kivi innovation drinks twente an ai encounter with …...braincreators applies 20+ years of...
TRANSCRIPT
KIVI Innovation Drinks Twente
An AI encounter with BrainCreators
Overview
➢ Short introduction BrainCreators➢ Example use-cases➢ AI maturity model➢ The Application Gap➢ Epilogue: pretty pictures
Discover value Deploy solutions Accelerate teamsCompile a strategic roadmap of viable
business cases
Implement scalable solutions with maximum
business impact
Inherit skills & best practices with
expert coaching
BrainCreators applies 20+ years of experience in artificial intelligence to business challenges across all verticals
TRUSTED BY
Use cases
➢ Smart Radio➢ Logistics➢ Fashion and Retail➢ Steel quality control➢ Genetics ➢ Telecommunication
Smart Radio
● 10 hours original radio broadcast per day ● Additional podcast creation● Inconsistent manual tagging of content● Course segmentation of topics
24/7 news radio● Automatically curated playlists of news content● Taylored to a listener’s preferences● On demand
Smart Radio
Smart Radio
● Audio feature detection● Detection of semantic overlap among existing labels● Semi-supervised refinement of existing dataset● Topic modeling● Segment classification
AI under the hood
Smart Radio
● Detect topics in segments● Cut audio segments ● Provide relevant user content
Results (v1 in the making)
Logistics
● Manual identification of address data for 15% of total volume● 4% delivered to wrong address● Geographical location of delivery points imprecise● Delivery window too coarse
Before application of AI
Logistics
● Fuzzy logic address matching● GPS delivery point prediction● Time window estimation & optimisation● Automated location mapping (inc. po-boxes)● Trained on historic data and self learning
AI under the hood
Logistics
● Manual correction reduced to <2% of total volume● Delivery failures reduced by 50%● 2000 man hours saved per month● Improved customer service through better time windows
Results
Fashion & retail
● Manual classification of products● Complex mappings to market place taxonomies● Poor quality of properties data● Basic recommendations
Before
Fashion & retail
● Training set of 20+ Million products● Combined Image & Text classifiers● Sorting of products using complex features● Human-in-the-loop data improvement
Under the hood
Fashion & retail
● Automated categorization >95% accurate● Auto-enrichment of product data● Product family recommendation● Cross & upselling automation
Result
● A major European steel producer● Total of 7.1 million tonnes of steel
products in 2016● High quality sheet and strip steel● Automotive, packaging, and construction
sectors
General
Steel quality control
● Kilometers of steel sheet each day● Accurate quality assessment enables
more profitable trading● Defects need to be detected to prevent
machine breaks● Manual inspection supported by
automatic camera system
Initial Situation
Steel quality control
● Infrared cameras inspect moving steel sheets on conveyor belts
● Basic image processing detects regions of interest
● Manual inspection often needed● Accuracy can still be improved
Camera system
Steel quality control
● Up to 50 different defect types● 5 million (!) new images each day● Currently only 25 thousand annotated
images available in total● Severely imbalanced data sets● Manual annotation is costly
Data sets
Steel quality control
● Deep Learning for robust image classification
● Ai & Active Learning approach for efficient image annotation
● Integration in existing systems● Knowledge transfer to customer’s own
tech team
Solution
Steel quality control
Genetics for livestockthe animal protein value chain
Genetics for livestockthe animal protein value chain
● Large scale selective breeding as an industrial optimization process
● Changing targets due to commercial, political, and environmental requirements
● Integration of different data sets, including genomics data
● Evaluation is either slow or imprecise
Selective breeding
Genetics for livestock
Breeding value = Genetics + Environment
● Predict carcass properties from measurements on live animals
● Numerical input data, e.g. weights at different ages
● Ultrascan visual data● How can the ultrascans be used more
effectively ?
Challenge
Genetics for livestock
● Information can get lost in the narrow passage of human interpretation
● Deep learning helps to extract useful information from complex visual data
● Less requirements for human understanding of the images
● End-to-end learning combines visual and non-visual data into one system
Narrow passage
Genetics for livestock
Human understanding Deep Learning
● Very large telco network
● Hybrid Fibre Coax● Up to 5M modems● Modems report their status
● Thousands of relay points● Diversity of legacy systems
Initial situation
Fault detection in telecom
…..
● Manage fleet of field technicians● Network errors and maintenance
● Detect & classify problems● Find problem root causes
● Collect useful data
Challenge
Fault detection in telecom
● Dedicated data labeling software● Network Anomaly Detection
● Generalize to all fault types● AI Roadmap
Solution
Fault detection in telecom
Human understanding Deep Learning
TRANSFORMING TO A DIGITAL ENTERPRISE
Exploring
TRANSFORMING TO A DIGITAL ENTERPRISE
Exploring Planning
TRANSFORMING TO A DIGITAL ENTERPRISE
Exploring Planning
Experimenting
TRANSFORMING TO A DIGITAL ENTERPRISE
Exploring Planning
Experimenting
Productizing
TRANSFORMING TO A DIGITAL ENTERPRISE
Exploring Planning
Experimenting
Productizing
Scaling
TRANSFORMING TO A DIGITAL ENTERPRISE
Exploring Planning
Experimenting
Productizing
Scaling
Data-Centric
TRANSFORMING TO A DIGITAL ENTERPRISE
The Application Gap
The Application Gap… between Research and Industry
The Application Gap… between Research and Industry
● When is something “solved” ?● Has it been demonstrated to work once,
under special circumstances?● Or is it ready and safe to deploy in
general, right now, for everyone ?
