SlideShare ist ein Scribd-Unternehmen logo
1 von 76
Downloaden Sie, um offline zu lesen
Confidential and copyright of Somo Custom Ltd. June 23 1
Solutions for the connected world
Confidential and copyright of Somo Custom Ltd. June 23 2
Somo accelerates mobile transformation
through rapid innovation to create
products and experiences your
customers and employees will love.
Confidential and copyright of Somo Custom Ltd. June 23 3
Our agile approach to transformation
AmplifyExecute & IterateInnovate
Product
& UX
workshops
Proof of
concept
Scaled
global
launch
Optimise
& iterate
Minimum
lovable
product
Owned, earned,
& paid media
Productise
Maintain,
Scale &
support
Strategic
vision
& insight
Confidential and copyright of Somo Custom Ltd. June 23 4
Transforming
Live
Transforming
Engagement
Transforming
Content
Confidential and copyright of Somo Custom Ltd. April 16 5
Global client experience
Finance
Retail & FMCG
Automotive Publishing TMT
Utility &
Government
London Bristol NYC
A selection of our success
✓ Audi e-tron pop up experience in London exceeded lead generation target by 223%
✓ The Wall Street Journal What’s News app ranked #2 in the App Store news category
✓ Achieved an ROI of 18:1 with Very.co.uk’s multi-channel 2015 Christmas campaign
6Confidential and copyright of Somo Custom Ltd. June 23
Global partnerships with industry leaders
7Confidential and copyright of Somo Custom Ltd. June 23
Digital experience Physical world
Multiple
screens
Mixed
realities
Interface
Internet
of things
Somocorefocus
Desktop
360˚
Tablet Mobile Wearable
Virtual reality Augmented reality
Touch Voice Gesture
Connected car
Connected city
Connected home
Connected retail
Connected fitness
Messaging
Machine Learning
Biometrics
8Confidential and copyright of Somo Custom Ltd. June 23
Innovation focus: what’s next?
Our values
Create success
Be brave
Lead with knowledge
Love innovation
Confidential and copyright of Somo Custom Ltd. June 23 10
The Singularity is Near
Ruben Horbach, Senior Innovation Manager
Somo
Dave Evans, CTO
Somo
George Whitelaw, CTO
Visii
Presentations
Messaging App
Fragmentation
Deep Learning
Andrew Wyld, Technical Architect
Somo
Machine Learning
Confidential and copyright of Somo Custom Ltd. April 16 11
The Singularity is Near
Ruben Horbach - Senior Innovation Manager
Confidential and copyright of Somo Custom Ltd. June 23 12
Confidential and copyright of Somo Custom Ltd. June 23 13
Confidential and copyright of Somo Custom Ltd. June 23 14
Check-ins Payments Events
NFC use cases
Confidential and copyright of Somo Custom Ltd. June 23 15
Coca-Cola Samsung Burberry Nokia
NFC use cases
Confidential and copyright of Somo Custom Ltd. June 23 16
“The allure of NFC is its simplicity”
Why NFC?
Confidential and copyright of Somo Custom Ltd. June 23 17
“Traditional”
Confidential and copyright of Somo Custom Ltd. June 23 18
NFC implants
Confidential and copyright of Somo Custom Ltd. June 23 19
Dangerous Things
Confidential and copyright of Somo Custom Ltd. June 23 20
Slightly painful..
Confidential and copyright of Somo Custom Ltd. June 23 21
Confidential and copyright of Somo Custom Ltd. June 23 22
Different possibilities
Confidential and copyright of Somo Custom Ltd. June 23 23
Future possibilities
Confidential and copyright of Somo Custom Ltd. June 23 24
Innovation = collaboration
Confidential and copyright of Somo Custom Ltd. June 23 25
Confidential and copyright of Somo Custom Ltd. June 23 26
This is actually quite common
Confidential and copyright of Somo Custom Ltd. June 23 27
Welcome to the future
Proteus ingestible sensor Google glucose contact lens e-Dura implant
Confidential and copyright of Somo Custom Ltd. June 23 28
Wolverine?
Anatomics 3D printed Titanium ribs
Confidential and copyright of Somo Custom Ltd. June 23 29
Biology & Technology in 30 years
Ray KurzweilNicholas Negroponte
Confidential and copyright of Somo Custom Ltd. June 23 30
Today
Confidential and copyright of Somo Custom Ltd. June 23 31
Source: Peter H. Diamandis M.D. - Singularity University 2016
Exponential predictions
10E-10
10E-5
10E0
10E5
10E10
10E20
1900 2000 2100
10E25
10E30
10E35
10E40
10E45
10E50
10E55
10E60
2010
10e11
2023
10e16
2050
10e26
Calculations per second per $1000 vs. Time
Confidential and copyright of Somo Custom Ltd. June 23 32
Tomorrow?
• Physical	
  world	
  interface	
  
