Upload
pietro-leo
View
441
Download
4
Embed Size (px)
Citation preview
@pieroleo
IBM activities to push forwardthe Future of the IT Technology
Executive Architect – IBM Italy CTO for Big Data Analytics & AI – Chief Scientist for Research & Business IBM Italy – IBM Academy of Technology Leadership Team www.pieroleo.com
3
The Future of Man & Machine Relation
The Future of Machine Intelligence
The Future of Transactions
The Future of Problem Solving
The Future of Cyber Security
5 Major Themes
A reading of IBM ResearchTechnologyHot Areas
www.pieroleo.com
4
4. The Future of Problem Solving
5. The Future of Cyber Security
2. The Future ofMachine Intelligence
1. The Future of Man & MachineRelation
3. The Future of Transactions
• Cognitive cybersecurity
• Homomorphicencryption
• Lattice cryptography
• Blockchaineconomies
• AI+Blockchain• Crypto anchors
• Quantum computing
• Edge computing• Nanotechnology
• Foundation of AI• World's smallest
computer• World’s Smallest
Magnet• In-memory
computing
• Artificial Interactions – Soul Machine
• AI systems & reasoning –Debating
• AI systems & sighting
• AI systems & voice
Artificial Intelligence will integrate seamlessly into and enhance the workplace
AI will have enhanced reasoning abilities and will be widely distributed -helping us make decisions instantly
Blockchainswill do for trusted transactions what the Internet did for Information
New methods of computing are emerging to solve problems that classic machines cannot even attempt or consider more effective ways to solve a problem
Within next few years most of the today's cybersecurity methods and protocols will be breached
5
Artificial Intelligence will integrate seamlessly into and enhance the workplace
The Future of Man & MachineRelation
Challenges
www.pieroleo.com
6
The Future of Man & MachineRelation
• Artificial Interactions – Soul Machine • AI systems & reasoning – Debating• AI systems & sighting• AI systems & voice
What we are doing?
www.pieroleo.com
7
Artificial Interactions – Soul Machine
Soul Machine’s Greg Cross Presentation https://www.facebook.com/bwnet.fans/videos/10154882117121837/
Soul Machines ‘digital humans’ have own unique personality and emotional intelligence.
• creating a realistic physiological model,
• the persona, and• the self-learning from each
human interaction.
The IBM Watson platform enables these digital humans to learn an organization's corpus of knowledge and Soul Machines’ human computing engine allows them to embody the core values of the organization they will represent.
Soul Machine Digital Humans: https://www.youtube.com/watch?time_continue=2&v=AzPs7GfOkew
Baby 5.0 - https://www.youtube.com/watch?v=yzFW4-dvFDA
8
The human face of Artificial
Intelligence
Air New Zealand Project: https://www.ibm.com/blogs/ibm-anz/digital-humans/
Soul Machines used neural networks and
biological brain models to bring Sophie to life, powered by a cloud
based Human Computing Engine.
IBM Watson not only did Sophie undergo domain-
specific training, but other aspects such as her Kiwi
accent and facial expressions were also
tweaked to ensure a high level of realism.
9IBM Research Debating Technologies Home Page: http://www.research.ibm.com/haifa/dept/vst/debating.shtml
“Argumentation and debating represent primary intellectual activities of the human mind. People in all societies argue and debate, not only to convince othersof their own opinions but also in order to explore the differences between multiple perspectives and conceptualizations, and to learn from this exploration.” ACL Berlin2016
AI system & reasoning - debatingwww.pieroleo.com
10
IBM Unraveling Language Patterns1. Opponents often argue that the open primary is
unconstitutional. [Open Primaries]
2. Prof. Smith suggested that affirmative action devalues the accomplishments of discriminated groups. [Affirmative Action]
3. The majority stated that the First Amendment does not guarantee the right to offend others. [Freedom of Speech]
IBM Research GRASP system, -http://www.research.ibm.com/haifa/dept/vst/grasp.shtmlPaper: http://aclweb.org/anthology/D17-1141 GRASP
[someone][argue/suggest/state][that][topic term][sentiment term].[noun][express][that][noun,topic][sentiment],
Language Patterns in
1-2-3
11
AI system & sighting
General Purpose Visual Services
Source IBM Research Computer Vision: http://www.research.ibm.com/cognitive-computing/computer-vision/
Medical Image Analysis
“a person holding a giraffe in their hand”
Video Content Analysis Image Captioning Low-power computer vision - Gesture Recognition
Multimodal Analysis
www.pieroleo.com
12
Source: IBM Research automatic sport highlights generation https://www.ibm.com/blogs/research/2017/06/scaling-wimbledons-video-production-highlight-reels-ai-technology/
13
Source: IBM Research Food Recognition - https://www.ibm.com/blogs/research/2017/05/training-watson-see-whats-plate
14Source: IBM Research Image Caption generation paper - https://arxiv.org/pdf/1612.00563.pdf
“a blue boat is sitting on the side of a building” “a person holding a giraffe in their hand”
15
Source: PowerAI DDL - https://arxiv.org/pdf/1708.02188.pdf
Source: https://www.ibm.com/blogs/research/2017/08/distributed-deep-learning
IBM Research beat Facebook’s time by training the model in 50 minutes, versus the 1 hour Facebook took.
Using this software, IBM Research achieved a new image recognition accuracy of 33.8% for a neural network trained on a very large data set (7.5M images).
The previous record published by Microsoft demonstrated 29.8% accuracy.
Training time: IBM=7 hours, MS=240 hours
ImageNet – 14.197.122 images, 21.841 categories
16
2,400 two-sided telephone conversationsamong 543 speakers (302 male, 241 female) from all areas of the United States.
