2. How do you get up-to-speed?
here’s a couple books
3. other ways to get up-to-speed
http://www.dataversity.net/category/data-topics/smart-data-data-topics/cognitive-
computing-news-articles-education/
http://www.dataversity.net/
#cognitivecomputing…….@ibmwatson thinks they own this one ;-)
#cogcomputing
#contextAwareness
#smartdata………………..coming on strong
#machinelearning…………lot’s of geeks here
#deeplearning
#machineintelligence…a fav because combines #ai + #machinelearning
#AmbientIntelligence…….emerging fast!
#LinkedData
#IoT
#InternetOfThings
#IoE
#M2M
etc.
4. other ways to get up-to-speed
subscribe to http://www.storminsights.com/
Available Now...A new annual service featuring: STORMWatch: The Commercial Cognitive Computing
Emerging Market Report - a bi-monthly guide to the cognitive computing ecosystem.
more on this in
ensuing slides
6. 6
Cognitive Computing in 3 bullets
from http://www.cognitivecomputingforum.com/ in August 2014
1. Context driven dynamic algorithms for automating pattern discovery
2. Reasons and to discern context for sense-making.
3. Learns, adapt, improve over time without direct modeling or programming.
Adrian Bowles: Learning is fundamental baseline requirement and
patterns, relationships, context are three words that come up.
Matt Sanchez: Cognitive Computing handles dark data i.e. all of the
unstructured data that makes up 80 percent of information that’s out
there today.
Jim Kobielus: It’s all about engagement to drive decision support and
guidance to humans trying to make decisions in various contexts.
Tony Sarris: Lots of focus on machine learning but should also include
ontologies and knowledge modeling i.e. Linked Open Data models
7. RT @Ag_Loop Can we stop calling it #Aerial bigdata already?! http://okt.to/5Rb4W3 #SmartData ”
• Humans ( modelers, information architects, taxonomists, ontologists ) still generate
smart data but algorithms fast becoming new star designers
• In 2015 major shift to machine learning with dynamic algorithms to convert to
smart data for contextual, personalized, actionable information
• Without smart data cannot do cognitive computing that learns, adapts, improves
over time.
my April 2015 update to Cognitive Computing in 3 bullets from August 2014
9. How does Smart Data & Cognitive Computing enhances this?
https://www.linkedin.com/pulse/big-data-predictive-analytics-healthcare-charles-gellman-mshi
10. by @cirrus_shakeri on @LinkedIn https://www.linkedin.com/pulse/from-big-data-
intelligent-applications-cirrus-shakeri
Smart Data
11. led by emergence of SMAC (social, mobile,
analytics, cloud)
Why the need for AI / machine learning
15. @IBMWatson most recognized name in cognitive computing
http://www.ibm.com/
smarterplanet/us/en/ibmwatson/
http://www-01.ibm.com/software/bluemix/
https://developer.ibm.com/bluemix/
@IBMbluemix
User Modeling
Machine Translation
Concept Expansion
Message Resonance
Question & Answer.
Visualization Rendering
SOLUTIONS
IBM SET TO EXPOSE 50 WATSON APIS
http://www.programmableweb.com/news/
ibm-set-to-expose-50-watson-apis/
2014/10/08?utm_content=bufferd2a47
IBM Looks to Enhance Bluemix Usership With
New Developer-Centric APIs for Watson
http://dataconomy.com/ibm-looks-to-
enhance-bluemix-usership-with-new-
developer-centric-apis-for-watson/
IBM’s IoT Foundation Available for Beta Test Drive http://fw.to/EgBJcpU
Using #IBMWatson on @IBMBluemix
http://slidesha.re/1tITgVE
IBM Reboots Research to Focus
on Cognitive Computing
19. IBM works to deliver on Watson's cognitive computing promise
http://searchdatamanagement.techtarget.com/news/2240238506/IBM-works-to-deliver-on-Watsons-cognitive-computing-promise
"The key point is cognitive computing is blending with machine learning, AI and ambient
intelligence said Steve Ardire, advisor to software startups. ”Watson and cognitive
computing are a greenfield opportunity for IBM to reinvent itself and it’s a huge bet but
others are in the hunt”.
