Microsoft Research conducts research across multiple domains to push the boundaries of computing. It has over 1,100 researchers working in locations around the world. Microsoft Research scientists have won over 320 major awards for their work. The document also provides details on Microsoft Research's joint research institutes and areas of research focus, including machine learning and intelligence.
Le potentiel du Machine Learning et de l’analyse prédictive à portée de votre entreprise
1.
2. Innovation: a definition
«Innovation is the ability to
create value while bringing
something new in the field and
ensuring that the appropriation
of this novelty is optimum.»
Arnaud Groff, Dr in
« Management de l'Innovation & de la
Créativité »
3. Microsoft Research (MSR)
Redmond (1991) Cambridge (1997) Beijing (1998)
Silicon Valley (2001) Bangalore (2005) New England (2008)
More than 1,100 brilliant scientists and engineers push the boundaries of
computing in multiple research areas and include contributions to Kinect for
Xbox 360, work to develop an HIV vaccine, and advancing education
techniques in rural communities.
4ème mondial toute industrie confondue
1er mondial dans l’industrie du logiciel
4.
5.
6. Microsoft Research scientists have
won more than 320 major awards,
including the Turing Award, MacArthur
Foundation Fellowship, MIT
Technology Review’s TR35 Award, the
Draper Prize, IEEE John von Neumann
Medal, IEEE Piore Award, the Kyoto
Prize, multiple Oscars and a British
knighthood.
Microsoft Research Awards
7. Joint research institutes
INRIA, France
Software security; Formal methods;
Applications of computer science
research to science
www.msr-inria.inria.fr
University of Trento, Italy
Computational tools for systems
biology
www.cosbi.eu
Barcelona Super
Computing Centre, Spain
Multi core systems; Architectures and
programming; Language runtimes
www.bscmsrc.eu
8. Algorithms and Theory
Exploring the theoretical foundations of computing, and efficient
algorithms for a wide variety of problems.
Communication and Collaboration
Enabling people to reach each other easily regardless of network
or device.
Computational Linguistics
Focusing on machine translation, multilingual systems
and natural-language processing.
Computational Science
Providing computational support to unravel the
mysteries of the universe.
Computer Systems and Networking
Improving efficiency in the deployment, operation management and
security of distributed applications.
Computer Vision
Teaching computers to see and
understand the visual world.
Data Mining and Management
Creating systems for accessing and managing large collections of data,
and algorithms for finding patterns and insights within the data.
Economics and Computation
Exploring the connections between economics and
computer science, and creating economic models of
online systems.
Education
Applying computing to help people learn. Expanding
programs in computer-science education.
Gaming
Exploring new technologies to enhance the gaming experience, and
identifying and developing innovative technologies and curricula to
aid in educational activities.
Graphics and Multimedia
Addressing challenges in displaying complex computer graphics
models, in multiresolution signal representations and enhancement,
and in compression of geometry and multimedia data.
Hardware and Devices
Building the hardware that will support the
next generation of software.
Health and Well-Being
Leading innovation in assisted cognition, bioinformatics,
synthetic biology, and biomedicine.
Human-Computer Interaction
Advancing the way users interact with computing
devices.
Machine Learning and Intelligence
Building software that automatically learns from data to create
more advanced, intelligent computer systems.
Mobile Computing
Exploring how to build mobile devices and services that
are efficient, responsive, and usable.
Quantum Computing
Exploiting quantum physics to create a new
generation of computing devices.
Search, Information Retrieval and
Knowledge Management
Exploring indexing and classification technologies, entity extraction, and user-
experience concepts that help people organize and find information.
Security and Privacy
Ensuring the privacy and integrity of our
computations and data.
Social Media
Exploring how digital media are changing the way people
work, play, and connect with each other.
Social Science
Exploring how people use
computing in their daily lives.
Software Development, Programming
Principles, Tools, and Languages
Improving quality and efficiency throughout the software-development
process.
Speech Recognition, Synthesis,
and Dialog Systems
Teaching computers how to both speak and listen.
Technology for Emerging Markets
Understanding how technologies can address the needs and
aspirations of people in the world’s developing communities.
Machine Learning and Intelligence
Building software that automatically learns from data to
create more advanced, intelligent computer systems.
