The conference BigData Paris was recently organized the 6th and 7th of March 2017. I though I would share my notes regarding some of the talks I attended.

Keynotes

Opening Keynote with Guillaume Poubard, ANSSI (Day 1 at 9:00AM)

• Agence nationale de la sécurité des systèmes d’information, ANSI is a state agency to support companies around cyber security
• Not to be mistaken by the CNIL (Commission nationale de l’informatique et des libertés)
• Emphasis on security while dealing with data. Sign that the market is maturing
• Key risks:
  • Economic spy activity (with data theft) - no usually detectable
  • Sabotage activities
• Security Principles
  • Governance (It is not only an IT problem)
  • Training people around simple security measures
  • IT needs to be design to be resilient to cyber attacks and have the capacity to detect such attacks
• Actionable plan?
  • See the ANSSI guides
  • Work with third-parties for audits
  • Training
• Quote “Company that will succeed are companies that are secured)

Big Data, Cloud and IoT, presented by IDC:

• Spending: Traditional IT - down / Cloud: Up / IoT, robotics: x2 Up
• Fatest growing IoT: Manufacturing + Retail
• By 2018: Cloud will be the preferred mechanism for analytics
• By 2020: 45% of IT infra + app will be cloud
• Next thing: collaboration across companies
• 40% of the company will be a digital / 40% of projects will monetize big data
• Uncertainty: cross-sharing of data between different companies + gap in skills ++ EU: provide free flow of data through rules and regulation
• Quote: “Use data for competitive advantage””

Netflix: how “Stranger Things” can happen with Visual Analytics, by Jasan Flittner, Netflix (Day 1 at 12:10PM)

• Use case of in-house Big Data analytic with Tableau
• Work on a 60PB DW on S3
• Toolkit:
  • Storage: s3, hadoop, redshift, teradata
  • Analytic: hive, spark, presto
  • Viz:: Tableau, sting (n internal project with Netflix in Python to access data), Microstrategy
• A couple of techniques to use Tableau in BigData:
  • hive + tableau: odbc connection (through the thrift server) / used of materialized views / odbc optimization
  • spark + tableau: less query reliability
  • presto + tableau
  • use of Tableau Data Extract
• Mentioned of addition offers worth to check out:
  • Druid
  • Hyper from Tableau

Deep Learning and Big Data applied to your Business: the key factor to success, by Luming Wang, UBER (Day 2 at 9:30AM)

• Big data challenges: data quality, privacy, common schema / data joining problem, to scale data processing platform, domain knowledge
• Machine learning for big data:
  • ML is a bigger hammer for getting value of Big Data
  • Better prediction (regression, random forest, XG boosted trees)
  • Better clustering (unsupervised learning)
• Deep learning: better results than ML for specifics domains
• Deep learning limitation:
  • Require large volume
  • Training takes times
  • Model serving performance is slower than ML
• Leadership, vision and future
  • Be a believer
  • Make proper investment
  • Understand limitation and risks
  • Ask questions
  • Do it now: customer won’t ask for it
• People and roles
  • Skill sets:: domain knowledge, understand ML best practice, turn solution into product
  • Roles: researcher/applied scientist , data scientist, data engineer, ml software engineer
• Products:
  • mknet
  • Microsoft Cognitive Toolkit - CNTK
  • Tensor Flow
  • MLlib

Use of data in companies and customer trust: Study by Boston Consulting group, by Elias Baltassis (Day 2 at 10:00AM)

• Companies fail to properly communicate with their customers around the data being collected
• Perceived data misuses => leading to lower revenue
• need to implement a cohesive data governance
• Keys:
  • Consumer’s trust will be a competitive advantage
  • Some data are more sensitive than others
  • Distrust of companies is prevailing
  • Company actions (or lack of) are often responsible for this situation
  • Distrust and misuse of data starts having significant financial impact
  • Only frequent communication and a holistic Data Governance is the solution

Innovation Trophy (Day 2 at 12h40)

• Tellmeplus focus on analytic for IoT (driven by the Business Object creator)
• Log Island, a open-source project to process log data
• Domlmen focuses on customer data acquisitions for physical stores
• datadome protect the content of web sites from bots
• zelros: “Turn Enterprise Data into conversational assistants”

Round-Tables

Round-Table around Security

• Fact: Less than 10% of companies made any kind of impact analysis.
• Hard to budget for loss inherent to security breach as oppose to budget investment to strength security
• Interesting use case regarding the security breach with TV5 with Yves Bigot.

