strata san jose 2016: deep learning is eating your lunch -- and mine
Post on 09-Feb-2017
13.433 Views
Preview:
TRANSCRIPT
H2O.ai Machine Intelligence
Afraid of the future?You should be. Deep Learning iseating your lunch — and mine.
San Jose, March 29 2016
Arno Candel, PhDChief Architect
H2O.ai Machine Intelligence
Who Am I?
Arno Candel Chief Architect, Physicist & Hacker at H2O.ai
PhD Physics, ETH Zurich 2005 10+ yrs Supercomputing (HPC) 6 yrs at SLAC (Stanford Linear Accelerator) 4+ yrs Machine Learning 2+ yrs at H2O.ai
Fortune Magazine Big Data All Star
Follow me @ArnoCandel
2
WhoamI?
H2O.ai Machine Intelligence
• What is Deep Learning? • What can it do? • What will it change? • What can you do?
3
Outline
H2O.ai Machine Intelligence 4
WhatisDeepLearning?
Machine Learning (ML)
Artificial Intelligence (A.I.)
Computer Science (CS)
H2O.ai
Deep Learning (DL)
H2O.ai Machine Intelligence 5
WhatisDeepLearning?
That’s the stuff that powers pretty much all of Google.
Deep Learning (DL)
http://www.wired.com/2016/02/ai-is-changing-the-technology-behind-google-searches/
H2O.ai Machine Intelligence 6
WhatisDeepLearning?
Or, more technically, it’s a computer algorithm that models high-level abstractions in data with multiple layers of non-linear transformations.
Deep Learning (DL)
H2O.ai Machine Intelligence 7
ExampleDLModel:Multi-Layer(Deep)ArtificialNeuralNetwork
income
loan amount
credit scoreborrower will not default on loan
borrower will default on loan
IN: data OUT: prediction
nodes : neuron activations (real numbers) — represent features arrows : connecting weights (real numbers) — learned during training
: non-linearity x -> f(x) — adds models complexity
mid 1970s
H2O.ai Machine Intelligence 8
DeepLearningProsandCons
• conceptuallysimple• nonlinear• highlyflexibleandconfigurable• learnedfeaturescanbeextracted• canbefine-tunedwithmoredata• efficientformulti-classproblems• world-classatpatternrecognition
Pros:
• hardtointerpret• theorynotwellunderstood• slowtotrainandscore• overfits,needsregularization• manyhyper-parameters• inefficientforcategoricalvariables• verydatahungry,learnsslowly
Cons:
Deep Learning got boosted recently by faster computers
H2O.ai Machine Intelligence 9
BriefHistoryofA.I.,MLandDL
John McCarthyPrinceton, Bell Labs, Dartmouth, later: MIT, Stanford
1955: “A proposal for the Dartmouth summer research project on Artificial Intelligence”
with Marvin Minsky (MIT), Claude Shannon (Bell Labs) and Nathaniel Rochester (IBM)
http://www.asiapacific-mathnews.com/04/0403/0015_0020.pdf
A step back: A.I. was coined over 60 years ago
H2O.ai Machine Intelligence 10
1955proposalfortheDartmouthsummerresearchprojectonA.I.
“We propose that a 2-month, 10-man study of artificial intelligence be carried out during the summer of 1956 at Dartmouth College in Hanover, New Hampshire. The study is to proceed on the basis of the conjecture that every aspect of learning and any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it. An attempt will be made to find how to make machines use language, form abstractions and concepts, solve kinds of problems now reserved for humans, and improve themselves. We think that a significant advance can be made in one or more of these problems if a carefully selected group of scientists work on it together for one summer.”
H2O.ai Machine Intelligence 11
It took a little longer…
McCarthy invents LISP in 1958, influences ALGOL, popularizes timesharing, wins Turing Award, Kyoto Price, National Medal of Science, etc.
A.I.takestime
H2O.ai Machine Intelligence 12
Almostthere!Ornot?