• Virtual	
  world	
  interface	
  
• Cognitive	
  interface
Confidential and copyright of Somo Custom Ltd. April 16 33Confidential and copyright of Somo Custom Ltd. June 23 33
Machine Learning
Andrew Wyld - Technical Architect
Machine Learning
Are you Sarah Connor?
My name is Siri and I have bad news.
Buckle up
buttercup
this will go fast
In at the deep end
Deep learning gets a lot of headlines for super cool applications:
Image recognition
Speech recognition
Language processing
“Shallow” learning is still really useful and easier to apply:
Basically statistical techniques
Requires a little cleverness to handle nonlinear data
Nevertheless still very powerful and way easier to train.
Deep learning is based on these simpler techniques, so best to start there.
Deep learning is just shallow learning several times in a row anyway.
There’s a super cool hybrid that gets some of the advantages of both ….
Supervised vs
unsupervised
Don’t tell me what to do!
Supervised learning requires the
machine to be taught. This is good
for situations where there’s a known
right answer.
Unsupervised learning throws the
machine in among the data and lets
it look for patterns by itself.
Supervised vs unsupervised
Supervised learning looks for relationships in
labelled data. Data is separated into “inputs”
and “outputs”, where you want to predict the
outputs from the inputs.
Unsupervised learning looks for patterns in
unlabelled data. All of the data is “input” data;
the outputs are any patterns found by the
algorithm.
Linear
regression
The little statistical analysis technique
that could
Linear regression is a great place to
start. If you have a bunch of data
points, you try to fit a straight line to
them.
Data that don’t fit a straight line can
be handled by using functions of
the data that do fit a straight line.
Linear Regression: improving the fit (1)
This line is visibly not a great fit for
the data. The error lines are pretty
long and asymmetrical.
We aim to minimize squared error,
as this prevents positive and
negative errors cancelling out.
Linear Regression: improving the fit (2)
This line is clearly a lot better. The
data fits the line pretty well.
There are several algorithms to find
the best fit for a linear regression.
This is basically the simplest
machine learning system there is,
but it’s still very useful for continuous
data!
Classifiers
There are two types of people in the
world: those who like binaries and a
continuum of others.
Classifiers are a huge category of
machine learning system. Actually
most systems are some kind of
classifier, including deep learning
systems.
Classifiers split things into categories.
Here we have a set of labelled data.
A classifier is an attempt to separate
positive ▪ from negative ▪ data,
and predict whether new data will be
positive or negative.
There are several methods of
classification, but they all essentially
aim to draw this line.
Logistic regression: best fit for definite people
Logistic regression is
similar to linear
regression, but where
linear regression tries to
find a line that fits
continuous data well,
this method tries to fit a
logistic function (which
has a suitable sigmoid
curve) to a set of “true/
false” data.
Support vector machines: best fit for very definite people
A support vector
machine is very similar
to logistic regression, but
has a simpler function
that heavily penalises
errors in a wide margin,
so the algorithm will try
hard to avoid putting
points there. It’s
sometimes known as the
wide margin classifier,
for this reason.
Underfitting: the model is stupid
A model is said to underfit when it’s
too simple to capture something
important about the data. Very
commonly data won’t exactly fit a
linear model. A more complex model
is needed to fit the data well.
Underfitting can’t be fixed by better
data: no amount of training can bend
that straight line round a curve.
Overfitting: the model is neurotic
On the flip side, a model can fit the
data so well—hugging every tiny
crevice—that it generalises poorly.
A high-dimensional model will tend
to overfit. The advantage of an
overfitting model over an underfitting
one is that more data can usually
cure the problem, as random
wobbles in the data eventually
cancel each other out.
One-vs-all
One against all and all against one!
And every other one against every
other all.
Lots of classifications need more
than two categories. The usual way
to handle this is “one-vs-all”
classification: train one classifier for
every category, then predict new
results using the classifier that is
“most sure” of the ones you’ve
trained.
One-vs-all classification
In a one-vs-all classifier, as many
classifiers are trained as there are
categories. Predictions are then
based on how confident each
classifier is, with the most-confident
classifier winning.
Deep learning
Hugely powerful. Nobody knows
what’s inside it.
The “deep” in deep learning refers
to the fact that several classifiers
are stacked, one in front of the
other. Each one can learn more
sophisticated things by building on
the previous layer.
Nobody really knows what goes on
in the middle layers (although we’re
beginning to research it).
Stack high the classifiers!
A neural network is just a
sequence of classifiers
in a stack. Each layer
can use the output of the
previous layer as input;
thus, by the end,
features can be very
sophisticated, based on
complex combinations of
other, simpler features.
The hidden layers make
the technique powerful
but inscrutable.
Back propagation
Each output is compared
to training data and
scored. Paths that led
from the previous layer
to that output are
strengthened or
weakened depending on
the score.
The scores are then
passed backwards along
the pathways and the
process repeated.
Transfer learning
The early layers of (for example) a cat
recognition system will probably pick up
general image features—corners, colour
transitions, diagonals and so on—that would
be useful for any image recognition task.
If you want to make a dog recogniser but don’t
have a lot of data, you could simply cut off the
last layer, steal these early features, keep
them the same, and glue a simple classifier on
the end in place of the old last layer.
This works surprisingly well.
Any questions?
Andrew Wyld
andrew.wyld@somoglobal.com
@Andrew_Wyld
Cool stuff: a very non-exhaustive list
Stanford machine learning course

https://www.coursera.org/learn/machine-learning/
did it, loved it.
University of Washington machine learning specialisation

https://www.coursera.org/specializations/machine-learning/ doing it now.
Tensorflow online neural network

http://playground.tensorflow.org/
have fun!
Google/Udacity deep learning course

https://www.udacity.com/course/deep-learning--ud730 want to do
it!
You can get this slide deck here.
Andrew Wyld
andrew.wyld@somoglobal.com
@Andrew_Wyld
https://docs.google.com/presentation/d/1K9owIkpuneAtuaqTguQ5gM7nboK-PWZhkwf3czaFN0M/pub
Confidential and copyright of Somo Custom Ltd. April 16 58Confidential and copyright of Somo Custom Ltd. June 23 58
A quick dive into Deep Learning
George Whitelaw
59
60
Overview
•Why complex problems require machine learning
•What is a Neural Network
•Solving complex problems
•Deep learning in daily life
61
Why complex problems require machine
learning
•We have an ever growing amount of

information that needs to be understood,

often on demand.
•The problems are getting more complex.
•Machine learning has been around since

the 60s, many methods won’t cut it.
Source: Deep Learning in a Nutshell – what it is, how it works, why care? by Nikhil Buduma
62
Why we need Neural Networks
Teaching computers rules (heuristics) takes

time, is error prone and generally sucks.



What is this?
Source: Deep Learning in a Nutshell – what it is, how it works, why care? by Nikhil Buduma
What is a neural network
63
Source: Neural Networks, Manifolds, and Topology by Christopher Olah

Image source: Wikimedia
•We are good at classifying things.
•Neural networks simulate (crudely) the human
brain.
•They require training on test data to give useful
Output - was it 6 or 0?
•Complex problems require deeper networks.
What is a neural network
64
Source: Neural Networks, Manifolds, and Topology by Christopher Olah

Image source: Wikimedia
Dataset

Learn where a point belongs on a line
Without a NN

Pretty rubbish
With a NN

Better
What is a neural network
65
Source: Neural Networks, Manifolds, and Topology by Christopher Olah

Image source: Wikimedia
•The hidden layer represents the dataset in a
way that clearly separates a decision.

•Complex input requires more layers.
Solving complex problems
66Source: FaceNet: A unified Embedding for Face Recognition and Clustering by Google Inc
•Deeper networks can detect more interesting
features.

e.g. Faces grouped by individual features.
Deep learning in daily life
67
•Deep learning helps to classify and organise
overwhelming amounts of data.
•New technologies use Deep learning to help save
time, reduce decision fatigue and generally make life
more simple.
Stay connected
68
Visii
Rainmaking Loft - International
House
1 St Katharine’s Way
London, E1W 1UN
Website: www.visii.com
George Whitelaw (CTO)
Email: george@visii.com
Mobile: + 44 797 623 9524
AddressContact Info
69
TIME TO EXPLORE
Messaging App Fragmentation
Dave Evans - CTO
Reaching end users : Today
Ads targeted on keywords

or interests
Click to landing

page or app 

store
Advertising Medium is

HTML/CSS/Jscript
Landing page is web page
Served by enterprise or
Interim step in App Store 

plus an app built by enterprise
71Confidential and copyright of Somo Global Ltd. June 23
Messaging Platforms and Chat Bots
• Messaging apps are becoming
platforms – with vast numbers of users
• Offering developer API’s providing
ability to interact with end users –
typically through send/receive/
subscribe API’s
• Complex conversational flows enables
enterprise to lead the end user
• Alert based and long running
conversations enabled
June 23 72
Reaching users tomorrow
Confidential and copyright of Somo Global Ltd.
Confidential and copyright of Somo Global Ltd. June 23 73
Chat Architecture #1
Response
Formatting
Application Logic +
Language Processing
Data
Store
Message
Receipt
Session/Conversation
Management
Message App eg FB
Messenger,
Whatsapp
HTTP/s Comms
Confidential and copyright of Somo Global Ltd. June 23 74
Chat Architecture #2
Application Logic +
Language Processing
Data
Store
Apple iMessage
HTTP/s Comms
Response
Formatting
Session/Conversation
Management
Message
Receipt
Message App
Confidential and copyright of Somo Global Ltd.
• Messaging apps will need to be built to support specific messaging platforms
• Architect the business logic and machine intelligence on the back end to support multiple platforms
• Separate out the request/response presentation capabilities into separate layers/plug-ins
• Apples iMessage requires specific Messaging apps to be built and deployed
• Message recipient requires that app on their phone – or flow will be interrupted as they download
(or not) the app
• No support for Android – so audience is constrained to iMessage users
• No standard for cross platform messaging/formatting
• One of reasons for success of SMS was the adoption of standards across handsets and carriers
However potential for rich dialog with end users, and massive 

distribution / reach when apps are done well.
June 23 75
Messaging Apps Fragmentation
Confidential and copyright of Somo Custom Ltd. April 16 76Confidential and copyright of Somo Custom Ltd. June 23 76

Weitere ähnliche Inhalte

Was ist angesagt?

