semana PESC
30/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Startup - Measurements and Planning
TGR - Tecnologia em Gestão e Planejamento de Redes
Measurements and Analysis
E. de Souza e Silva Semana PESC
31/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Parceria Universidade/Empresa
Assunto científico de interesse internacional:Artigos recentes têm sido publicados em veículosinternacionais de renome (2012, 2013)"much remains to be done to improve our understanding ofbroadband services"Transferência de novas técnicas desenvolvidas no LANDpara TGRDesenvolvimento teórico aliado a experimentação emcampoGanhos para a sociedade: tópico de interesse paraformuladores de políticas públicas e consumidoresPesquisa de ponta: Tese de mestrado 2015: Caracterização emodelos para avaliar o desempenho de redes de acesso residencialbaseados em aprendizado de máquina .
E. de Souza e Silva Semana PESC
31/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Parceria Universidade/Empresa
Assunto científico de interesse internacional:Artigos recentes têm sido publicados em veículosinternacionais de renome (2012, 2013)"much remains to be done to improve our understanding ofbroadband services"Transferência de novas técnicas desenvolvidas no LANDpara TGRDesenvolvimento teórico aliado a experimentação emcampoGanhos para a sociedade: tópico de interesse paraformuladores de políticas públicas e consumidoresPesquisa de ponta: Tese de mestrado 2015: Caracterização emodelos para avaliar o desempenho de redes de acesso residencialbaseados em aprendizado de máquina .
E. de Souza e Silva Semana PESC
31/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Parceria Universidade/Empresa
Assunto científico de interesse internacional:Artigos recentes têm sido publicados em veículosinternacionais de renome (2012, 2013)"much remains to be done to improve our understanding ofbroadband services"Communications of the ACM - November 2012Transferência de novas técnicas desenvolvidas no LANDpara TGRDesenvolvimento teórico aliado a experimentação emcampoGanhos para a sociedade: tópico de interesse paraformuladores de políticas públicas e consumidoresPesquisa de ponta: Tese de mestrado 2015: Caracterização emodelos para avaliar o desempenho de redes de acesso residencialbaseados em aprendizado de máquina .
E. de Souza e Silva Semana PESC
31/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Parceria Universidade/Empresa
Assunto científico de interesse internacional:Artigos recentes têm sido publicados em veículosinternacionais de renome (2012, 2013)"much remains to be done to improve our understanding ofbroadband services"Impact of Home Networks on End-to-End Performance(HomNets 2010)Transferência de novas técnicas desenvolvidas no LANDpara TGRDesenvolvimento teórico aliado a experimentação emcampoGanhos para a sociedade: tópico de interesse paraformuladores de políticas públicas e consumidoresPesquisa de ponta: Tese de mestrado 2015: Caracterização emodelos para avaliar o desempenho de redes de acesso residencialbaseados em aprendizado de máquina .E. de Souza e Silva Semana PESC
31/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Parceria Universidade/Empresa
Assunto científico de interesse internacional:Artigos recentes têm sido publicados em veículosinternacionais de renome (2012, 2013)"much remains to be done to improve our understanding ofbroadband services"Broadband Internet Performance: A View From theGateway (ACM/SIGCOMM 2011)Transferência de novas técnicas desenvolvidas no LANDpara TGRDesenvolvimento teórico aliado a experimentação emcampoGanhos para a sociedade: tópico de interesse paraformuladores de políticas públicas e consumidoresPesquisa de ponta: Tese de mestrado 2015: Caracterização emodelos para avaliar o desempenho de redes de acesso residencialbaseados em aprendizado de máquina .E. de Souza e Silva Semana PESC
31/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Parceria Universidade/Empresa
Assunto científico de interesse internacional:Artigos recentes têm sido publicados em veículosinternacionais de renome (2012, 2013)"much remains to be done to improve our understanding ofbroadband services"Speed Measurements of Residential Internet Access(IEEE/PAM 2012)Transferência de novas técnicas desenvolvidas no LANDpara TGRDesenvolvimento teórico aliado a experimentação emcampoGanhos para a sociedade: tópico de interesse paraformuladores de políticas públicas e consumidoresPesquisa de ponta: Tese de mestrado 2015: Caracterização emodelos para avaliar o desempenho de redes de acesso residencialbaseados em aprendizado de máquina .E. de Souza e Silva Semana PESC
31/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Parceria Universidade/Empresa
Assunto científico de interesse internacional:Artigos recentes têm sido publicados em veículosinternacionais de renome (2012, 2013)"much remains to be done to improve our understanding ofbroadband services"Measuring Home Broadband Performance(Communications of the ACM - November 2012)Transferência de novas técnicas desenvolvidas no LANDpara TGRDesenvolvimento teórico aliado a experimentação emcampoGanhos para a sociedade: tópico de interesse paraformuladores de políticas públicas e consumidoresPesquisa de ponta: Tese de mestrado 2015: Caracterização emodelos para avaliar o desempenho de redes de acesso residencialbaseados em aprendizado de máquina .E. de Souza e Silva Semana PESC
31/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Parceria Universidade/Empresa
Assunto científico de interesse internacional:Artigos recentes têm sido publicados em veículosinternacionais de renome (2012, 2013)"much remains to be done to improve our understanding ofbroadband services"The Realities of Home Broadband (CACM Nov 2012)Transferência de novas técnicas desenvolvidas no LANDpara TGRDesenvolvimento teórico aliado a experimentação emcampoGanhos para a sociedade: tópico de interesse paraformuladores de políticas públicas e consumidoresPesquisa de ponta: Tese de mestrado 2015: Caracterização emodelos para avaliar o desempenho de redes de acesso residencialbaseados em aprendizado de máquina .