10. Qu’est-ce que le Machine Learning ?
Des méthodes et des systèmes qui …
en fonction
des données
collectées
de nouvelles
données en
fonction des
données
collectées
une action
étant donné
une fonction
d’utilité
une structure
cachée des
données
les données
en des
descriptions
concises
s’adaptent prédisent optimisent extraient résument
Champ d’études qui donne aux ordinateurs la capacité
d’apprendre sans avoir besoin d’être explicitement programmés
11. 20 ans de Machine Learning chez Microsoft
1992
début de la
reconnaissance vocale
2000
système de
recommandation dans
Commerce Server
2005
Data Mining dans
SQL Server 2005
2008
Kinect pour XBOX
2009
Flash Fill pour Excel 2013
2014
Microsoft Azure
Machine Learning
12.
13. from Machine Learning to Predictive Analysis
In business, predictive models exploit patterns
found in historical and transactional data to
identify risks and opportunities
Crime Fighting
Fraud Detection
Marketing
Advertising
Family and
Personal Life
Human Resources
Financial Risk
Insurance
Healthcare
Fault Detection for
Safety and Efficiency
16. Prédire les prochains souscripteurs de crédit automobile
Modèle
comporte-
mental
Caractéristiques
Succession
d’événements
Contexte
Social
Je suis un cadre dans
l’informatique de 42 ans,
propriétaire de ma
résidence, avec 2
enfants…
… j’ai réalisé deux
dépenses de puériculture
supérieures à 200€
chacune dans les trois
derniers mois…
…mes amis viennent de
souscrire des crédits
automobile…
…dans trois semaines
aura lieu le salon de
l’automobile Porte de
Versailles…
19. Pier 1 Imports
Pier 1 Imports discuss how they predict which product the customer
might want to purchase next, helping to build a better relationship
with their customers.
24. Ambiant Intelligence for a better Customer Experience
“Consistent, Personalized, and Self-learning”
Customer
Business Operations
Orders / CRM
Inventory / IOT
Finance
Services
External sources
Rating
Social / Weather
Demographics
Partners
Integrated Enterprise Data
Single View of the Customer
Information as a service
Scores Segmentation
High-Value Services
Sales
Campaign
Churn
Prices
Interaction
Management
Channels
Web
Stores
Support
Devices
Lounges
Partners
Learning
25. Ambiant Intelligence for a better Customer Experience
“Consistent, Personalized, and Self-learning”
Customer
Business Operations
Orders / CRM
Inventory / IOT
Finance
Services
External sources
Rating
Social / Weather
Demographics
Partners
Integrated Enterprise Data
Single View of the Customer
Information as a service
Scores Segmentation
High-Value Services
Sales
Campaign
Churn
Prices
Interaction
Management
Channels
Web
Stores
Support
Devices
Lounge
Partners
Learning
Advanced and Innovative Dashboards from any device
26. Crunching des données
internes / externes2
Mode opératoire standard pour un projet ML
Compréhension du métier et
des données de nos clients1
Vérification itérative
avec les métiers4Mise en production du
modèle prédictif final5
Mise au point des
modèles mathématiques3
30. BIG DATA / MACHINE LEARNING : un état d’esprit
Typologie simplifiée des projets Big Data / Machine Learning
Expérimentation
Big Data (Data Lab)
Industrialisation de la
production d’indicateurs
Focalisé sur la production
rapide de résultats
Focalisé sur les moyens
Scientifique
(Exploratoire)
Ingénieur
(Top-Down ou Bottom-Up)
Disruption,
Accepter l’erreur
Continuité,
Aversion au risque
Métiers
« Shadow IT »
IT
« Core IT »
Métiers & IT
« Fast IT »
≠
31. Business Value Workshop
La Data Science au service de vos métiers
MICROSOFT
SERVICES
De la Data aux Insights : quels scénarios innovants pour mieux exploiter les données ?
Introduction autour des nouvelles tendances et enjeux
du marché ainsi que de la vision de Microsoft sur la Data
Science
Compréhension des enjeux métiers du client et des données
manipulées par celui-ci
Recensement des intuitions du client
Identification des questions « Machine Learning » intéressant
le client et valorisation de celles-ci
Choix de la question la plus pertinente et proposition de
pilote pour y répondre
Agenda – ½ journée
Problématique
Accompagnement sur la mise en place des scénarios identifiés
Objectif de l’atelier :
• Présenter les tendances et nouveaux usages autour des données
ainsi que les opportunités offertes par la Data Science avec
Microsoft
• Imaginer et formaliser un ou plusieurs scénarios cibles pour
répondre à vos problématiques métiers
Vue d’ensemble
Préparation :
• Identification d’un sponsor client, puis des participants à inviter
• Rendez-vous de qualification avec le sponsor,
1h pour identifier ses enjeux et définir sa problématique
Audience attendue :
Pour plus d’informations : BVW@MICROSOFT.COM
32.