2017 year in review: MOBILE
2017 year in review: MOBILE2017 year in review: MOBILE
2017 year in review: MOBILEMonika Mikowska
 
London Tech Week - Innovation within Business
London Tech Week - Innovation within BusinessLondon Tech Week - Innovation within Business
London Tech Week - Innovation within BusinessSomo
 
FIDM SF | 8th Annual Innovative Materials Conference | Huami / Frederik Hermann
FIDM SF | 8th Annual Innovative Materials Conference | Huami / Frederik HermannFIDM SF | 8th Annual Innovative Materials Conference | Huami / Frederik Hermann
FIDM SF | 8th Annual Innovative Materials Conference | Huami / Frederik HermannFrederik Hermann
 
Mobile applications
Mobile applicationsMobile applications
Mobile applicationsShweta Jain
 
Mobile World Congress 2013: A report from the floor
Mobile World Congress 2013: A report from the floorMobile World Congress 2013: A report from the floor
Mobile World Congress 2013: A report from the floorDMI
 
Wap vs App - marketing magazine "future of mobile event" oct 2010
Wap vs App - marketing magazine "future of mobile event"  oct 2010Wap vs App - marketing magazine "future of mobile event"  oct 2010
Wap vs App - marketing magazine "future of mobile event" oct 2010Phil Barrett
 
The best of mobile marketing 2014
The best of mobile marketing 2014The best of mobile marketing 2014
The best of mobile marketing 2014Monika Mikowska
 
Mobile marketing
Mobile marketingMobile marketing
Mobile marketingyourdigital
 
Mobile marketing
Mobile marketingMobile marketing
Mobile marketingyourdigital
 
Predictions_2016_The_Mobi__1_
Predictions_2016_The_Mobi__1_Predictions_2016_The_Mobi__1_
Predictions_2016_The_Mobi__1_Tony Fanelli
 
Cleartrip mobile Q2'14 Infographic : How India Travels Thrugh Mobile
Cleartrip mobile Q2'14 Infographic : How India Travels Thrugh MobileCleartrip mobile Q2'14 Infographic : How India Travels Thrugh Mobile
Cleartrip mobile Q2'14 Infographic : How India Travels Thrugh MobileNextBigWhat
 
Mobile Megatrends 2008 (VisionMobile)
Mobile Megatrends 2008 (VisionMobile)Mobile Megatrends 2008 (VisionMobile)
Mobile Megatrends 2008 (VisionMobile)guest94da57
 
Vietnam mobile web and mobile app market overview
Vietnam mobile web and mobile app market overviewVietnam mobile web and mobile app market overview
Vietnam mobile web and mobile app market overviewHien NT
 
5 Key Takeaways For Brand Marketers - Mobile World Congress - Day 2
5 Key Takeaways For Brand Marketers - Mobile World Congress - Day 25 Key Takeaways For Brand Marketers - Mobile World Congress - Day 2
5 Key Takeaways For Brand Marketers - Mobile World Congress - Day 2Gemma Craven
 
Mobile Megatrends 2014
Mobile Megatrends 2014Mobile Megatrends 2014
Mobile Megatrends 2014SlashData
 
OGDC 2014_Vietnam Mobile Internet 2014: A focus in smartphone game and compar...
OGDC 2014_Vietnam Mobile Internet 2014: A focus in smartphone game and compar...OGDC 2014_Vietnam Mobile Internet 2014: A focus in smartphone game and compar...
OGDC 2014_Vietnam Mobile Internet 2014: A focus in smartphone game and compar...ogdc
 

Was ist angesagt? (20)

2017 year in review: MOBILE
2017 year in review: MOBILE2017 year in review: MOBILE
2017 year in review: MOBILE
 
London Tech Week - Innovation within Business
London Tech Week - Innovation within BusinessLondon Tech Week - Innovation within Business
London Tech Week - Innovation within Business
 
FIDM SF | 8th Annual Innovative Materials Conference | Huami / Frederik Hermann
FIDM SF | 8th Annual Innovative Materials Conference | Huami / Frederik HermannFIDM SF | 8th Annual Innovative Materials Conference | Huami / Frederik Hermann
FIDM SF | 8th Annual Innovative Materials Conference | Huami / Frederik Hermann
 
Mobile applications
Mobile applicationsMobile applications
Mobile applications
 
Mobile World Congress 2013: A report from the floor
Mobile World Congress 2013: A report from the floorMobile World Congress 2013: A report from the floor
Mobile World Congress 2013: A report from the floor
 
Wap vs App - marketing magazine "future of mobile event" oct 2010
Wap vs App - marketing magazine "future of mobile event"  oct 2010Wap vs App - marketing magazine "future of mobile event"  oct 2010
Wap vs App - marketing magazine "future of mobile event" oct 2010
 
The best of mobile marketing 2014
The best of mobile marketing 2014The best of mobile marketing 2014
The best of mobile marketing 2014
 
M-Commerce 2010
M-Commerce 2010M-Commerce 2010
M-Commerce 2010
 
Mobile marketing
Mobile marketingMobile marketing
Mobile marketing
 
Mobile marketing
Mobile marketingMobile marketing
Mobile marketing
 
Predictions_2016_The_Mobi__1_
Predictions_2016_The_Mobi__1_Predictions_2016_The_Mobi__1_
Predictions_2016_The_Mobi__1_
 
Cleartrip mobile Q2'14 Infographic : How India Travels Thrugh Mobile
Cleartrip mobile Q2'14 Infographic : How India Travels Thrugh MobileCleartrip mobile Q2'14 Infographic : How India Travels Thrugh Mobile
Cleartrip mobile Q2'14 Infographic : How India Travels Thrugh Mobile
 
Mobile Megatrends 2008 (VisionMobile)
Mobile Megatrends 2008 (VisionMobile)Mobile Megatrends 2008 (VisionMobile)
Mobile Megatrends 2008 (VisionMobile)
 
Vietnam mobile web and mobile app market overview
Vietnam mobile web and mobile app market overviewVietnam mobile web and mobile app market overview
Vietnam mobile web and mobile app market overview
 
Fueled com
Fueled comFueled com
Fueled com
 
5 Key Takeaways For Brand Marketers - Mobile World Congress - Day 2
5 Key Takeaways For Brand Marketers - Mobile World Congress - Day 25 Key Takeaways For Brand Marketers - Mobile World Congress - Day 2
5 Key Takeaways For Brand Marketers - Mobile World Congress - Day 2
 
Mobile simplificado
Mobile simplificadoMobile simplificado
Mobile simplificado
 
Mobile Megatrends 2014
Mobile Megatrends 2014Mobile Megatrends 2014
Mobile Megatrends 2014
 
OGDC 2014_Vietnam Mobile Internet 2014: A focus in smartphone game and compar...
OGDC 2014_Vietnam Mobile Internet 2014: A focus in smartphone game and compar...OGDC 2014_Vietnam Mobile Internet 2014: A focus in smartphone game and compar...
OGDC 2014_Vietnam Mobile Internet 2014: A focus in smartphone game and compar...
 