A few thousand years ago Today: still plowing along
https://en.wikipedia.org/wiki/Ploughhttp://www.processindustryinformer.com/latest-news/news-events/industry-news/robot-helps-farmers-monitor-fertiliser-water-usage
We can automate many tasks, general A.I. is still far away.
H2O.ai Machine Intelligence 13
Step1:GreatAlgorithms+FastComputers
Raw computing power can automate complex tasks!
http://nautil.us/issue/18/genius/why-the-chess-computer-deep-blue-played-like-a-human
Earlier he said: “No computer will ever beat me.”
1997: Playing Chess (IBM Deep Blue beats Kasparov)
ComputerScience 30customCPUs,60billionmovesin3mins
H2O.ai Machine Intelligence 14
Step2:MoreData+Real-TimeProcessing
http://cs.stanford.edu/group/roadrunner/old/presskit.html
2005: Self-driving CarsDARPA Grand Challenge, 132 miles (won by Stanford A.I. lab*)
Sensors&ComputerSciencevideo,radar,laser,GPS,7Pentiumcomputers
Automating automobiles into autonomous automata!
*Established by McCarthy et al. in the early 60s
H2O.ai Machine Intelligence 15
Step3:BigData+In-MemoryClusters
2011: Jeopardy (IBM Watson)
In-MemoryAnalytics/ML4TBofdata(incl.wikipedia),90servers,16TBRAM,Hadoop,6millionlogicrules
Automating question answering and information retrieval!
https://www.youtube.com/watch?v=P18EdAKuC1U https://en.wikipedia.org/wiki/Watson_(computer)
Note: IBM Watson received the question in electronic written form, and was often able to (electronically) press the answer button faster than the competing humans.
H2O.ai Machine Intelligence 16
Step4:DeepLearning
2014: Atari Games (DeepMind)
2016: AlphaGo (Google DeepMind)
DeepLearning +reinforcementlearning,treesearch,
MonteCarlo,GPUs,playingagainstitself,…
https://deepmind.com
Deep Learning + Smart Algorithms = Master Gamer
Go board has approx. 200,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000 (2E170) possible positions.
trainedfromrawpixelvalues,nohumanrules
H2O.ai Machine Intelligence 17
Step5:ImproveTrainingEfficiency
2015: MIT, NYU, Toronto
BayesianProgramLearning(NOTDeepLearning)
New algorithm learns handwriting of unseen symbols from very few training examples (unlike typical Deep Learning)
http://www.nytimes.com/2015/12/11/science/an-advance-in-artificial-intelligence-rivals-human-vision-abilities.html
H2O.ai Machine Intelligence 18
MicrosoftA.I.goesrogue
https://twitter.com/TayandYou/with_replies
P.S. Microsoft won the Visual Recognition challenge: http://image-net.org/challenges/LSVRC/2015/
H2O.ai Machine Intelligence 19
WhatELSEcanDeepLearningdo?
http://www.cs.toronto.edu/~graves/handwriting.cgi
Deep Learning can generate handwriting
H2O.ai Machine Intelligence 20
WhatELSEcanDeepLearningdo?
http://karpathy.github.io/2015/05/21/rnn-effectiveness/
Image captioning:
“black and white dog jumps over bar.”
Deep Learning can generate code, captions, language, etc.
DL twitter bot
Generated (bogus) math proof:
H2O.ai Machine Intelligence 21
http://techcrunch.com/2015/01/14/amaaaaaazing/
Quest Visual (acquired by Google)
WhatELSEcanDeepLearningdo?
Deep Learning can translate any language
H2O.ai Machine Intelligence 22
WhatELSEcanDeepLearningdo?
Deep Learning can create masterpieces: Semantic Style Transfer
https://github.com/alexjc/neural-doodle
H2O.ai Machine Intelligence 23
StateofAItoday
Today’s A.I. can simulate human behavior, but only for the tasks it was programmed for.
It’s just very good at some board games. And at driving cars. And at speaking any language. And at handwriting. And at much more, but it doesn’t generalize to arbitrary capabilities.
