Human Brain Project Unifying our understanding of the human brain
Co-funded by the European Union
Marc-Oliver Gewaltig Ecole Polytechnique Fédérale de Lausanne, Blue Brain Project, Neurorobotics Disruptive tools and technologies…
April 25, 2017 12 … empower Information Technology industry and/or give rise to new Medicine industries (after – sometimes – a long delay!)
Biotech
April 25, 2017 13 The Vision of the Human Brain Project
R. Feynman : What I cannot create I do not understand
To understand the brain (better) we need a
• large-scale, interdisciplinary, integrating infrastructure
• for performing multi-level studies of brain and body
• from analytics and neuroscientific data by way of synthetic modeling for partial/full brain simulation, brain reconstruction,
• and the design of new computer architectures and robots.
HBP is a European FET Flagship project to create and operate collaborative research tools for experimental and virtualized brain research, and for developing brain-derived technologies.
April 25, 2017 14 HBP at a Glance – Facts and Figures
• 10-year, EUR 1 billion Research Roadmap (50% Core Project, 50% Partnering Projects) • Core project 400+ scientists, 116 institutions, 19 countries • 6 prototype research platforms released in March 2016 • Embedded in previous and existing national and international initiatives: Blue Brain, BrainScaleS, Supercomputing and Modeling the Human Brain, SpiNNaker, PRACE, etc. • 23 industry collaborations; 121 research collaborations with non-HBP research groups (61 with universities and institutes in 3rd countries)
April 25, 2017 15 Research Branches within the Human Brain Project
Accelerated Medicine Contribute to understanding, diagnosing and treating diseases of the brain.
Accelerating Neuroscience Accelerated Future Computing Integrate everything we know about the Learn and derive from the brain to build the brain into computer models and supercomputers and robots of tomorrow. simulations.
April 25, 2017 16 The HBP Platform Universe supports the science Brain Simulation: In-silico behavior Collaborative integration of neuroscience data into and cognition multi-scale scaffold models and simulations of brain Neurorobotics regions
Neurorobotics: Scaffold Models Medical Informatics Testing brain models and simulations in
dynamic virtual environments Medicine Mouse Human Neuroinformatics: Organizing neuroscience data, mapping to brain Brain Simulation atlases
Medical Informatics:
Neuroscience Bringing together information on brain diseases Neuroinformatics
Neuromorphic Computing: ICT that mimics the functioning of the brain High Performance Analytics constraints andComputing Computing, High Performance Analytics and Computing: predictions Neuromorphic Computing Hardware and software to support the other capability Platforms
April 25, 2017 17 Neuroinformatics Platform
Brain Atlases
April 25, 2017 18 Brain Simulation Platform
Detailed reconstruction and simulation of brain regions
Markram et al., 2015
April 25, 2017 19 High Performance Analytics and Computing Platform
Simulation technology Extending the functionality of brain simulation codes: concepts, numerical algorithms and software technologies Data-intensive supercomputing Linking extreme scale data processing challenges to the exploitation of scalable computer resources Interactive visualization Visual analysis of large-scale neural simulation data Dynamic resource management Novel approaches for managing the resources in a supercomputer across applications
April 25, 2017 20 Neuromorphic Computing Platform
SpiNNaker BrainScaleS / HICANN
Many-Core Machine Physical Model Machine Base Chip with stacked DRAM Base Chip 18 Cores 512 Neurons 115k Synapses
Intel Free Press, CC BY-SA 2.0
April 25, 2017 21 Neurorobotics Platform
Closed-Loop Simulation of Soft Biological Bodies: The HBP Mouse
Mouse body Simulated activation of S1 with tactile stimulation
Trunk
Hindlimbs Forelimbs Whiskers
Mouth Nose
April 25, 2017 22 Project Timeline
Submission of HBP is selected as one Passed review FET proposals of two FET Flagship successfully for pilots projects by the European Commission Transformation into a European Research Infrastructure Start of the ramp-up phase New Users
Pilot Ramp-Up Phase Operational Phase
2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023
Public New Collaborators Start of platform HBP pilot release Passed 6 pilots review Ethics Advisory selected successfully Board formed
April 25, 2017 23 Summary
• HBP is a European Flagship project that builds an integrated ICT-based research infrastructure for brain research, cognitive neuroscience and brain-inspired computing: • Gather, organise and disseminate data describing the brain and its diseases • Build multi-scale scaffold models and theory for the brain • Simulate the brain • Develop brain-inspired computing, data analytics and robotics • Ensure that the HBP's work is undertaken responsibly and that it benefits society
• The project promotes collaboration across the globe.
• The next HBP Summit will be held in Glasgow in October 2017
April 25, 2017 24 Video
Link
April 25, 2017 25
Deloitte Digital Series April 2017
29 James mission is
To bring the best of data science to every risk office
Artificial Intelligence
30 Artificial Intelligence is a reality today
31 So why isn’t AI into credit risk?
32 Well, because....
1. It is a time 3. Cannot rely 2. Regulatory 4. It is hard to find consuming and on black constraints the right talent complex process boxes
33 What is the solution?
Easy to use by experts Regulation Ready
Easy to integrate
Credit Risk AI
Some banks are already building their own AI!
34 The time is now
Higher cost of risk
This is where we Lose market share
are today Adoption of Artificial intelligence in credit risk in credit intelligence Artificial of Adoption
time Adoption Pioneers Early adopters Early majority Late majority Late tiers
Effort to catch up increases as adoption spreads
35 James is the first credit risk AI and he helps risk officers
James is your credit risk AI.