● Speech-to-text ? ● Self-driving cars ? ● … …
Solved? .... really?
Andrew Ng: “AI is the new electricity”
“We have enough papers. Stop publishing, and start transforming people’s lives with technology!”
The competitive landscape
Thank you!Maarten Stol [email protected]
BrainCreators Prinsengracht 697 1017JV Amsterdam+31 (0)20 369 7260
Epilogue
BigGAN
Large Scale GAN Training for High Fidelity Natural Image SynthesisAndrew Brock, Jeff Donahue, Karen Simonyan
BigGAN
Large Scale GAN Training for High Fidelity Natural Image SynthesisAndrew Brock, Jeff Donahue, Karen Simonyan
BigGAN
Large Scale GAN Training for High Fidelity Natural Image SynthesisAndrew Brock, Jeff Donahue, Karen Simonyan
BigGAN
Large Scale GAN Training for High Fidelity Natural Image SynthesisAndrew Brock, Jeff Donahue, Karen Simonyan
BigGAN
Large Scale GAN Training for High Fidelity Natural Image SynthesisAndrew Brock, Jeff Donahue, Karen Simonyan
BigGAN
Large Scale GAN Training for High Fidelity Natural Image SynthesisAndrew Brock, Jeff Donahue, Karen Simonyan
BigGAN
Large Scale GAN Training for High Fidelity Natural Image SynthesisAndrew Brock, Jeff Donahue, Karen Simonyan
BigGAN
Large Scale GAN Training for High Fidelity Natural Image SynthesisAndrew Brock, Jeff Donahue, Karen Simonyan
BigGAN
Large Scale GAN Training for High Fidelity Natural Image SynthesisAndrew Brock, Jeff Donahue, Karen Simonyan
BigGAN
Large Scale GAN Training for High Fidelity Natural Image SynthesisAndrew Brock, Jeff Donahue, Karen Simonyan
BigGAN
Large Scale GAN Training for High Fidelity Natural Image SynthesisAndrew Brock, Jeff Donahue, Karen Simonyan
BigGAN
Large Scale GAN Training for High Fidelity Natural Image SynthesisAndrew Brock, Jeff Donahue, Karen Simonyan
BigGAN
Large Scale GAN Training for High Fidelity Natural Image SynthesisAndrew Brock, Jeff Donahue, Karen Simonyan
BigGAN
Large Scale GAN Training for High Fidelity Natural Image SynthesisAndrew Brock, Jeff Donahue, Karen Simonyan
BigGAN
Large Scale GAN Training for High Fidelity Natural Image SynthesisAndrew Brock, Jeff Donahue, Karen Simonyan
BigGAN
Large Scale GAN Training for High Fidelity Natural Image SynthesisAndrew Brock, Jeff Donahue, Karen Simonyan
BigGAN
Large Scale GAN Training for High Fidelity Natural Image SynthesisAndrew Brock, Jeff Donahue, Karen Simonyan
BigGAN
Large Scale GAN Training for High Fidelity Natural Image SynthesisAndrew Brock, Jeff Donahue, Karen Simonyan
BigGAN
Large Scale GAN Training for High Fidelity Natural Image SynthesisAndrew Brock, Jeff Donahue, Karen Simonyan
BigGAN
Large Scale GAN Training for High Fidelity Natural Image SynthesisAndrew Brock, Jeff Donahue, Karen Simonyan
BigGAN
Large Scale GAN Training for High Fidelity Natural Image SynthesisAndrew Brock, Jeff Donahue, Karen Simonyan
BigGAN
Large Scale GAN Training for High Fidelity Natural Image SynthesisAndrew Brock, Jeff Donahue, Karen Simonyan
BigGAN
Large Scale GAN Training for High Fidelity Natural Image SynthesisAndrew Brock, Jeff Donahue, Karen Simonyan
BigGAN
Large Scale GAN Training for High Fidelity Natural Image SynthesisAndrew Brock, Jeff Donahue, Karen Simonyan
GAN Progress
Source: www.eff.org/files/2018/02/20/malicious_ai_report_final.pdf
Source: PROGRESSIVE GROWING OF GANS FOR IMPROVED QUALITY, STABILITY, AND VARIATION Tero Karras, et.al. NVIDIA
Source: PROGRESSIVE GROWING OF GANS FOR IMPROVED QUALITY, STABILITY, AND VARIATION Tero Karras, et.al. NVIDIA