Post on 19-Apr-2018
Beautiful Ideas
Dr. Dario Gil Vice President Science and Solutions IBM Research
“The thing that differentiates scientists is purely an artistic ability to discern
what is a good idea, what is a beautiful idea, what is worth spending time on, and most importantly,
what is a problem that is sufficiently interesting, yet sufficiently difficult, that it hasn't yet been solved,
but the time for solving it has come now.”
-- Professor Savas Dimopoulos, Stanford University
Brazil
T.J. Watson Almaden
Austin
Ireland Zurich
Haifa
Kenya
India
China
Tokyo
Australia
IBM Research: 3,000 scientists & 12 labs
Six Nobel Laureates
Ten Medals of Technology
Five National Medals of Science
Six Turing Awards
Time
Com
pute
r “In
telli
genc
e”
Counting Machine Circa 1820
ENIAC circa 1945 Antikythera
Astronomical Computer
circa 87 BC
Abacus circa
3500 BC
Napier’s Rods circa 1600
System/360 1964
Deep Blue 1997
Calculators Watson
2011 Calculating Paradigm
Programmable Paradigm
Cognitive Paradigm
The History of Computing
Claim #1
You can not afford to ignore the learning systems trend
Trend #1: Better Machine Learning Algorithms Tom
Mitchell (CMU)
“A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured by P, improves with experience E.”
Introduction of large scale neural networks
Introduction of large scale neural networks
650,000 neurons
5,000,000 neurons
Human Error
Ref: O. Russakovsky et al., arXiv:1409.0575v1 [cs.CV], 1 Sep 2014
Deep Learning
http://www.slideshare.net/NVIDIA/nvidia-ces-2016-press-conference
This challenge evaluates algorithms for object localization/detection and image/scene classification from images and videos at large scale.
‘What does it feel like to teach a machine?’
Jonathan Connell PhD
Trend #2: Massive Datasets (for training)
Social Media
Web Sites
Video Sharing Sites
Curated Data Sets
The Web and the Internet of Things are digitizing the world and the human experience
Trend #3: Performance and Cost of Computing
2011 2,400% improvement in performance and 90% smaller
2015
$17 billion in AI investments since 2009* Trend #4: Massive Talent & Investments Flows
*Source: Quid
1/2
“The business plans of the next 10,000
startups are easy to forecast:
Take X and add AI. This is a big deal, and
now it's here.”
Kevin Kelly (Wired)
© 2016 IBM Corporation
Trend #4: Massive Talent & Investments Flows
2/2
NYC
Cambridge, MA
Munich
Claim #2
The future of expertise will be defined by people and learning systems working collaboratively
Physical limitations
Connectivity limitations
Productivity limitations
Complexity limitations We need enhanced cognitive abilities
Enhancing human capability
“System 1” The automated you
“System 2” The reflective you
Humans incur many cognitive errors and biases
THE AUTOMATIC YOU / THE REFLECTIVE YOU
• Fast • Parallel • Automatic • Effortless • Associative • Slow-learning
• Slow • Serial • Controlled • Effortful • Rule-governed • Flexible
! Cognitive ease " Illusions of truth ! Infers & invents causes and intentions ! Neglects ambiguity ! Is biased to believe and confirm ! Exaggerates emotional consistency ! Focuses on existing evidence and ignores
absent evidence ! Responds more strongly to losses than to gains
Cognitive Optical Errors A few “System 1” challenges
Insights from Behavioral Science
Finance Operations Marketing & Sales Medical Mergers, Acquisitions &
Divestitures Crisis and Emergency
Management Product Pricing &
Launch Discovery
Investment Decisions Project Planning Selection of Markets & Geos Diagnosis
Strategic Planning & Scenario Analysis Discovery & Diagnosis Competitive Analysis Treatment
For Institutions…
Education Large Purchases Financial Investments Medical
Selecting a college Purchasing a home Retirement investment decisions
Selecting medical plans
Financing education Purchasing a car Stock market investments
Deciding on treatment options
For the Individual…
Decisions with a high degree of cognitive complexity
Human Our expertise
Self-directed goals
Common sense
Value judgment
Machine +
All digital knowledge
Large-scale math
Pattern recognition
Statistical reasoning
23
Pushing the frontiers of IT
Source: Kurzweil 1999 – Moravec 1998
1E-5
1E-3
1E+0
1E+3
1E+6
1E+9
1E+12
$100
0 B
uys:
Com
puta
tions
per
sec
ond
1E+15
1900 1920 1940 1960
Integrated Circuit
Discrete Transistor
Vacuum Tube
Electro- Mechanical
Mechanical
2020 and Beyond 2000 1980
Carbon Nanotube
Quantum Technology
A Century of Progress A technological achievement without equal
Min
Dim
ensi
on
Fabr
icat
ed (µ
m)
Moore’s Law (1965)
Semiconductors in Production 100 nm
7nm technology
A unique period in history
© 2016 IBM Corporation
Nanoscopes developed by IBM