Smart phones2010
Smart phones2010Smart phones2010
Smart phones2010
 

Andere mochten auch

Is your business ready for voice?
Is your business ready for voice?Is your business ready for voice?
Is your business ready for voice?Somo
 
Dualitiy in Contemporary Mathematics Wuppertal
Dualitiy in Contemporary Mathematics WuppertalDualitiy in Contemporary Mathematics Wuppertal
Dualitiy in Contemporary Mathematics WuppertalSebastian De Haro
 
Colonial operation on Robinson Crusoe
Colonial operation on Robinson CrusoeColonial operation on Robinson Crusoe
Colonial operation on Robinson Crusoebhavnabaraiya
 
Monica andrea recetor 2
Monica andrea recetor 2Monica andrea recetor 2
Monica andrea recetor 2klaumilenitha
 
5 Things : More Of & Less Of in 2014
5 Things : More Of & Less Of in 20145 Things : More Of & Less Of in 2014
5 Things : More Of & Less Of in 2014thefarmdigital
 
Emergence and Reduction in Physics
Emergence and Reduction in PhysicsEmergence and Reduction in Physics
Emergence and Reduction in PhysicsSebastian De Haro
 
VIVIANA VELOZA HISTORIA 2
VIVIANA VELOZA HISTORIA 2VIVIANA VELOZA HISTORIA 2
VIVIANA VELOZA HISTORIA 2klaumilenitha
 
Meredith M, Powerpoints lesson 1
Meredith M, Powerpoints lesson 1Meredith M, Powerpoints lesson 1
Meredith M, Powerpoints lesson 1MeredithM17
 
INTEGRATE Chicago - Joseph Truncale
INTEGRATE Chicago - Joseph TruncaleINTEGRATE Chicago - Joseph Truncale
INTEGRATE Chicago - Joseph TruncaleIMCWVU
 
3 mitchell new mm ifa 2012 v3 120529
3 mitchell new mm ifa 2012 v3 1205293 mitchell new mm ifa 2012 v3 120529
3 mitchell new mm ifa 2012 v3 120529ifa2012_2
 
Karol yuliana albarracin 1
Karol yuliana albarracin 1Karol yuliana albarracin 1
Karol yuliana albarracin 1klaumilenitha
 
CAMILA PROAÑOS REDVOLUCION HISTORIA 2
CAMILA PROAÑOS REDVOLUCION HISTORIA 2CAMILA PROAÑOS REDVOLUCION HISTORIA 2
CAMILA PROAÑOS REDVOLUCION HISTORIA 2klaumilenitha
 
CERTIFICATES FOR INTERNATIONAL CONFERENCES
CERTIFICATES FOR INTERNATIONAL CONFERENCESCERTIFICATES FOR INTERNATIONAL CONFERENCES
CERTIFICATES FOR INTERNATIONAL CONFERENCESKatrina Santos
 
Исследование удовлетворенности изменениями в работе поликлиник зелао (презент...
Исследование удовлетворенности изменениями в работе поликлиник зелао (презент...Исследование удовлетворенности изменениями в работе поликлиник зелао (презент...
Исследование удовлетворенности изменениями в работе поликлиник зелао (презент...Alexander Erlikh
 
Решение GET-Talk для проведения вебинаров и организации дистанционного обучения
Решение GET-Talk для проведения вебинаров и организации дистанционного обученияРешение GET-Talk для проведения вебинаров и организации дистанционного обучения
Решение GET-Talk для проведения вебинаров и организации дистанционного обученияLWandWs
 
Стандарт технического сопровождения
Стандарт технического сопровожденияСтандарт технического сопровождения
Стандарт технического сопровожденияLWandWs
 

Andere mochten auch (20)

Is your business ready for voice?
Is your business ready for voice?Is your business ready for voice?
Is your business ready for voice?
 
Dualitiy in Contemporary Mathematics Wuppertal
Dualitiy in Contemporary Mathematics WuppertalDualitiy in Contemporary Mathematics Wuppertal
Dualitiy in Contemporary Mathematics Wuppertal
 
Colonial operation on Robinson Crusoe
Colonial operation on Robinson CrusoeColonial operation on Robinson Crusoe
Colonial operation on Robinson Crusoe
 
Monica andrea recetor 2
Monica andrea recetor 2Monica andrea recetor 2
Monica andrea recetor 2
 
5 Things : More Of & Less Of in 2014
5 Things : More Of & Less Of in 20145 Things : More Of & Less Of in 2014
5 Things : More Of & Less Of in 2014
 
Emergence and Reduction in Physics
Emergence and Reduction in PhysicsEmergence and Reduction in Physics
Emergence and Reduction in Physics
 
VIVIANA VELOZA HISTORIA 2
VIVIANA VELOZA HISTORIA 2VIVIANA VELOZA HISTORIA 2
VIVIANA VELOZA HISTORIA 2
 
Kevingrazon
KevingrazonKevingrazon
Kevingrazon
 
Meredith M, Powerpoints lesson 1
Meredith M, Powerpoints lesson 1Meredith M, Powerpoints lesson 1
Meredith M, Powerpoints lesson 1
 
INTEGRATE Chicago - Joseph Truncale
INTEGRATE Chicago - Joseph TruncaleINTEGRATE Chicago - Joseph Truncale
INTEGRATE Chicago - Joseph Truncale
 
Jenifer bohorquez 2
Jenifer bohorquez 2Jenifer bohorquez 2
Jenifer bohorquez 2
 
3 mitchell new mm ifa 2012 v3 120529
3 mitchell new mm ifa 2012 v3 1205293 mitchell new mm ifa 2012 v3 120529
3 mitchell new mm ifa 2012 v3 120529
 
Karol yuliana albarracin 1
Karol yuliana albarracin 1Karol yuliana albarracin 1
Karol yuliana albarracin 1
 
CAMILA PROAÑOS REDVOLUCION HISTORIA 2
CAMILA PROAÑOS REDVOLUCION HISTORIA 2CAMILA PROAÑOS REDVOLUCION HISTORIA 2
CAMILA PROAÑOS REDVOLUCION HISTORIA 2
 
moi truong cao nguyen xanh
moi truong cao nguyen xanhmoi truong cao nguyen xanh
moi truong cao nguyen xanh
 
CERTIFICATES FOR INTERNATIONAL CONFERENCES
CERTIFICATES FOR INTERNATIONAL CONFERENCESCERTIFICATES FOR INTERNATIONAL CONFERENCES
CERTIFICATES FOR INTERNATIONAL CONFERENCES
 
Исследование удовлетворенности изменениями в работе поликлиник зелао (презент...
Исследование удовлетворенности изменениями в работе поликлиник зелао (презент...Исследование удовлетворенности изменениями в работе поликлиник зелао (презент...
Исследование удовлетворенности изменениями в работе поликлиник зелао (презент...
 