33. Fully
managed
Integrated Flexible Deploy in
minutes
No software to install,
no hardware to
manage, all you need is
an Azure subscription.
Drag, drop and
connect interface.
Data sources with just
a drop down; run
across any data.
Built-in collection of
best of breed
algorithms with no
coding required. Drop
in custom R or use
popular CRAN
packages.
Operationalize models
as web services with a
single click.
Monetize in Machine
Learning Marketplace.
34. Business users access results from anywhere, on any device
Delivering Advanced Analytics
• HDInsight
• SQL Server VM
• SQL DB
• Blobs & Tables
Devices Applications Dashboards
Data Microsoft Azure Machine Learning
Storage space
Integrated development
environment for Machine
Learning
ML
Studio
Business challenge Business valueModeling Deployment
• Desktop files
• Excel spreadsheet
• Other data
files on PC
Cloud
Local
Data to model to web services in minutes
http://studio.azureml
.net
Web
Clients
API
Model is now a web svc
Monetize this API
36. API examples
Green Score, Wealth Score, Giving Score
Frequently Bought Together API
Recommendations API
Anomaly Detection API
Lexicon Based Sentiment Analysis
Forecasting-Exponential Smoothing
Forecasting - ETS+STL
Forecasting-AutoRegressive Integrated
Moving Average (ARIMA)
Binary Classifier API
Cluster Model API
Survival Analysis API
Multivariate Linear Regression API
Survival Analysis API
Multivariate Linear Regression API
Normal Distribution Quantile Calculator
Binomial Distribution Quantile Calculator
datamarket.azure.com
Notes de l'éditeur
Family and Personal Life
Nokia : prédire avec un jour d’avance la position géographique des consommateurs à partir de l’historique des déplacements et des interactions sur les réseaux sociaux
Marketing Advertising
FedEx : arrive à prédire quels clients vont se tourner vers la concurrence avec une efficacité entre 65 et 90% !
Amazon : 35% des revenus proviennent du système de recommandations
Financial Risk Insurance
Allstate : multiplication par trois de la pertinence de la prédiction d’accidents corporels calculée uniquement à partir des caractéristiques des véhicules
Citygroup : a créé des modèles de risques pour les prêts bancaires personnalisés au niveau d’une région et d’une industrie (à partir de 30 ans d’historique)
Healthcare
RiskPrediction : prédit les risques de décès lors d’une opération chirurgicale (mineure, importante ou complexe)
Travaux d’universités : ont mis au point un modèle capable de prédire 80% des naissances prématurées à partir d’un simple examen de sang dès la 24ème semaine
Human Resources
US Air Force : prédit les capacités des candidats à être capable d’occuper des positions
LinkedIn : est capable d’enrichir votre profil avec des capacités déduites du contenu écrit
Fault Detection for Safety and Efficiency
Con Edison : à NYC, prédiction des taux de panne des câbles de distribution d’électricity avec une mise à jour, trois fois par heure, du niveau de risque sur les panneaux de contrôle des opérateurs
High-Tech : prédisent la probabilité d’échec d’imprimantes, disques durs… afin de pouvoir charger les camions de réparation en avance de phase
Crime Fighting Fraud Detection
Los Angeles : des villes comme LA arrivent à prédire les zones dans lesquelles un crime va avoir lieu et peuvent ainsi diriger en avance de phase les patrouilles de police
Chicago : les caractéristiques du crime et des victimes aident la police à prédire si le crime va pouvoir être résolu
I’m sure you’ve heard Satya talk about the Microsoft focus on enabling a cloud-first, mobile-first world. This shouldn’t be translated as abandoning the on-premises world, which is certainly not going anywhere, but instead putting the right tools in the cloud to enable modern business in ways really impossible before. Machine Learning is a great example of this.
I spoke earlier about the vast amounts of data that companies are generating and attempting to consume and make sense out of today. Not only do companies have to think about ingesting that data and buying and managing all the systems to do so – they then have to think about buying and managing additional hardware and software for advanced analytics. That’s a show stopper for many companies today.