Atomic manipulation with the scanning tunneling microscope (STM), single Xe adatoms; D. M. Eigler and E. K. Schweizer, Nature (1990) Olympicene radical (C19H11) imaged with the atomic force microscope (AFM), to scale; L. Gross et al. Science (2009); B. Schuler et al. Phys. Rev. Lett. (2013)
© 2016 IBM Corporation
Physical Analysis: Pushing the Limits of Measurement First complete Measurement of Atomic Structures: atoms, bonds, charge distribution, bond order
Bond Order
A081023.112303.dat Ch: 3 Biasvoltage: 0.16950V Current: 1.1E-10A Temperature: 4.71737 [K]
0 5 10 15 20 250
2
4
6
8
10
12
140
0.1
0.2
0.3
0.4
0.5
AFM
Atomic Positions and Chemical Bonds
Charge Distribution within Molecular Switch Reaction Intermediates (Arynes) Science 337, 1326 (2012) Nature Chemistry 7, 623–628 (2015) Nature Nanotech. 7, 227 (2012)
Science 325, 1110 (2009)
© 2016 IBM Corporation
Molecular identification
First application of AFM for molecular structure identification
A091218.170556.dat C
h: 3 Biasvoltage: -0.15010V Current: 1.2E-12A
Temperature: 4.79412 [K]
02
46
810
1214
1618
024681012141618
0 0.2
0.4
0.6
0.8
1
A091218.170556.dat C
h: 3 Biasvoltage: -0.15010V Current: 1.2E-12A
Temperature: 4.79412 [K]
02
46
810
1214
1618
024681012141618
0 0.2
0.4
0.6
0.8
1
Mariana Trench Challenger Deep (-10911 m)
Kaiko – lost in 2003
Leo Gross et al. Nature Chemistry (2010)
© 2016 IBM Corporation April TK, 2016 | dgil@us.ibm.com | Copyright 2016 IBM Corporation
Bruno Schuler et al., “Reversible Bergman cyclization by atomic manipulation”, Nature Chemistry 8, 220–224 (25 Jan. 2016) doi:10.1038/nchem.2438
Reversible Bergman cyclization by atomic manipulation
31
The$Quantum$Fron.er$$
$$
ROLF LANDAUER
$$
CHARLES BENNETT
ORIGINS OF QUANTUM INFORMATION SCIENCE “Is there a fundamental limit to the energy efficiency of computation?”
“Nature isn’t classical, dammit, and if you want to make a simulation of nature, you’d
better make it quantum mechanical, and by golly, it’s a wonderful problem, because it
doesn’t look so easy.” $
/Richard$P.$Feynman$
NATURE ISN’T CLASSICAL, DAMMIT, AND IF YOU WANT TO MAKE A SIMULATION OF NATURE, YOU’D BETTER MAKE IT QUANTUM MECHANICAL, AND BY GOLLY, IT’S A WONDERFUL PROBLEM, BECAUSE IT DOESN’T LOOK SO EASY.”
RICHARD P. FEYNMAN
“$
Photo: Bomazi
0
1
QUBIT
‘1’
‘0’
‘0’ + ‘1’
BITS
Shor’s'algorithm'(1994)'
Exponential speed-up: A task taking 2100 seconds (1025 days) on a
classical computer might take 100 seconds on a quantum computer
The problem of multiplication vs factoring
937 x 947 = N (easy)
887339 = p x q (harder)
Modulus (1024 bits): de b7 26 43 a6 99 85 cd 38 a7 15 09 b9 cf 0f c9
c3 55 8c 88 ee 8c 8d 28 27 24 4b 2a 5e a0 d8 16 fa 61 18 4b cf 6d 60 80 d3 35 40 32 72 c0 8f 12 d8 e5 4e 8f b9 b2 f6 d9 15 5e 5a 86 31 a3 ba 86 aa 6b c8 d9 71 8c cc cd 27 13 1e 9d 42 5d 38 f6 a7 ac ef fa 62 f3 18 81 d4 24 46 7f 01 77 7c c6
2a 89 14 99 bb 98 39 1d a8 19 fb 39 00 44 7d 1b 94 6a 78 2d 69 ad c0 7a 2c fa d0 da 20 12 98 d3
Public key example:
= p × q
(just short of impossible)
One'of'the'following'must'be'true*:'
– Strong$Church/Turing**$thesis$is$false$
– Factoring$is$easy$
– Quantum$mechanics$is$wrong$
* Scott Aaronson, PhD thesis, UC Berkeley
Shor’s algorithm jumpstarted the interest in quantum computing
** Church-Turing thesis: anything that can be simulated efficiently can be simulated efficiently on existing digital computers
Shor’s algorithm
best classical algorithm (number field sieve)
Classical Record: 230 digits
39$
0$ 1$ 1$ 0$
• Each bit is in a definite state, 0 or 1 • Reading a bit does not change the state • You can copy a bit • All of the information of a bit is stored in that bit
bit 1 bit 2 bit 3 bit 4
Classical$Informa.on$
40$
?$ ?$ ?$ ?$
Entangled$Quantum$Informa.on$
qubit 1
Correlations • Each qubit is in a definite state
• Can be in superposition state – |0> and |1> • Reading a qubit can change the state • You cannot copy a qubit state (no cloning) • Information can be stored in correlations of qubits
• Entanglement: non-classical correlation
qubit 2 qubit 3 qubit 4
Quantum$Informa.on$
41$
• Superposi.on:$$each$qubit$in$2$states$simultaneously$• Entanglement:$qubits$share$informa.on,$so$$
– N$entangled$qubits$in$superposi.on$states$span$all$2N$states$– N$bits$in$a$classical$machine$represent$ONLY$1$out$of$2N$states$$
• Interference:$$– Construc.ve$interference$enhances$correct$answer$–$ONLY$0$or$1$at$end.$– Destruc.ve$interference$suppresses$incorrect$answers$
• Power$grows$exponen.ally$with$number$of$qubits$– Doubles$with$each$added$qubit$
• Compare$to$linear$growth$of$conven.onal:$double$by$doubling$the$number$of$bits…$
Why$are$Some$Quantum$Algorithms$So$Fast?$
Why Quantum Information is Fragile!