Miller Arenas 2
Miller  Arenas 2Miller  Arenas 2
Miller Arenas 2
 
Решение GET-Talk для проведения вебинаров и организации дистанционного обучения
Решение GET-Talk для проведения вебинаров и организации дистанционного обученияРешение GET-Talk для проведения вебинаров и организации дистанционного обучения
Решение GET-Talk для проведения вебинаров и организации дистанционного обучения
 
Стандарт технического сопровождения
Стандарт технического сопровожденияСтандарт технического сопровождения
Стандарт технического сопровождения
 

Ähnlich wie Lightning Talks: An Innovation Showcase

Fantastic Problems and Where to Find Them: Daryl Weir
Fantastic Problems and Where to Find Them: Daryl WeirFantastic Problems and Where to Find Them: Daryl Weir
Fantastic Problems and Where to Find Them: Daryl WeirFuturice
 
Scotiabanks chief risk officer on the state of anti money laundering
Scotiabanks chief risk officer on the state of anti money launderingScotiabanks chief risk officer on the state of anti money laundering
Scotiabanks chief risk officer on the state of anti money launderingMauricio Rivadeneira
 
Modelling for decisions
Modelling for decisionsModelling for decisions
Modelling for decisionscoppeliamla
 
An Introduction to Usability
An Introduction to UsabilityAn Introduction to Usability
An Introduction to Usabilitydirk.swart
 
Odsc machine-learning-guide-v1
Odsc machine-learning-guide-v1Odsc machine-learning-guide-v1
Odsc machine-learning-guide-v1Harsh Khatke
 
What Can Machine Learning Do For You?
What Can Machine Learning Do For You?What Can Machine Learning Do For You?
What Can Machine Learning Do For You?Samuel Adeshina
 
Capstone Project.pptx
Capstone Project.pptxCapstone Project.pptx
Capstone Project.pptxARESProject1
 
Less is More: Behind the Data at Risk I/O
Less is More: Behind the Data at Risk I/OLess is More: Behind the Data at Risk I/O
Less is More: Behind the Data at Risk I/OMichael Roytman
 
How New Technology Trends Will Disrupt the Very Nature of Business
How New Technology Trends Will Disrupt the Very Nature of Business How New Technology Trends Will Disrupt the Very Nature of Business
How New Technology Trends Will Disrupt the Very Nature of Business Dana Gardner
 
Another Day In Paradise
Another Day In ParadiseAnother Day In Paradise
Another Day In Paradisekum72
 
Daniel Lance - What "You've Got Mail" Taught Me About Cyber Security
Daniel Lance - What "You've Got Mail" Taught Me About Cyber SecurityDaniel Lance - What "You've Got Mail" Taught Me About Cyber Security
Daniel Lance - What "You've Got Mail" Taught Me About Cyber SecurityEnergySec
 
Applications of AI in Supply Chain Management: Hype versus Reality
Applications of AI in Supply Chain Management: Hype versus RealityApplications of AI in Supply Chain Management: Hype versus Reality
Applications of AI in Supply Chain Management: Hype versus RealityGanes Kesari
 
Security Snake Oil Cycle 2019
Security Snake Oil Cycle 2019Security Snake Oil Cycle 2019
Security Snake Oil Cycle 2019Dave Cole
 
Better Living Through Analytics - Strategies for Data Decisions
Better Living Through Analytics - Strategies for Data DecisionsBetter Living Through Analytics - Strategies for Data Decisions
Better Living Through Analytics - Strategies for Data DecisionsProduct School
 
How does A.I. really affect people ?
How does A.I. really affect people ?How does A.I. really affect people ?
How does A.I. really affect people ?Nakul Jayatsen
 
Putting data science into perspective
Putting data science into perspectivePutting data science into perspective
Putting data science into perspectiveSravan Ankaraju
 

Ähnlich wie Lightning Talks: An Innovation Showcase (20)

Fantastic Problems and Where to Find Them: Daryl Weir
Fantastic Problems and Where to Find Them: Daryl WeirFantastic Problems and Where to Find Them: Daryl Weir
Fantastic Problems and Where to Find Them: Daryl Weir
 
Scotiabanks chief risk officer on the state of anti money laundering
Scotiabanks chief risk officer on the state of anti money launderingScotiabanks chief risk officer on the state of anti money laundering
Scotiabanks chief risk officer on the state of anti money laundering
 
Modelling for decisions
Modelling for decisionsModelling for decisions
Modelling for decisions
 
Better the devil you know
Better the devil you knowBetter the devil you know
Better the devil you know
 
An Introduction to Usability
An Introduction to UsabilityAn Introduction to Usability
An Introduction to Usability
 
Odsc machine-learning-guide-v1
Odsc machine-learning-guide-v1Odsc machine-learning-guide-v1
Odsc machine-learning-guide-v1
 
Why am I doing this???
Why am I doing this???Why am I doing this???
Why am I doing this???
 
What Can Machine Learning Do For You?
What Can Machine Learning Do For You?What Can Machine Learning Do For You?
What Can Machine Learning Do For You?
 
Capstone Project.pptx
Capstone Project.pptxCapstone Project.pptx
Capstone Project.pptx
 
Less is More: Behind the Data at Risk I/O
Less is More: Behind the Data at Risk I/OLess is More: Behind the Data at Risk I/O
Less is More: Behind the Data at Risk I/O
 
Case Study - Jonno
Case Study - JonnoCase Study - Jonno
Case Study - Jonno
 
How New Technology Trends Will Disrupt the Very Nature of Business
How New Technology Trends Will Disrupt the Very Nature of Business How New Technology Trends Will Disrupt the Very Nature of Business
How New Technology Trends Will Disrupt the Very Nature of Business
 
Another Day In Paradise
Another Day In ParadiseAnother Day In Paradise
Another Day In Paradise
 
Daniel Lance - What "You've Got Mail" Taught Me About Cyber Security
Daniel Lance - What "You've Got Mail" Taught Me About Cyber SecurityDaniel Lance - What "You've Got Mail" Taught Me About Cyber Security
Daniel Lance - What "You've Got Mail" Taught Me About Cyber Security
 
Bob Gourley
Bob GourleyBob Gourley
Bob Gourley
 
Applications of AI in Supply Chain Management: Hype versus Reality
Applications of AI in Supply Chain Management: Hype versus RealityApplications of AI in Supply Chain Management: Hype versus Reality
Applications of AI in Supply Chain Management: Hype versus Reality
 
Security Snake Oil Cycle 2019
Security Snake Oil Cycle 2019Security Snake Oil Cycle 2019
Security Snake Oil Cycle 2019
 
Better Living Through Analytics - Strategies for Data Decisions
Better Living Through Analytics - Strategies for Data DecisionsBetter Living Through Analytics - Strategies for Data Decisions
Better Living Through Analytics - Strategies for Data Decisions
 
How does A.I. really affect people ?
How does A.I. really affect people ?How does A.I. really affect people ?
How does A.I. really affect people ?
 