Chief Data Officer: new advantage in your strategy? (Day 2 at 10:50AM)

• Characteristics:
  • Has often a team (data lab type)
  • Need to propose solution to improve performance
  • Can come from the start-up world, or from the consulting world (as already familiar with transformation project)
• Use Case #1: Havas - No Need to have a CDO (as each function within HAVAS is different in term of data.
• Use Case #2: Sociéte Genérale:
  • 1 CDO per business function piloted by a lead CDO
  • help business with quality and security of the data, and the development of the data culture
  • View: CDO is a temporary function while the “data culture” becomes more wide spread.
• Note that one of the panel member came from etalab, a public body aimed at opening public data. One of the projects they worked on is a web site that predict which companies are likely to hire.

Use Cases

A couple of the sessions were dedicated to how companies were building analytic and big-data related capabilities.

Trainline (Day 1 at 3:00PM)

• Trainline is a web site that enables you to buy tickets everywhere in Europe.
• Presentation of the “Data Initiative” after it became clear that too much time was spent on extracting the data, and not enough time on the prediction.
• Automation of the Data Lab using Dataiku
• Data analysis to perform targeted marketing regarding the generation of online ads.

Under Armour (Day 1 at 4:00PM)

• Under Armour, on top of providing sportswear, provides connected apps (around sport and well being)
• How to create value: Data + Machine Learning + Infrastructure
• Different cases ++ Find calories in restaurant menus (as to track your calorie intake through an app) ++ Send coach messages at the right time

Auchan: Direct - Day 1/16h20

• Auchan: Direct deliver groceries.
• Case for building a data lab
• Goal: “Provide insights to collaborators as to improve customer experience”
• Virtual circle: Learn -> Data viz -> Big Data (and repeat)

SeLoger.com (Day 1 at 4:40PM)

• SeLoger.com facilitate the buy or sell of houses.
• Use case with AlloMedia to analyse the content of voice calls
• No interruption in data collection between web interactions and phone interactions

AirFrance-KLM (Day 1 at 5:00PM)

• Objectives: Discover, Improve Knowledge, and facilitate a better Interaction with Customer

Feedback from Consulting Work

Trends and leassons learnt from Acenture (Day 1 at 11h00)

• Process Offer: Accenture Connected Analytics Experience
  • Collaboration with multiple actors as to prototype new uses around the data
  • design-thinking, analytics
  • Actors: business, data scientist, IT
  • Few clients “blend” the three actors
• Software Offer: Accenture Insights Platform: offer between IaaS and PaaS
• Advice: “Do not do POC”
  • Need to create value, not assess technology
  • POC without constraint are not usually transformed
• Quote “As for the analytics transform really the processes, the organisation needs to change”

Leasons learnt from an IoT project, by PwC (Day 2 at 11:00AM)

• Goal: to have the pulse on the customer in continuous
• Challenges:
  • #1: manage complexity of an IoT workflow
  • #2: Scale up (from 500 devices to 100k devices)
• Responses:
  • Set-up of a monitoring solution (see PwC control room)
    • Detect trend
    • Set-up of a Data Lab to be used by the control room
  • Train and support of the build and support team by subject exports (such as IoT, Big Data, and Cloud)
• Profiles to address in such projects: end-user, support, manager
• Impacts from audit and risk management

Exhibits

There were 400+ exhibits. Below are a couple of concepts I spotted (non exhaustive):

• DataValue: A marketing concept of Audit IMMA (Information Management Maturity Audit)
• Data Altares: has a solution around payment recovery
• SoftComputing: focus on Marketing data
• Xebia: Kit to work on use cases around data analysis