Qubits “trap” and control state of single microwave photon or atom Energy state >> Thermal energy, RF noise OR you lose qubit state. - This is why we operate at 0.02 Kelvin… - Why cryostat (refrigerator) is well-shielded… Qubit chip
Cryostat (refrigerator)
0.02 K
0.10 K
0.30 K
Qubit Transistor Quantum System
Conventional System
Vision: A new device defines a new computational system
43
IBM Quantum Processor
A “Small” Quantum Computer
Quantum Computing on the Cloud: The Quantum Experience
Since launch: ! >30,000 users for over 100 countries ! 200,000 experiments ! 8 scientific publications ! >350 major media articles ! 20,000 media mentions ! 138 million Twitter impressions
http://www.research.ibm.com/quantum/
A$final$reflec.on$on$culture$
The Culture of Science The Culture of the Road Map The Culture of Agile
A Story of Three Cultures
DGil@US.IBM.COM | IBM Confidential | © 2016 IBM Corporation 52 25 August, 2016
The Culture of Science
53 DGil@US.IBM.COM | IBM Confidential | © 2016 IBM Corporation
Science “suggests a process of uncommon rationality, inspired observation, and near-saintly tolerance for failure. ...
“The term ‘science’ also entails people aiming high. ...
“In both theory and practice, science … is perceived as a noble endeavor.”1
The culture of science is a culture of engagement with a broad, worldwide community that seeks truths by using proven methods of
enquiry.
In industry research labs, the context of the science is the business. 1 Kevin Kelly, “The Third Culture.” Science, 13 Feb 1998. Vol. 279, Issue 5353, pp. 992-993.
Available at http://science.sciencemag.org/content/279/5353/992
25 August, 2016
Road map culture is always looking up to a vision of where to be 3, 5, and even 10 years in the future.
Experts commit to future targets and delivery dates without fully knowing how to reach them.
Road map culture is intensely team-oriented and unrelenting. A missed target means dropping out of the race.
The Culture of the Road Map
54 DGil@US.IBM.COM | IBM Confidential | © 2016 IBM Corporation 25 August, 2016
Semiconductor Technology Roadmap
55
1998
2014
Area: ~80X ↓ Power: ~20,000X ↓ Speed: ~7-10X ↑
Area: ~14.3X ↓ Power: ~4,000X ↓
Speed: ~7.5X ↑
p substrate, doping α*NA
Scaled Device
L/αxd/α
GATE
n+ source
n+ drain
WIRINGVoltage, V / α
W/αtox/α
R.H. Dennard, IEDM, 72
25 August, 2016
Agile culture practices, encourages and rewards collaboration, speed and value.
Agile is a culture of openness and sharing, where people build for reuse and consumption is measured.
Agile focuses on the rapid innovation of bite-sized pieces that are easy to deploy and consume.
Agile culture is outcome-driven with processes informed by user/value centered design and continuous feedback/learning.
Agile culture is always asking “What is the Minimum Viable Product (MVP)?”
The Culture of Agile
56 DGil@US.IBM.COM | IBM Confidential | © 2016 IBM Corporation 25 August, 2016
The Culture of the Road Map
57 25 August, 2016
The Culture of Agile
The Culture of Science
To solve the most difficult problems in business and the world, we must
incorporate the best in these three complimentary ways of working.
IBM T.J.Watson Research Center, Yorktown Heights, New York
59 DGil@US.IBM.COM | IBM Confidential | © 2016 IBM Corporation 25 August, 2016