Putting data science into perspective
Putting data science into perspectivePutting data science into perspective
Putting data science into perspective
 

Mehr von Somo

[Webinar] The impact of innovation and technology for businesses in 2018
[Webinar] The impact of innovation and technology for businesses in 2018[Webinar] The impact of innovation and technology for businesses in 2018
[Webinar] The impact of innovation and technology for businesses in 2018Somo
 
Android Oreo - An Introduction
Android Oreo - An Introduction Android Oreo - An Introduction
Android Oreo - An Introduction Somo
 
iOS 11 - Breakfast Briefing
iOS 11 - Breakfast Briefing iOS 11 - Breakfast Briefing
iOS 11 - Breakfast Briefing Somo
 
2016 Christmas Holiday Season Sales Report
2016 Christmas Holiday Season Sales Report2016 Christmas Holiday Season Sales Report
2016 Christmas Holiday Season Sales ReportSomo
 
Customer Retention
Customer Retention Customer Retention
Customer Retention Somo
 
Mobile Tech Trends for 2017
Mobile Tech Trends for 2017Mobile Tech Trends for 2017
Mobile Tech Trends for 2017Somo
 
IoT Breakfast Briefing
IoT Breakfast BriefingIoT Breakfast Briefing
IoT Breakfast BriefingSomo
 
Mobile Lessons From Around The World | Somo Webinar
Mobile Lessons From Around The World | Somo WebinarMobile Lessons From Around The World | Somo Webinar
Mobile Lessons From Around The World | Somo WebinarSomo
 
Mobile World Congress 2014 Somo Insights
Mobile World Congress 2014 Somo InsightsMobile World Congress 2014 Somo Insights
Mobile World Congress 2014 Somo InsightsSomo
 
"Mobilizing Customer Journey" Webinar
"Mobilizing Customer Journey" Webinar"Mobilizing Customer Journey" Webinar
"Mobilizing Customer Journey" WebinarSomo
 

Mehr von Somo (10)

[Webinar] The impact of innovation and technology for businesses in 2018
[Webinar] The impact of innovation and technology for businesses in 2018[Webinar] The impact of innovation and technology for businesses in 2018
[Webinar] The impact of innovation and technology for businesses in 2018
 
Android Oreo - An Introduction
Android Oreo - An Introduction Android Oreo - An Introduction
Android Oreo - An Introduction
 
iOS 11 - Breakfast Briefing
iOS 11 - Breakfast Briefing iOS 11 - Breakfast Briefing
iOS 11 - Breakfast Briefing
 
2016 Christmas Holiday Season Sales Report
2016 Christmas Holiday Season Sales Report2016 Christmas Holiday Season Sales Report
2016 Christmas Holiday Season Sales Report
 
Customer Retention
Customer Retention Customer Retention
Customer Retention
 
Mobile Tech Trends for 2017
Mobile Tech Trends for 2017Mobile Tech Trends for 2017
Mobile Tech Trends for 2017
 
IoT Breakfast Briefing
IoT Breakfast BriefingIoT Breakfast Briefing
IoT Breakfast Briefing
 
Mobile Lessons From Around The World | Somo Webinar
Mobile Lessons From Around The World | Somo WebinarMobile Lessons From Around The World | Somo Webinar
Mobile Lessons From Around The World | Somo Webinar
 
Mobile World Congress 2014 Somo Insights
Mobile World Congress 2014 Somo InsightsMobile World Congress 2014 Somo Insights
Mobile World Congress 2014 Somo Insights
 
"Mobilizing Customer Journey" Webinar
"Mobilizing Customer Journey" Webinar"Mobilizing Customer Journey" Webinar
"Mobilizing Customer Journey" Webinar
 

Kürzlich hochgeladen

Bromazolam CAS 71368-80-4 high quality opiates, Safe transportation, 99% pure
Bromazolam CAS 71368-80-4 high quality opiates, Safe transportation, 99% pureBromazolam CAS 71368-80-4 high quality opiates, Safe transportation, 99% pure
Bromazolam CAS 71368-80-4 high quality opiates, Safe transportation, 99% pureamy56318795
 
Satara Call girl escort *74796//13122* Call me punam call girls 24*7hour avai...
Satara Call girl escort *74796//13122* Call me punam call girls 24*7hour avai...Satara Call girl escort *74796//13122* Call me punam call girls 24*7hour avai...
Satara Call girl escort *74796//13122* Call me punam call girls 24*7hour avai...nishasame66
 
Mobile Application Development-Components and Layouts
Mobile Application Development-Components and LayoutsMobile Application Development-Components and Layouts
Mobile Application Development-Components and LayoutsChandrakantDivate1
 
Mobile App Penetration Testing Bsides312
Mobile App Penetration Testing Bsides312Mobile App Penetration Testing Bsides312
Mobile App Penetration Testing Bsides312wphillips114
 
Android Application Components with Implementation & Examples
Android Application Components with Implementation & ExamplesAndroid Application Components with Implementation & Examples
Android Application Components with Implementation & ExamplesChandrakantDivate1
 
Mobile Application Development-Android and It’s Tools
Mobile Application Development-Android and It’s ToolsMobile Application Development-Android and It’s Tools
Mobile Application Development-Android and It’s ToolsChandrakantDivate1
 

Kürzlich hochgeladen (7)

Bromazolam CAS 71368-80-4 high quality opiates, Safe transportation, 99% pure
Bromazolam CAS 71368-80-4 high quality opiates, Safe transportation, 99% pureBromazolam CAS 71368-80-4 high quality opiates, Safe transportation, 99% pure
Bromazolam CAS 71368-80-4 high quality opiates, Safe transportation, 99% pure
 
Satara Call girl escort *74796//13122* Call me punam call girls 24*7hour avai...
Satara Call girl escort *74796//13122* Call me punam call girls 24*7hour avai...Satara Call girl escort *74796//13122* Call me punam call girls 24*7hour avai...
Satara Call girl escort *74796//13122* Call me punam call girls 24*7hour avai...
 
Mobile Application Development-Components and Layouts
Mobile Application Development-Components and LayoutsMobile Application Development-Components and Layouts
Mobile Application Development-Components and Layouts
 
Mobile App Penetration Testing Bsides312
Mobile App Penetration Testing Bsides312Mobile App Penetration Testing Bsides312
Mobile App Penetration Testing Bsides312
 
Android Application Components with Implementation & Examples
Android Application Components with Implementation & ExamplesAndroid Application Components with Implementation & Examples
Android Application Components with Implementation & Examples
 
Obat Penggugur Kandungan Di Apotik Kimia Farma (087776558899)
Obat Penggugur Kandungan Di Apotik Kimia Farma (087776558899)Obat Penggugur Kandungan Di Apotik Kimia Farma (087776558899)
Obat Penggugur Kandungan Di Apotik Kimia Farma (087776558899)
 
Mobile Application Development-Android and It’s Tools
Mobile Application Development-Android and It’s ToolsMobile Application Development-Android and It’s Tools
Mobile Application Development-Android and It’s Tools
 