”Watson is getting the most attention because IBM has the most marketing dollars.
“Watson Analytics is an impressive app, but there are other ways to build impressive
cognitive analytics apps”
• Microsoft Officially Launches #Azure #MachineLearning Platform, Turns Heads and Bing
Distill #machinelearning project echoes IBM's plans with Watson
• Sentient.ai to Live in Tata Data Centers
• Accenture and ipsoft.com Amelia
• Wipro spending $200M on cognitive technology, machine learning and smart devices.
• Infosys raises start-up fund size to $500M focusing on AI and IoT in move to automation and
hiring just 30,000 next year the lowest in three years
• Bridgewater Rolls Out New Artificial Intelligence Division that “learns as markets change and
adapts to new information” as opposed to following static instructions.”
• and much more
20.
21.
22. Machine Learning and Deep Learning Google Trends
Smart Data and Cognitive Computing Google Trends
23. The Basic Recipe For Machine Learning Explained In A
Single PowerPoint Slide
1) First thing I ask is “does it do well on the training data?” If it doesn’t, then I would build a bigger network,
or “rocket engine”, so you have more neurons, more weights to try and fit the training data well.
2) Then you see if it fits the test data or development data. If it doesn’t do well on the test data but you’re
doing well on the training data, that means you’re overfitting. The most reliable cure for overfitting is to
get more data, to get more rocket fuel.
3) And then you keep going around and around until eventually it does well in the training data, it does
well in the test data and then hopefully you’re done.
Ng advised this is
highly simplified
explanation of what
his job entails and
often computer
scientists still run into
problems even after
following these steps.
At this point, you need
to modify the network
architecture. Or cast
some black magic
24. ML is where algorithms process data, draw conclusions, make
inferences, change, adapt, become better
main categories
• Supervised that takes place when the training instances are labelled with the correct result, which
gives feedback about how learning is progressing. Most learning is still supervised.
• Unsupervised discovering internal structure of data, mutual dependencies between input
variables, disentangling explanatory factors of variations to create way for machine to learn to
understand the world the way a human does.
• Reinforcement learns by trial and error e.g. A baby learns that when you put a toy behind a box,
the toy is still in the world”
• Deep learning - uses multi-layer neural network algorithms to understand patterns, make non-
linear transformations, inferences, adapt, become better
Practical Examples
• Recognition - Speech to text, voice, face, fingerprint, parts, etc.
• Natural Language Processing - Translation, sentiment analysis, etc.
• Recommendation - If you like this, you may love this too.
• Diagnostics - For medical or mechanical systems.
• Categorization - For categorizing text, images, audio, spam filtering, etc.<
• Prediction - What stock will go up or down? Who may be a terrorist?
• Analysis - Share of basket, viewer profiling, fraud detection, etc.
• Yield Management - Programmatic advertising auctions, airline seat pricing, etc.
• Autonomous Vehicles - Cars, robots, drones, etc.
26. Data is increasingly being delivered in streams
requires machine learning with dynamic algorithms to convert to smart data
http://www.dataminr.com/technology/
27. Data streams integrated with web, RDF, LinkedData
enterprise SQL, NoSQL, Taxonomies, Ontologies
for Real-time Information Discovery
28. can no longer use ‘traditional’ ETL/data prep DW/BI architecture for massive
amounts of unstructured + streaming data across 4V’s esp. great velocity.
Implications
29. multi-layer convolutional neural network algorithms to understand patterns, make
non-linear transformations, inferences, adapt, and become better
Deep Learning is New Black
30. RT @ArnoCandel: I don't always learn, but when I do, I learn deeeeeply
#mlconfsf @hexadata @MLconf #h2oworld #bigdata
31. Generalized vs Specialized Deep Learning for Small Data and Big Data
http://www.kdnuggets.com/2015/03/small-data-specialized-deep-learning-yann-lecun.html
Data Scientists in field without Big
Data (Quadrant C) simply cannot
wait to gather sufficient data to
implement Deep Learning in the
way that works in your industry
(Quadrant A).