E. de Souza e Silva Semana PESC
31/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Parceria Universidade/Empresa
Assunto científico de interesse internacional:Artigos recentes têm sido publicados em veículosinternacionais de renome (2012, 2013)"much remains to be done to improve our understanding ofbroadband services"Mixture Models of Endhost Network Traffic (Infocom 2013)Transferência de novas técnicas desenvolvidas no LANDpara TGRDesenvolvimento teórico aliado a experimentação emcampoGanhos para a sociedade: tópico de interesse paraformuladores de políticas públicas e consumidoresPesquisa de ponta: Tese de mestrado 2015: Caracterização emodelos para avaliar o desempenho de redes de acesso residencialbaseados em aprendizado de máquina .
E. de Souza e Silva Semana PESC
31/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Parceria Universidade/Empresa
Assunto científico de interesse internacional:Artigos recentes têm sido publicados em veículosinternacionais de renome (2012, 2013)"much remains to be done to improve our understanding ofbroadband services"Predicting user dissatisfaction with Internet applicationperformance at end-hosts (Infocom 2013)Transferência de novas técnicas desenvolvidas no LANDpara TGRDesenvolvimento teórico aliado a experimentação emcampoGanhos para a sociedade: tópico de interesse paraformuladores de políticas públicas e consumidoresPesquisa de ponta: Tese de mestrado 2015: Caracterização emodelos para avaliar o desempenho de redes de acesso residencialbaseados em aprendizado de máquina .E. de Souza e Silva Semana PESC
31/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Parceria Universidade/Empresa
Assunto científico de interesse internacional:Artigos recentes têm sido publicados em veículosinternacionais de renome (2012, 2013)"much remains to be done to improve our understanding ofbroadband services"Transferência de novas técnicas desenvolvidas no LANDpara TGRDesenvolvimento teórico aliado a experimentação emcampoGanhos para a sociedade: tópico de interesse paraformuladores de políticas públicas e consumidoresPesquisa de ponta: Tese de mestrado 2015: Caracterização emodelos para avaliar o desempenho de redes de acesso residencialbaseados em aprendizado de máquina .
E. de Souza e Silva Semana PESC
31/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Parceria Universidade/Empresa
Assunto científico de interesse internacional:Artigos recentes têm sido publicados em veículosinternacionais de renome (2012, 2013)"much remains to be done to improve our understanding ofbroadband services"Transferência de novas técnicas desenvolvidas no LANDpara TGRDesenvolvimento teórico aliado a experimentação emcampoGanhos para a sociedade: tópico de interesse paraformuladores de políticas públicas e consumidoresPesquisa de ponta: Tese de mestrado 2015: Caracterização emodelos para avaliar o desempenho de redes de acesso residencialbaseados em aprendizado de máquina .
E. de Souza e Silva Semana PESC
31/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Parceria Universidade/Empresa
Assunto científico de interesse internacional:Artigos recentes têm sido publicados em veículosinternacionais de renome (2012, 2013)"much remains to be done to improve our understanding ofbroadband services"Transferência de novas técnicas desenvolvidas no LANDpara TGRDesenvolvimento teórico aliado a experimentação emcampoGanhos para a sociedade: tópico de interesse paraformuladores de políticas públicas e consumidoresPesquisa de ponta: Tese de mestrado 2015: Caracterização emodelos para avaliar o desempenho de redes de acesso residencialbaseados em aprendizado de máquina .