2. He is compliant with 1. He automates time 3. He provides you with 4. He is proactive, Basel Committee consuming processes intelligible state-of-the-art autonomous and directives algorithms easy going
(plays well with other softwares and platforms)
36 James brings the best of data science to every risk team
1. State of the art 2. Basel-compliant 3. Seamless model 4. Proactive model modeling techniques validation reports deployment monitoring
37 James has a recognized experience in the credit risk space Runs & Operates successfully at:
Gini Index (discriminating Benchmark capacity) Using James
70
60
50 Tested with positive results at:
40 Lender 1 Lender 2 Lender 3
630 mln AUM 842 bln AUM 4258.3 mln AUM Default Rate: 2% DR: 11.13% DR: 1.77%
38 What results can you obtain?
Team of experts in James provides Results obtained Artificial intelligence provides
1. State of the art classification algorithms Best machine learning credit Increase acceptance rate risk support 2. Best optimization and up to 10% validation techniques On-demand data cleansing
3. Easy model management On going analysis of 4. Automate model validation Reduce default rate monitoring alerts Up to 30% 5. Automated performance On-demand reporting reporting
39 The results speak for themselves
Goal: To decrease the default rate without impacting the acceptance rate.
Default rate Default rate incumbent model James model
2,69% 2,44% Potential upside per year
1.5M (aprox.) Reduction in default rate
9,3%
40 james.finance
New York +1 (347) 305-9110 London +44 20 3287 4132 Lisbon +351 912 250 990 João Menano [email protected]
41
Henry White @pixoneye CONTENTS
Intro Deep Learning and evolution of Computer Vision 1
Current user understanding on mobile devices 2
Pixoneye’s Computer Vision capabilities 3
Pixoneye’s Product solutions and Use cases 4 LANDSCAPE AI 1950’s a broad concept established - can machines one day think MACHINE like humans? LEARNING DEEP one path of AI, rather than trying to hard code or LEARNING develop a theoretical is a branch of machine learning based model teach by example on a set of algorithms that attempt to model high level abstractions in data
deep learning is the primary driver and the most important approach to AI and will drive enterprise
1950 1960 1970 1980 1990 2000 2010 2020 COMPUTER VISION
Computer Vision
The aim is to imitate the functionality of human eye and brain components responsible for your sense of sight
This can provide essential data to process, analyse and utilise in fields ranging from transport to facial recognition to marketing [2008] Image Detection [2010] Image Recognition [2013] Image Understanding
What’s in the image: Friends, skiing, on the slopes, snow, outdoors
Tags: Snow; Winter; Skiing; Couple; Friends; Mountains; Holliday. [2015-2016] CONTEXTUAL UNDERSTANDING
Demographics: Lifestyle: Location: London Past-time: Cycling – 30% | Hiking – User: Male 25-30 30% |Rugby – 30%| BBQ – 10% Marital Status: Engaged Fashion: Casual Family: children 0 Income level: 5/6 (0-8) Relationship: Female 25-30 Pets: Dog – 1 Breed: Great dane Work environment: Interests: Cycling, Beach, Rugby, business/Casual Friends, Social events. Relevant details: Traveler, young couple, outdoor lifestyle THE CURRENT MOBILE MARKETING PROBLEM
NO ONE KNOWS THEIR MOBILE CUSTOMERS…
…AND THEREFORE, NO ONE CAN TARGET OR RECOMMEND TO THEM EFFECTIVELY UNDERSTANDING CUSTOMERS
81% 22% Of companies say they v Of consumers on mobile have a holistic view of say the average retailer their mobile customers understands them as an individual PERSON PERSON 1 2 ? ? 54
What You Know
Male Married twice Grown Children Young Grand Children
English Countryside Holiday in Alps Extensive Travellers
Born 1948 Dog Lovers Sports Cars Fanatics
Wealthy
Why personal galleries? It is effectively a data set along a timeline
People take >250 photos each month
The average camera 1,500 photos and 24 videos
OFFLINE understanding documents real life <2% of images are shared on social media HOW IT WORKS…
4 1 2 3
Contextual 6 5
4 Understanding 3 2
Amount Amount of images of personal 1 0 0 50 100 150 200 250 300 350 400 galleries Categories Engaged
Skier
Cyclist
Dog owner FEATURE VECTOR
Engaged Skier Cyclist Dog Owner
6
5
4
3
2
Amount Amount of images 1
0 0 50 100 150 200 250 300 350 400 Categories 3 Main products 1 Analytics:
150 Characteristics 2 Recommendation Engine Case Study - Advertising Optimization Capability
“Before meeting Pixoneye we typically generated sales from 4 in 1,000 digital ad impressions by targeting pet owners. Using the Pixoneye technology, we generated 40 in 1,000 digital impressions by targeting cat and dog owners specifically.” 3 Triggers: Life changing 6
5
4 3 events 2
Amount Amount of images 1
0 0 50 100 150 200 250 300 350 400 Categories ACCURATE DECISION MAKING USING PIXONEYE’S AI Case Study 1 92%
User User 1 User 2 User 3 NEW
6
5
4
3
2
Amount Amount of images 1
0 0 50 100 150 200 250 300 350 400 Categories Thank You
Henry White @pixoneye