Ideally all efforts should gravitate
to Holy Grail (Quadrant B), where
Specialized Deep Learning
converges with Big Data to give
us amazing insights about our
fields. Unfortunately that’s time
dependent and we’ll have to work
within our means until the data
catches up.
32. Google Creates Artificial Intelligence Program Smart Enough To Learn From Its
Mistakes and Thinks And Plays Video Games Like We Do Many people out there
are probably thinking "Eh. So what. It's for video games."
Well, currently yes, but DeepMind Technologies says that "The ultimate goal is to
build general purpose smart machines", and essentially machines capable of
learning like we do.
This may be a decade or more away, but this is still an advancement, and a very
significant small step.
Google acquired
for $500M
Why your company should focus on Machine Intelligence
(AI + machine learning)
33. Facebook uses #DeepLearning to identify the content and
context of videos and sentences
http://techcrunch.com/2015/03/26/a-i-book/#JUvykc:sBTl
Facebook pouring money into AI and language technology and last year it hired one of the top
deep learning researchers Yann LeCun to start a whole AI lab at NYU.
AI to identify different types of sports to recognize tiny differences between activities and read
sentences to understand content of lines from fantasy novels like characters in Lord Of The Rings
without any prior knowledge using Memory Network, which Facebook says “enables a machine to
perform relatively sophisticated question and answering”.
Why Does This Matter?
Because understanding the meaning of content makes it much easier to route that content to the
relevant audience. If it knows you love football and hate baseball, the ability to recognize which
sport is in a video means it can only show you clips you care about.
34. Hadoop ecosystems players use ML/MI
top players
Pivotal CEO says open source Hadoop tech is coming
https://gigaom.com/2015/02/06/exclusive-pivotal-ceo-says-open-source-hadoop-tech-is-coming/ via @gigaom
Well newsflash open source Hadoop and Spark solutions already here with smart
data-as-as-service and intelligent apps some of which are here at EDW
35. Spark now 5 yrs old surpasses Hadoop so creating new
ecosystems and opportunities esp for clever startups
open source machine & deep
learning, unveils AI developer
program to make apps smarter
Apache Spark: 3 Promising Use-Cases http://goo.gl/nMzcoJ by
@jameskobielus who says Spark is the latest shiny new big-data bauble
IMO a hard line b/c @ApacheSpark threat to status quo players with
proprietary solutions who could lose big!
36. ApacheSpark, DataBricks, open source ML / MI
@ApacheSpark with Data Frames, Spark SQL, and MLlib Improvements looks like the
future of big data http://bit.ly/17BYqfl with 8 Reasons #Apache #Spark is So Hot http://
buff.ly/1NKduJb
@databricks Cloud and DataFrames in Spark
• provides "unified platform with one API for batch, streaming, interactive queries
• Hadoop vendors threatened by Spark because its success would diminish their analytical role
and prospects for components such as Cloudera Impala.
• Intel partners with Databricks to push adoption of the Spark data-processing engine
@amplab The AMPLab at UC Berkeley is integrating Algorithms, Machines, People to
make sense of Big Data.