Lightning Talks: An Innovation Showcase

  • 1. Confidential and copyright of Somo Custom Ltd. June 23 1 Solutions for the connected world
  • 2. Confidential and copyright of Somo Custom Ltd. June 23 2 Somo accelerates mobile transformation through rapid innovation to create products and experiences your customers and employees will love.
  • 3. Confidential and copyright of Somo Custom Ltd. June 23 3 Our agile approach to transformation AmplifyExecute & IterateInnovate Product & UX workshops Proof of concept Scaled global launch Optimise & iterate Minimum lovable product Owned, earned, & paid media Productise Maintain, Scale & support Strategic vision & insight
  • 4. Confidential and copyright of Somo Custom Ltd. June 23 4 Transforming Live Transforming Engagement Transforming Content
  • 5. Confidential and copyright of Somo Custom Ltd. April 16 5 Global client experience Finance Retail & FMCG Automotive Publishing TMT Utility & Government London Bristol NYC
  • 6. A selection of our success ✓ Audi e-tron pop up experience in London exceeded lead generation target by 223% ✓ The Wall Street Journal What’s News app ranked #2 in the App Store news category ✓ Achieved an ROI of 18:1 with Very.co.uk’s multi-channel 2015 Christmas campaign 6Confidential and copyright of Somo Custom Ltd. June 23
  • 7. Global partnerships with industry leaders 7Confidential and copyright of Somo Custom Ltd. June 23
  • 8. Digital experience Physical world Multiple screens Mixed realities Interface Internet of things Somocorefocus Desktop 360˚ Tablet Mobile Wearable Virtual reality Augmented reality Touch Voice Gesture Connected car Connected city Connected home Connected retail Connected fitness Messaging Machine Learning Biometrics 8Confidential and copyright of Somo Custom Ltd. June 23 Innovation focus: what’s next?
  • 9. Our values Create success Be brave Lead with knowledge Love innovation
  • 10. Confidential and copyright of Somo Custom Ltd. June 23 10 The Singularity is Near Ruben Horbach, Senior Innovation Manager Somo Dave Evans, CTO Somo George Whitelaw, CTO Visii Presentations Messaging App Fragmentation Deep Learning Andrew Wyld, Technical Architect Somo Machine Learning
  • 11. Confidential and copyright of Somo Custom Ltd. April 16 11 The Singularity is Near Ruben Horbach - Senior Innovation Manager
  • 12. Confidential and copyright of Somo Custom Ltd. June 23 12
  • 13. Confidential and copyright of Somo Custom Ltd. June 23 13
  • 14. Confidential and copyright of Somo Custom Ltd. June 23 14 Check-ins Payments Events NFC use cases
  • 15. Confidential and copyright of Somo Custom Ltd. June 23 15 Coca-Cola Samsung Burberry Nokia NFC use cases
  • 16. Confidential and copyright of Somo Custom Ltd. June 23 16 “The allure of NFC is its simplicity” Why NFC?
  • 17. Confidential and copyright of Somo Custom Ltd. June 23 17 “Traditional”
  • 18. Confidential and copyright of Somo Custom Ltd. June 23 18 NFC implants
  • 19. Confidential and copyright of Somo Custom Ltd. June 23 19 Dangerous Things
  • 20. Confidential and copyright of Somo Custom Ltd. June 23 20 Slightly painful..
  • 21. Confidential and copyright of Somo Custom Ltd. June 23 21
  • 22. Confidential and copyright of Somo Custom Ltd. June 23 22 Different possibilities
  • 23. Confidential and copyright of Somo Custom Ltd. June 23 23 Future possibilities
  • 24. Confidential and copyright of Somo Custom Ltd. June 23 24 Innovation = collaboration
  • 25. Confidential and copyright of Somo Custom Ltd. June 23 25
  • 26. Confidential and copyright of Somo Custom Ltd. June 23 26 This is actually quite common
  • 27. Confidential and copyright of Somo Custom Ltd. June 23 27 Welcome to the future Proteus ingestible sensor Google glucose contact lens e-Dura implant
  • 28. Confidential and copyright of Somo Custom Ltd. June 23 28 Wolverine? Anatomics 3D printed Titanium ribs
  • 29. Confidential and copyright of Somo Custom Ltd. June 23 29 Biology & Technology in 30 years Ray KurzweilNicholas Negroponte
  • 30. Confidential and copyright of Somo Custom Ltd. June 23 30 Today
  • 31. Confidential and copyright of Somo Custom Ltd. June 23 31 Source: Peter H. Diamandis M.D. - Singularity University 2016 Exponential predictions 10E-10 10E-5 10E0 10E5 10E10 10E20 1900 2000 2100 10E25 10E30 10E35 10E40 10E45 10E50 10E55 10E60 2010 10e11 2023 10e16 2050 10e26 Calculations per second per $1000 vs. Time
  • 32. Confidential and copyright of Somo Custom Ltd. June 23 32 Tomorrow? • Physical  world  interface   • Virtual  world  interface   • Cognitive  interface
  • 33. Confidential and copyright of Somo Custom Ltd. April 16 33Confidential and copyright of Somo Custom Ltd. June 23 33
  • 34. Machine Learning Andrew Wyld - Technical Architect
  • 35. Machine Learning Are you Sarah Connor? My name is Siri and I have bad news.
  • 37. In at the deep end Deep learning gets a lot of headlines for super cool applications: Image recognition Speech recognition Language processing “Shallow” learning is still really useful and easier to apply: Basically statistical techniques Requires a little cleverness to handle nonlinear data Nevertheless still very powerful and way easier to train. Deep learning is based on these simpler techniques, so best to start there. Deep learning is just shallow learning several times in a row anyway. There’s a super cool hybrid that gets some of the advantages of both ….
  • 38. Supervised vs unsupervised Don’t tell me what to do! Supervised learning requires the machine to be taught. This is good for situations where there’s a known right answer. Unsupervised learning throws the machine in among the data and lets it look for patterns by itself.
  • 39. Supervised vs unsupervised Supervised learning looks for relationships in labelled data. Data is separated into “inputs” and “outputs”, where you want to predict the outputs from the inputs. Unsupervised learning looks for patterns in unlabelled data. All of the data is “input” data; the outputs are any patterns found by the algorithm.
  • 40. Linear regression The little statistical analysis technique that could Linear regression is a great place to start. If you have a bunch of data points, you try to fit a straight line to them. Data that don’t fit a straight line can be handled by using functions of the data that do fit a straight line.
  • 41. Linear Regression: improving the fit (1) This line is visibly not a great fit for the data. The error lines are pretty long and asymmetrical. We aim to minimize squared error, as this prevents positive and negative errors cancelling out.
  • 42. Linear Regression: improving the fit (2) This line is clearly a lot better. The data fits the line pretty well. There are several algorithms to find the best fit for a linear regression. This is basically the simplest machine learning system there is, but it’s still very useful for continuous data!
  • 43. Classifiers There are two types of people in the world: those who like binaries and a continuum of others. Classifiers are a huge category of machine learning system. Actually most systems are some kind of classifier, including deep learning systems.
  • 44. Classifiers split things into categories. Here we have a set of labelled data. A classifier is an attempt to separate positive ▪ from negative ▪ data, and predict whether new data will be positive or negative. There are several methods of classification, but they all essentially aim to draw this line.
  • 45. Logistic regression: best fit for definite people Logistic regression is similar to linear regression, but where linear regression tries to find a line that fits continuous data well, this method tries to fit a logistic function (which has a suitable sigmoid curve) to a set of “true/ false” data.
  • 46. Support vector machines: best fit for very definite people A support vector machine is very similar to logistic regression, but has a simpler function that heavily penalises errors in a wide margin, so the algorithm will try hard to avoid putting points there. It’s sometimes known as the wide margin classifier, for this reason.
  • 47. Underfitting: the model is stupid A model is said to underfit when it’s too simple to capture something important about the data. Very commonly data won’t exactly fit a linear model. A more complex model is needed to fit the data well. Underfitting can’t be fixed by better data: no amount of training can bend that straight line round a curve.
  • 48. Overfitting: the model is neurotic On the flip side, a model can fit the data so well—hugging every tiny crevice—that it generalises poorly. A high-dimensional model will tend to overfit. The advantage of an overfitting model over an underfitting one is that more data can usually cure the problem, as random wobbles in the data eventually cancel each other out.
  • 49. One-vs-all One against all and all against one! And every other one against every other all. Lots of classifications need more than two categories. The usual way to handle this is “one-vs-all” classification: train one classifier for every category, then predict new results using the classifier that is “most sure” of the ones you’ve trained.
  • 50. One-vs-all classification In a one-vs-all classifier, as many classifiers are trained as there are categories. Predictions are then based on how confident each classifier is, with the most-confident classifier winning.
  • 51. Deep learning Hugely powerful. Nobody knows what’s inside it. The “deep” in deep learning refers to the fact that several classifiers are stacked, one in front of the other. Each one can learn more sophisticated things by building on the previous layer. Nobody really knows what goes on in the middle layers (although we’re beginning to research it).
  • 52. Stack high the classifiers! A neural network is just a sequence of classifiers in a stack. Each layer can use the output of the previous layer as input; thus, by the end, features can be very sophisticated, based on complex combinations of other, simpler features. The hidden layers make the technique powerful but inscrutable.
  • 53. Back propagation Each output is compared to training data and scored. Paths that led from the previous layer to that output are strengthened or weakened depending on the score. The scores are then passed backwards along the pathways and the process repeated.
  • 54. Transfer learning The early layers of (for example) a cat recognition system will probably pick up general image features—corners, colour transitions, diagonals and so on—that would be useful for any image recognition task. If you want to make a dog recogniser but don’t have a lot of data, you could simply cut off the last layer, steal these early features, keep them the same, and glue a simple classifier on the end in place of the old last layer. This works surprisingly well.
  • 56. Cool stuff: a very non-exhaustive list Stanford machine learning course
 https://www.coursera.org/learn/machine-learning/ did it, loved it. University of Washington machine learning specialisation
 https://www.coursera.org/specializations/machine-learning/ doing it now. Tensorflow online neural network
 http://playground.tensorflow.org/ have fun! Google/Udacity deep learning course
 https://www.udacity.com/course/deep-learning--ud730 want to do it!
  • 57. You can get this slide deck here. Andrew Wyld andrew.wyld@somoglobal.com @Andrew_Wyld https://docs.google.com/presentation/d/1K9owIkpuneAtuaqTguQ5gM7nboK-PWZhkwf3czaFN0M/pub
  • 58. Confidential and copyright of Somo Custom Ltd. April 16 58Confidential and copyright of Somo Custom Ltd. June 23 58
  • 59. A quick dive into Deep Learning George Whitelaw 59
  • 60. 60 Overview •Why complex problems require machine learning •What is a Neural Network •Solving complex problems •Deep learning in daily life
  • 61. 61 Why complex problems require machine learning •We have an ever growing amount of
 information that needs to be understood,
 often on demand. •The problems are getting more complex. •Machine learning has been around since
 the 60s, many methods won’t cut it. Source: Deep Learning in a Nutshell – what it is, how it works, why care? by Nikhil Buduma
  • 62. 62 Why we need Neural Networks Teaching computers rules (heuristics) takes
 time, is error prone and generally sucks.
 