E. de Souza e Silva Semana PESC
31/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Parceria Universidade/Empresa
Assunto científico de interesse internacional:Artigos recentes têm sido publicados em veículosinternacionais de renome (2012, 2013)"much remains to be done to improve our understanding ofbroadband services"Transferência de novas técnicas desenvolvidas no LANDpara TGRDesenvolvimento teórico aliado a experimentação emcampoGanhos para a sociedade: tópico de interesse paraformuladores de políticas públicas e consumidoresPesquisa de ponta: Tese de mestrado 2015: Caracterização emodelos para avaliar o desempenho de redes de acesso residencialbaseados em aprendizado de máquina .
E. de Souza e Silva Semana PESC
32/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Produtos Iniciais
Ferramenta de medição e diagnóstico da qualidade dabanda larga.Software para diagnosticar problemas de instalação emanutençãoSoftware para dimensionamento da capacidade da redeConhecer o tráfego do clienteEntender o comportamento do usuário de banda larga
E. de Souza e Silva Semana PESC
32/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Produtos Iniciais
Ferramenta de medição e diagnóstico da qualidade dabanda larga. Garantia de Qualidade→ economia derecursosSoftware para diagnosticar problemas de instalação emanutençãoSoftware para dimensionamento da capacidade da redeConhecer o tráfego do clienteEntender o comportamento do usuário de banda larga
E. de Souza e Silva Semana PESC
32/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Produtos Iniciais
Ferramenta de medição e diagnóstico da qualidade dabanda larga.Software para diagnosticar problemas de instalação emanutençãoSoftware para dimensionamento da capacidade da redeConhecer o tráfego do clienteEntender o comportamento do usuário de banda larga
E. de Souza e Silva Semana PESC
32/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Produtos Iniciais
Ferramenta de medição e diagnóstico da qualidade dabanda larga.Software para diagnosticar problemas de instalação emanutenção Redução de custosSoftware para dimensionamento da capacidade da redeConhecer o tráfego do clienteEntender o comportamento do usuário de banda larga
E. de Souza e Silva Semana PESC
32/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Produtos Iniciais
Ferramenta de medição e diagnóstico da qualidade dabanda larga.Software para diagnosticar problemas de instalação emanutençãoSoftware para dimensionamento da capacidade da redeConhecer o tráfego do clienteEntender o comportamento do usuário de banda larga
E. de Souza e Silva Semana PESC
32/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Produtos Iniciais
Ferramenta de medição e diagnóstico da qualidade dabanda larga.Software para diagnosticar problemas de instalação emanutençãoSoftware para dimensionamento da capacidade da redePlanejamento futuro→ economia de recursosConhecer o tráfego do clienteEntender o comportamento do usuário de banda larga
E. de Souza e Silva Semana PESC
32/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Produtos Iniciais
Ferramenta de medição e diagnóstico da qualidade dabanda larga.Software para diagnosticar problemas de instalação emanutençãoSoftware para dimensionamento da capacidade da redeConhecer o tráfego do clienteEntender o comportamento do usuário de banda larga
E. de Souza e Silva Semana PESC
32/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Produtos Iniciais
Ferramenta de medição e diagnóstico da qualidade dabanda larga.Software para diagnosticar problemas de instalação emanutençãoSoftware para dimensionamento da capacidade da redeConhecer o tráfego do cliente Planejamento→ economiade recursosEntender o comportamento do usuário de banda larga
E. de Souza e Silva Semana PESC
32/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Produtos Iniciais
Ferramenta de medição e diagnóstico da qualidade dabanda larga.Software para diagnosticar problemas de instalação emanutençãoSoftware para dimensionamento da capacidade da redeConhecer o tráfego do clienteEntender o comportamento do usuário de banda larga
E. de Souza e Silva Semana PESC
33/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Projeto
Implementação do software no firmware da INTEL(gateway embedded solution)
INTELImplementação do software no firmware em escala global
E. de Souza e Silva Semana PESC
34/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Gateway Inteligente
E. de Souza e Silva Semana PESC
35/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Research/Development in our Group
Fault Tolerance is EssencialPerformance always matterStuart FeldmanACM’s Software System Award, Vice-President Eng.Google
E. de Souza e Silva Semana PESC
35/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Research/Development in our Group
Fault Tolerance is EssencialPerformance always matterStuart FeldmanACM’s Software System Award, Vice-President Eng.Google
E. de Souza e Silva Semana PESC
35/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
Research/Development in our Group
Fault Tolerance is EssencialPerformance always matterStuart FeldmanACM’s Software System Award, Vice-President Eng.Google
E. de Souza e Silva Semana PESC
36/38
Preliminaries Performance Evaluation Examples Modeling Machine Learning Projects Startup - TGR Summary
FIM
OBRIGADO!PERGUNTAS?
www.abc.org.br/~edmundowww.land.ufrj.br/~edmundo
E. de Souza e Silva Semana PESC
Recommended