It doesn’t pour itself a scotch (but it could). It doesn’t practice putting (but it could). It doesn’t fall in love (we think).
H2O.ai Machine Intelligence 24
Whatnext?
Anything that can be automated will be automated.
Jobs of the past: assembly line work, teller (ATMs), taxi-firm receptionist
Jobs being automated away now: resume matching, driving, language translation, education
Jobs being automated away soon: healthcare, arts & crafts, entertainment, design, decoration, software engineering, politics, management, professional gaming, financial planning, auditing, real estate agent
Jobs of the future: professional sports, food & wine reviewer
H2O.ai Machine Intelligence 25
Maybewe’llfinallygettheeatingmachine?
https://www.youtube.com/watch?v=n_1apYo6-Ow
1936 Modern Times (Charlie Chaplin)
H2O.ai Machine Intelligence 26
Whatwillchange?
Jerry Kaplan, Stanford http://research.microsoft.com/apps/video/default.aspx?id=258318
Future headlines
“In a few decades a cheap computer may compute as much as all human brains combined — and everything is going to change; every aspect of civilization is going to be affected and transformed by that.”
- Jürgen Schmidhuber (Deep Learning pioneer)http://www.ibtimes.co.uk/what-are-ai-neural-networks-how-are-they-applied-financial-markets-1545896
H2O.ai Machine Intelligence 27
Technologyhasalwayscreatedjobs
U.S. workforce in agriculture 1800: 90% 2000: <2%
A.I. in software: Today: 1% of all software apps use A.I. 2018: 50% of developers will use A.I. in their software (IDC)
“Intelligent software applications will become commonplace. And machine learning will touch every industry.”
— Jeff Dean, Google Brainhttp://www.nytimes.com/2016/03/26/technology/the-race-is-on-to-control-artificial-intelligence-and-techs-future.html
H2O.ai Machine Intelligence 28
H2O.ai-MakersofH2O
H2O-SoftwareProduct• ApacheV2opensource(github.com/h2oai)• ScalableDataSciencePlatform(AIEngineforBusinessTransformation)• DistributedColumnarDataFrameandMap/ReduceBackend• DeepLearning,GradientBoosting,RandomForest,GeneralizedLinearModeling,K-Means,PCA,GLRM,…
EasytoTrainandOperationalizeModels• h2o.ai/downloadandrunanywhere,immediately• R,Python,Java,Scala,REST,GUIclientAPIs• Spark(cf.SparklingWater),Hadoop,Standalone• Auto-generatedJavascoringcode• Vibrantusercommunity
H2O.ai Machine Intelligence
ClientAPIs:R,Python,Java,Scala,Flow
29
library(h2o)h2o.init()h2o.deeplearning(x=1:4,y=5,as.h2o(iris))
importh2o
fromh2o.estimators.deeplearningimportH2ODeepLearningEstimator
h2o.init()
dl=H2ODeepLearningEstimator()
dl.train(x=list(range(1,4)),y="Species",training_frame=iris.hex)
import_root_.hex.deeplearning.DeepLearningimport_root_.hex.deeplearning.DeepLearningParametersvaldlParams=newDeepLearningParameters()dlParams._train=iris.hexdlParams._response_column=‘Speciesvaldl=newDeepLearning(dlParams)valdlModel=dl.trainModel.get
Allheavyliftingisdonebythebackend!
Built-inwebGUIandNotebook
H2O.ai Machine Intelligence 30
Gradient Boosting Tree Model
Auto-generated Java scoring code
to Operationalize Data Science
EasilyBringH2OModelsintoProduction
READ MORE
H2O.aiMachine Intelligence
Hear what our customers are saying about H2O…
31
H2O.aiMachine Intelligence
WhatProgressiveSaysAboutH2O
WATCH NOW
WATCH NOW
“H2O is an enabler in how people are thinking about data.”
“We have many plans to use H2O across the different business units.”