@ApacheIgnite In-Memory Data Fabric is high-performance, integrated, distributed in-
memory platform for computing and transacting on large-scale data sets in real-time
@ApacheAurora aurora.incubator.apache.org is a Mesos framework with Docker
Apache Tajo brings data warehousing to Hadoop
DeepDive http://deepdive.stanford.edu/ DARPA Offers Free Watson-Like AI
deeplearning4j with Spark & GPU accelerations and language agnostic text analysis
38. Shivon Zilis Bloomberg Beta spent 3 mo learning about AI, machine learning, data startups and
complied a list of 2,500 and said I’d use this #machineintelligence landscape to help figure out
what core and supporting technologies I could package into a novel industry application”
“The real battle isn’t being fought over the underlying machine learning technology, it’s in
building support systems to make it usable.” …Stephen Purpura CEO Context Relevant
this means smart data acquisition, data
harmonization with semantic enrichment
using both machine learning and rules
Applications Drive The Biggest Money In Big Data
http://readwrite.com/2014/12/31/big-data-companies--applications-money-startups
The "Big winners" in Big Data are infrastructure providers like Hadoop and NoSQL vendors
and "Bigger winners" are Apps and Analytics vendors that abstract complexity of working with
very complicated underlying technologies into a user friendly front end
39. 39
AlchemyAPI $2 Million in 1 Round (Acquired by IBM on March 4, 2015)
Ayasdi $106M with $55M on 3/25/15 to Blend AI & Machine Learning
Blekko $60M in 9 Rounds ( IBM acquires ‘certain assets’ on Mar 27, 2015 )
ClearStory Data $30 Million in 3 Rounds from 5 Investors
Context Relevant $44.3 Million in 5 Rounds
Databricks (Apache Spark creators) $47M in 2 Rounds $33M Series B June 2014
DataRobot $21M Series A at $60M pre-money before launching
Dataminr $179.6 Million in 5 Rounds with $130M / Series D Mar 17, 2015
Datasift $77.9 Million in 6 Rounds
Digital Reasoning $29M in 4 Rounds with $24M Series C on October 9, 2014
H20.ai $13.6M open-source in-memory scalable machine learning for smarter applications
MetaMind $8M wants to make Deep Learning accessible to everyone
Palantir Technologies $1 Billion in 15 Rounds from 10 Investors
Quid $24 Million in 2 Rounds
ScaledInference $13.6M Big Machine Intelligence for Al
Sentient.ai $143.8 Million in 5 Rounds for massively parallel AI computing
Skytree $22M The Machine Learning Company®
Vicarious $72 Million in 4 Rounds without a product yet
Viv.ai (Siri’s Creators) $22.5M for AI That Can Teach Itself
Wise.io | Machine Learning as a Service $2.8M Series A
VC $ flowing into #machinelearning and
#machineintelligence #ai startups
Here's secret to how Silicon Valley calculates the value of its hottest companies
The numbers are sort of made-up http://bloom.bg/1wV4oE7
40. RT@Ayasdi $55M Series C takes us close to $100M in total. Why is
#machineintelligence so hot? Std. #Analytics broken.
41. 41
Engagement Layer UI’s contingent on vertical use case and
targeted user e.g. clinician, knowledge worker, consumer
Life Sciences
Oil & Gas
Public Sector
Financial Services
Manufacturing
Retail
Collaboration
Customer Service
Data Shape Has Meaning
Virtual AI with Conversational Speech capabilities
42. ex-Siri people creating an intelligent,
conversational interface to anything
with predictive capabilities
42
Siri with Watson now has apps
Microsoft infuses Cortana with a
personality
Google to develop "fully reasoning" AI but
real-life Skynet still a few years off
advanced psychobiological simulation with
emotion intelligence which learns and interacts
in real time with neuroscience models
Cognitive AI
artificial general intelligence (AGI) should also be
“human-like" Professor says Murray Shanahan, AI
expert and consultant on the film Ex Machina
Facebook’s deep learning “Memory Networks to
Unleash Deep Learning and Make Machines Smarter
Future of #AI is like Smart Lego blocks ( much less
coding ) as blocks only need to be built once.
Open source 'Sirius' joins the likes of
Google Now, Siri and Cortana
Has ability to understand user intent then generate
algorithms on the fly and continually update its
capabilities based on new content added to global brain.
Yann LeCun on Samantha in Her still “totally
out of reach of current technology. We will
need to invent new concepts, new principles,
new paradigms, new algorithms.”