5. BIG DATA SERVICES
Start-to-finish scalable and flexible solutions for all your data:
Architecture & design
Implementation & integration
Platform setup & maintenance
Consulting & training
30-11-2017
Anchormen @ Talent Event
6. DATA PLATFORM SERVICES
Unified approach towards data management and integration
that can improve scalability and flexibility.
Consulting on design, installation, and configuration of
complex data platforms.
Focus on large-scale databases.
Both greenfield projects and existing situations.
30-11-2017
Anchormen @ Talent Event
7. DATA SCIENCE & A.I.
Creating valuable new insights through prediction models and
A.I.:
High caliber data consulting
Recommendations & chatbots & predictive maintenance
Process & text mining
Computer vision
30-11-2017
Anchormen @ Talent Event
8. HIGH POTENTIAL PROGRAM
We will find the right people for you and guide them to success.
Data engineers & scientists
1-year program
Training & guidance
30-11-2017
Anchormen @ Talent Event
9. THE ROAD SO FAR
30-11-2017
Anchormen @ Talent Event
13. “LEARNING FROM DATA MEANS INFERRING
WHAT WE DON’T KNOW FROM WHAT WE KNOW.”
CHRIS POOL
LEAD DATA SCIENTIST @ ANCHORMEN
30-11-2017
Anchormen @ Talent Event
14. LEARNING TO PREDICT
For a given object you are asked to make a prediction
Is tomorrow a good day for playing football?
Is this tweet positive or negative?
Does this image contain a car?
Learning is making such predictions after observing data.
MACHINE LEARNING?
30-11-2017
Anchormen @ Talent Event
15. WHAT HAPPENS IN LEARNING
The learning algorithm observes given examples
It tries to find common patterns that explain the data: it tries to generalize so that predictions can be made for new examples
Exactly how this is done depends on what algorithm we are using
30-11-2017
Anchormen @ Talent Event
16. HOW GOOD IS AI?
30-11-2017
Anchormen @ Talent Event
18. NATURAL LANGUAGE
PROCESSING
Speech Recognition is at 5.1% (Xiong et. al 2017)
Statistical Machine Translation around 85% human (BLEU
score)
Neural Machine Translation is the new contender
30-11-2017
Anchormen @ Talent Event
19. “THE BOTTLENECK NOW IS IN MANAGEMENT,
IMPLEMENTATION, AND BUSINESS IMAGINATION.”
BRYNJOLFSSON & MCAFEE
HARVARD BUSINESS REVIEW (JULY 2017)
30-11-2017
Anchormen @ Talent Event
21. THE CASE (1)
A large part of any online advertising campaign is determining how much a click and its conversion is worth.
While bidding for advertising space is typically a blind auction with special sets of rules, optimizing your bid is difficult.
External influences like accidents, weather conditions, news, PR activities and events, tv subtitles or tv guides, all have impact on
online behaviour and activities
30-11-2017
Anchormen @ Talent Event
22. THE CASE (2)
Google trends is a good indicator for populair topics
Can we use Twitter data to explain these peaks?
Challenges:
Lots of spam on Twitter
Keyword != topic
30-11-2017
Anchormen @ Talent Event
23. STRUKTON RAIL
Strukton rail provides solutions for rail infrastructure and
electrical systems in rolling equipment.
Contracted by ProRail for maintaining (parts of) the Dutch
railroads
Performance contract
High competitive market
30-11-2017
Anchormen @ Talent Event
24. VITENS
The largest water company in the Netherlands
Goal to become data-driven
A.I. is relatively new
30-11-2017
Anchormen @ Talent Event
25. AI CHAT-BOT
One of the most challenging parts of a chat-bot (and NLP) is
interpreting a sentence because:
Ambiguity
People use slang / abbreviations
Grammar errors
Need knowledge of the world
30-11-2017
Anchormen @ Talent Event
27. METHODOLOGY
Define clear goals
Avoid searching for patterns in the data without a clear goal.
Use the right tools
Focus on tools that provide quick results.
Test and validate often
Create a data driven mindset within your organization.
Fail fast and fail forward
Data Science is highly innovative. Plans have to be changed all
the time.
30-11-2017
Anchormen @ Talent Event
29. A/B TESTING
Achieve a baseline
Measure your improvement
30-11-2017
Anchormen @ Talent Event
30. CONTINOUS
INTEGRATION /
DELIVERY
DS/AI work results in code
Code should adhere quality standards
Include quantitative model performance
30-11-2017
Anchormen @ Talent Event
31. GOVERNANCE
What data was used to train the model?
What where the model parameters?
Which customers interacted with it?
30-11-2017
Anchormen @ Talent Event
32. “ORGANIZATIONS NEED TO OPEN UP TO NEW INSIGHTS”
GEOFFREY VAN MEER
HEAD OF DATA INTELLIGENCE @ RTL
30-11-2017
Anchormen @ Talent Event
36. CLASSIFICATION
Predict a class using examples, for example the traffic light.