32User Based Insurance
H2O.aiMachine Intelligence
WhatMarketShareSaysAboutH2O
“H2O gave us the capability to do Big Modeling. There is no limit to scaling in H2O.”
“Working with the H2O team has been amazing.”
“The business value that we have gained from advanced analytics is enormous.”
WATCH NOW
WATCH NOW
33Digital Marketing
H2O.aiMachine Intelligence
WATCH NOW
WATCH NOW
WhatNielsenCatalinaSaysAboutH2O
“Unlike other systems where I had to buy the whole package and just use 10-20%, I can customize H2O to suit my needs.”
“I am a big fan of open source. H2O is the best fit in terms of cost as well as ease of use and scalability and usability.”
34Consumer Behavior Analytics
H2O.aiMachine Intelligence
WATCH NOW
WATCH NOW
WhatZurichSaysAboutH2O
“Predictive analytics is the differentiator for insurance companies going forward in the next couple of decades.”
“Advanced analytics was one of the key investments that we decided to make.”
35Risk Analytics
H2O.aiMachine Intelligence
WhatCapitalOneSaysAboutH2O
“H2O is a great solution because it's designed to be enterprise ready and can operate on very large datasets.”
”H2O has been a one-stop shop that helps us do all our modeling in one framework.”
”H2O is the best solution to be able to iterate very quickly on large datasets and produce meaningful models.”
WATCH NOW
WATCH NOW
36Financial Services
H2O.ai Machine Intelligence 37
H2ODeepLearningCommunityQuotes
READ MORE
Kaggle challenge 2nd place winner Colin Priest
READ MORE“FormyfinalcompetitionsubmissionIusedanensembleofmodels,including3deeplearningmodelsbuiltwithRandh2o.”
“IdidreallylikeH2O’sdeeplearningimplementationinR,though-theinterfacewasgreat,thebackendextremelyeasytounderstand,anditwasscalableandflexible.DefinitelyatoolI’llbegoingbackto.”
H2O.ai Machine Intelligence 38
READ MORE WATCH NOW
“H2ODeepLearningmodelsoutperformotherGleasonpredictingmodels.”
READ MORE
“…combineADAMandApacheSparkwithH2O’sdeeplearningcapabilitiestopredictanindividual’spopulationgroupbasedonhisorhergenomicdata.Ourresultsdemonstratethatwecanpredicttheseverywell,withmorethan99%accuracy.”
H2ODeepLearningCommunityQuotes
H2O.ai Machine Intelligence 39
BecomeaTop1%-er
Learn the tricks of the trade (changing fast)!
H2O.ai Machine Intelligence 40
Top13Tips&TricksforH2ODeepLearning
KnowyourMath
UseEarlyStopping
UseCheckpointing
UseRegularization
UseN-foldCross-Validation
PerformHyperParameterSearch
KnowyourData(Sparsity/Categoricals)
TuneCommunicationonMulti-Node
EstablishBaselineonHoldoutDataUnderstandModelComplexity
ControlScoringOverhead
DoEnsembling
InspectModelsinFlow
READ MORE WATCH NOW
H2O.ai Machine Intelligence 41
HigherAccuracywithBetterAlgorithms
Feature Engineering, Cross-Validation, Hyper-Parameters, Ensembles <— 2 talks today by Erin LeDell!
http://www.meetup.com/R-Users/events/229156095/
H2O.ai Machine Intelligence 42
H2OBooklets
DOWNLOAD
Get your booklets at the H2O.ai booth!
R Python Deep Learning GLMGBMSparkling Water
H2O OPEN TOUR W W W . O P E N . H 2 O . A I
#GOH2O
H2O.ai Machine Intelligence
Deep Learning is indeed eatingyour lunch — and mine.
Keep up by making your own smarter applications and operationalizing your
data science with H2O!
44
h2o.ai/download https://www.youtube.com/user/0xdata/videos
https://github.com/h2oai/h2o-3 H2O Google Group
@h2oai
KeyTakeaways
Visit our booth #1225 for live demos!
top related