Azure Machine Learning is a fully managed service in the cloud. That means your company literally needs an Azure subscription to utilize this – that’s all. And you don’t even need that to give it a spin, off azure.com/ml, we offer a free version that only requires a Microsoft account ID (LiveID) to get started right away. You can see now how we’re making the traditionally huge expense of this technology a part of the past.
Another issue solved with the use of the cloud is the problem around integration. Azure data sources like HDInsight, Azure SQL Database, SQL Server in a Virtual Machine and more can be brought into the modeling environment with a simple drop down. And data that’s not in the cloud can be modeled against by simply dropping a training set from on-premises into the built in storage space in Azure ML. The only data that must be in the cloud is the training set, but once the model is live as a web service it can literally be run across any data, anywhere.
We also understand that in terms of the talent base for machine learning implementation, there is a wide range of skills to serve with our service. That is why we have built a tool that is flexible to address the user brand new to machine learning as well as the seasoned programmer in the open source language R. With our battle tested algorithms from businesses like Bing, Xbox, etc., a nascent advanced analytics professional can drag and drop in data sources and experiment with machine learning without writing a line of code. More experienced implementers can drag and drop in their custom R code as well as use one of the over 350 R packages built into Azure ML. Essentially the use of Azure ML can be as simple or complex as your talent and business require.
Lastly – and this is the one thing I want to be sure you take away from the presentation – implementers can deploy a model as a web service in minutes. That is unique in the market to Azure ML. What that means is that models that took weeks or months to put into production and apply to their business problem can now be applied to the business problem right away. Additionally, I mentioned how models are typically so hard to put into production due to disconnected systems and languages that once they’re live they tend to stay live – and get stale and useless? Once the developer has wired up the web service to the company dashboard, or PowerBI or the app or any number of uses you can seamlessly adjust the model – that human input I was talking about – with zero additional development work. Thus, the CMO contacts the data scientist and says they’d like to adjust the output on their PowerBI dashboard to account for new variables as a result of a competitors movements – the implementer can make that change and the dashboard can reflect that in hours – not days.
So what does that look like from an architectural perspective? Machine learning is a technology in which you work from business problem backwards. Let’s say I have an issue of customer churn. I don’t know why my best customers are leaving and I need to find out. I have things like Twitter/Facebook/Blog entries in HDInsight – our Hadoop implementation in the cloud – and it’s streaming in daily from the web. On premises I have my customer sales data and buying behavior.
I can then bring in the training set data from HDInsight and a subset of my on-premises customer data into the built-in storage space. I can then model against that training set in ML Studio – which is the playground for the data scientist or advanced analytic developer. In this space the implementer trains and tests the model until she is satisfied that the model will deliver the answer to the question of customer churn. Not only why the customers are departing, but predictive analytics to tell the company which ones are currently at risk based on past data. That way the sales and marketing departments can target those specific customers with the right activities to solve for why they’re leaving in the first place.
The implementer then literally pushes a “Yes” button in the tool to send the finished model into staging, with a flag on the Microsoft Azure portal letting the owner of the all-up portal experience know the model is ready to go. Again – this is a unique and differentiated experience with Azure ML – we are the only ones who offer the ability to push a customized model to production this easily and quickly. Once pushed live, this is now surfaced as a web service which can run over any data, anywhere. If this is running over on-premises data, the data is never persisted in the cloud, so again the only data that must be in the cloud is the original training set, which can be anonymized and removed once the modeling is done for those customers with compliance/security concerns around data in the cloud.
This finished web service can now be called from the company dashboard, where the CMO can easily consume the results and advise the teams accordingly. And, as the company needs change, the implementer need only to revisit the model in ML Studio, adjust it and push it to staging again to literally have the model swap out underneath the live web service.
But what if the company doesn’t have an implementer in house? In that case, they can go right to the Azure Machine Learning Marketplace, where there are live hosted web services already existing to solve common problems such as this. They can be simply hooked up to apps, services and dashboards for this type of solution. This is also a value-add for companies and implementers looking to monetize their own machine learning solutions. Off azure.com/ml on Machine Learning Center we have detailed instructions on how to leverage this to create, monetize and scale your own ML offerings here.
Combining World Class Algorithms plus a very simple drag and drop interface.