Other examples:
Sentiment analysis
Predictive maintenance
30-11-2017
Anchormen @ Talent Event
37. CLUSTERING
Group observations in “meaningful” groups
Group a set of documents into groups about topic
Group similar observations.
30-11-2017
Anchormen @ Talent Event
39. FINANCIAL FORECASTING
Client: International Travel Agency
I
•Initial Business Question
What factors determine the right price for a house? (e.g. WiFi, Swimming pool, distance to beach)
II
•Technical solutions
Time series analysis (Season, Trend & Cyclical decomposition)
Regression and Trigger Models (Explaining residuals with non-frequent events)
Transfer Learning (Combining models from different regions)
III
•Primary data sources
Financial ledger (SAP)
Social media data (Twitter Firehose)
Web traffic data (Google Analytics)
IV
•Business impact
Insight in how investing in one region effects sales in neighboring regions
Presenting C-level managers with improved financial predictions
Revision of online marketing strategies
V
•Spin-off
Photo analysis, what makes a good photo? Clear guidelines for photographers?
Consumer segmentation, website stability, search
Competitor analysis, pricing models, AdWords and keywords
Anchormen @ Talent Event
40. MARKETING SPEND ANALYSIS
Client: Global Consumer Goods
I
•Initial Business Question
What is the best moment to buy ads? How to align TV campaign with online events?
II
•Technical solutions
Time series analysis (Season, Trend & Cyclical decomposition)
Regression and Trigger Models (Explaining residuals with rare events)
Transfer Learning (Combining models from different regions)
III
•Primary data sources
Marketing spend (SAP)
Social Media (Brand24)
Weather data (Weather Underground)
IV
•Business impact
Higher return on TV campaigns
Insight in what triggers people to buy flu related products
V
•Spin-off
Mosquito index
Anchormen @ Talent Event
41. LOGISTICS & STOCK OPTIMIZING
Client: Logistical Organisation
I
•Initial Business Question
How can we achieve better usage of machines?
II
•Technical solutions
Process mining
Simple reports deliberately targeted across intra-organizational boundaries
Programmed simulations to analyze complex interaction of simple processes
III
•Primary data sources
Maintenance logs (relational database)
Logs and standing data of supplies and procurements (ERP system)
Machine logs
IV
•Business impact
Maintenance planners now have access to historical data while planning
Impact on planning is now calculated real-time during Long-Term Planning meetings
List of parts were identified where higher supply levels will increase availability
V
•Spin-off
Promotion of a data-driven culture
Anchormen @ Talent Event
source: ministry of Defence
Editor's Notes
The speed, variety and volume of data is evergrowing. By integrating and unlocking information and making it accessible in a big data platform we help organizations work more effective.
A big data platform is characterized by the following features; it supports diverse types of data and use cases, it’s horizontally scalable (in both storage and computing power) and it’s flexible (components can easily be added or removed).
Anchormen provides a wide range of Big Data services. From advising on the infrastructure, hosting (on-premise or in the cloud) and actually setting up the platform, to integrating sources and developing data-driven applications.
For a worry-free big data platform we offer support.
boost your business
By making smart use of available techniques Anchormen developed a range of products that feed A.I. systems with valuable data streams. Products, techniques and platforms we know and trust include recommendation engines, A.I. chatbots, predictive maintenance, process mining and text mining.
The business value of A.I. solutions can be powerfully demonstrated.
Go to our booth for our atonomious cars en face recognation analysis
Creating valuable new insights starts with the investigation of structured and unstructured data.
From here we create models that can recognize and explore underlying patterns with razor-sharp efficiency. But this is just the beginning.
These valuable insights must be implemented in a real-time production environment.
By making smart use of available techniques Anchormen developed a range of products that feed A.I. systems with valuable data streams. Products, techniques and platforms we know and trust include recommendation engines, A.I. chatbots, predictive maintenance, process mining and text mining.
The business value of A.I. solutions can be powerfully demonstrated.
Go to our booth for our atonomious cars en face recognation analysis
With our High Potential Program Anchormen offers a high quality and valuable solution for companies that are looking for (big) data engineers, data scientists and A.I. experts.
As most companies lack experience with data professionals, it’s hard to select the appropriate job profile, let alone check the quality of their work.
In our High Potential Program we conclude a 1-year contract with graduates from IT-related courses.
During this year they will work as trainees at our client (4 days a week) and at the same time follow various technical trainings at one of our offices (1 day a week).
By the end of the year – in which our senior experts guide them and assess their work – they are highly skilled and can be employed by the client.
Large IT providers often deliver ICT projects in one go.
After a long development process, companies finally get to see results, but it is often the long wait between approval and delivery of the project that causes problems.
Our methodology is different:
Be more productive by guaranteeing quality
Test your code, but also your models
Make code reviews part of your delivery pipeline
Governance touches upon many topics
Data access for data scientists
Lineage involves the flow of data, but also log your model parameters for reconstruction
With Deep Learning this is a challenge
GDPR also affects models trained on consumer data?
Organisations and technology need to grow towards each other
Our collecitve experiences
Add strings beginning to end
One by one
First a tone, then music