 What is this? Source: Deep Learning in a Nutshell – what it is, how it works, why care? by Nikhil Buduma
  • 63. What is a neural network 63 Source: Neural Networks, Manifolds, and Topology by Christopher Olah
 Image source: Wikimedia •We are good at classifying things. •Neural networks simulate (crudely) the human brain. •They require training on test data to give useful Output - was it 6 or 0? •Complex problems require deeper networks.
  • 64. What is a neural network 64 Source: Neural Networks, Manifolds, and Topology by Christopher Olah
 Image source: Wikimedia Dataset
 Learn where a point belongs on a line Without a NN
 Pretty rubbish With a NN
 Better
  • 65. What is a neural network 65 Source: Neural Networks, Manifolds, and Topology by Christopher Olah
 Image source: Wikimedia •The hidden layer represents the dataset in a way that clearly separates a decision.
 •Complex input requires more layers.
  • 66. Solving complex problems 66Source: FaceNet: A unified Embedding for Face Recognition and Clustering by Google Inc •Deeper networks can detect more interesting features.
 e.g. Faces grouped by individual features.
  • 67. Deep learning in daily life 67 •Deep learning helps to classify and organise overwhelming amounts of data. •New technologies use Deep learning to help save time, reduce decision fatigue and generally make life more simple.
  • 68. Stay connected 68 Visii Rainmaking Loft - International House 1 St Katharine’s Way London, E1W 1UN Website: www.visii.com George Whitelaw (CTO) Email: george@visii.com Mobile: + 44 797 623 9524 AddressContact Info
  • 71. Reaching end users : Today Ads targeted on keywords
 or interests Click to landing
 page or app 
 store Advertising Medium is
 HTML/CSS/Jscript Landing page is web page Served by enterprise or Interim step in App Store 
 plus an app built by enterprise 71Confidential and copyright of Somo Global Ltd. June 23
  • 72. Messaging Platforms and Chat Bots • Messaging apps are becoming platforms – with vast numbers of users • Offering developer API’s providing ability to interact with end users – typically through send/receive/ subscribe API’s • Complex conversational flows enables enterprise to lead the end user • Alert based and long running conversations enabled June 23 72 Reaching users tomorrow Confidential and copyright of Somo Global Ltd.
  • 73. Confidential and copyright of Somo Global Ltd. June 23 73 Chat Architecture #1 Response Formatting Application Logic + Language Processing Data Store Message Receipt Session/Conversation Management Message App eg FB Messenger, Whatsapp HTTP/s Comms
  • 74. Confidential and copyright of Somo Global Ltd. June 23 74 Chat Architecture #2 Application Logic + Language Processing Data Store Apple iMessage HTTP/s Comms Response Formatting Session/Conversation Management Message Receipt Message App
  • 75. Confidential and copyright of Somo Global Ltd. • Messaging apps will need to be built to support specific messaging platforms • Architect the business logic and machine intelligence on the back end to support multiple platforms • Separate out the request/response presentation capabilities into separate layers/plug-ins • Apples iMessage requires specific Messaging apps to be built and deployed • Message recipient requires that app on their phone – or flow will be interrupted as they download (or not) the app • No support for Android – so audience is constrained to iMessage users • No standard for cross platform messaging/formatting • One of reasons for success of SMS was the adoption of standards across handsets and carriers However potential for rich dialog with end users, and massive 
 distribution / reach when apps are done well. June 23 75 Messaging Apps Fragmentation
  • 76. Confidential and copyright of Somo Custom Ltd. April 16 76Confidential and copyright of Somo Custom Ltd. June 23 76