120 unscripted 30-minute telephoneconversations between native speakers of English.
Most participants called family members or close friends.
Conversation among notknown individuals
SwitchBoard Corpus Call Home Corpus
AI systems & voice
Source: https://catalog.ldc.upenn.edu/ldc97s62 Source: https://catalog.ldc.upenn.edu/ldc97s42
www.pieroleo.com
17
Source: IBM Research - https://www.ibm.com/blogs/research/2017/03/speech-recognition/
Speech recognition
“SWITCHBOARD” corpus “CallHome” corpus
2000: 31.4% (Cambridge University)
2015:14.1% (IBM)2016:12.2% (IBM) - Apr2016: 11.3% (Microsoft) – Oct2017: 11% (Microsoft) – Feb2017: 10.3% IBM (Mar)
Human Parity 6.8% (to 11.3%)
1992: 70%-80%
2000: 19.3% (Cambridge University)2014: 15% (IBM)2015: 8% (IBM)2016: 6,9% (IBM) - Apr2016: 5,9% (Microsoft) - Oct2017: 5.8% (Microsoft) – Feb2017 5.5% IBM (Mar)
Human Parity 5.1% ( to 5.9%)
ErrorRate
ErrorRate
25 years 25 years
www.pieroleo.com
18
IBM Research Voice: http://www.research.ibm.com/haifa/dept/imt/ivvc.shtml
IBM Research Virtual Voice Creator Demo: https://ivva-tts.sl.haifa.il.ibm.com/
US patent application 15/594606. “Text-to-Speech Synthesis with Dynamically-Created Virtual Voices”. Filed on 14 May 2017.A. Sorin, S. Shechtman, and A. Rendel, “Semi Parametric Concatenative TTS with Instant Voice Modification Capabilities”, in Proc. Of INTERSPEECH 2017
The IBM Virtual Voice Creator takesText-To-Speech synthesis technologyto the next level, letting enterprisecustomers and users create uniquevoice personas on-demand in a fast and easy way.
The IBM Virtual Voice Creator letsautomatically create a voiceover for a multi-character game, animation or educational video, without the hassleof hiring voice actors and audio recording studios.
Synthetic speechwww.pieroleo.com
19
AI will have enhanced reasoning abilities and will be widely distributed - helping us make decisions instantly
The Future of Machine Intelligence
Challenges
www.pieroleo.com
20
The Future of Machine Intelligence
• Foundation of AI• World's smallest computer• World’s Smallest Magnet• In-memory computing
What we are doing?
www.pieroleo.com
24
Data
Algorithms
Compute
Foundation of AI
CPU GPU
Deep Neural Networks
A twin-engine aircraft every 12hrGenerate 844TB of Data
www.pieroleo.com
25
Data
Algorithms
Compute
Foundation of AI
CPU GPU
Deep Neural Networks
A twin-engine aircraft every 12hr Generate 844TB of Data
1 Atom =1 bit
= new acceleratingareas
26
Currently, hard disk drives use about 100,000 atoms to store a
single bit.
The storage feat couldsomeday lead to the ability
to store all 35 million songsin the iTunes library on an
area the size of a credit card.
IBM Researchers StoreData on World’s Smallest
Magnet -- a Single Atom
Source: IBM Research - https://www-03.ibm.com/press/us/en/pressrelease/51787.wss
Magnetic bits lie at the heartof hard-disk drives, tape and
next-generation magneticmemory
Details on Nature paper:: http://www.nature.com/articles/nature21371
27
Data
Algorithms
Compute
Foundation of AI
CPU GPU
Deep Neural Networks
Deep Reasoning(Transfer, Trust, Symbolic…)
A twin-engine aircraft every 12hr Generate 844TB of Data
1 Atom =1 bit
= new acceleratingareas
28
Perception
Deep Reason
Classification
Explain
InterpretabilitySymbolicReasoning
Observe
Common-Sense
PlanningPatterns & Sub-patternsObservation
AI Algorithms
….. ….. …..
Ethics
www.pieroleo.com
29
Perception
Deep Reason
Classification
Explain
InterpretabilitySymbolicReasoning
Observe
Common-Sense
PlanningPatterns & Sub-patternsObservation
AI Algorithms
….. ….. …..
EthicsDeep Neural Learning
www.pieroleo.com
31
Perception
Deep Reason
Classification
Explain
InterpretabilitySymbolicReasoning
Observe
Common-Sense
PlanningPatterns & Sub-patternsObservation
AI Algorithms
….. ….. …..
Ethics
See: https://www.ibm.com/watson/advantage-reports/future-of-artificial-intelligence/ai-innovation-equation.html
32
Source: https://www.ibm.com/blogs/think/2017/01/ibm-cognitive-principles/
Source: https://www.research.ibm.com/software/IBMResearch/multimedia/AIEthics_Whitepaper.pdf
www.pieroleo.com
33
Perception
Deep Reason
Classification
Explain
InterpretabilitySymbolicReasoning
Observe
Common-Sense
PlanningPatterns & Sub-patternsObservation
AI Algorithms
….. ….. …..
EthicsDeep Neural Learning
www.pieroleo.com
34
Deep learning basic process
Forward Propagation
Backward Propagation
Multiply + Add
SigmoidSoftMaxreLu
Multiply + Add
Errors
www.pieroleo.com
35
Data
Algorithms
Compute
Foundation of AI
CPU GPUApproximateComputing
In-MemoryComputation
SynapticComputing
5nm – 30B Transistors
Deep Neural Networks
Deep Reasoning(Transfer, Trust, Symbolic…)
A twin-engine aircraft every 12hr Generate 844TB of Data
1 Atom =1 bit
= new acceleratingareas
www.pieroleo.com
36
5 nanometre chips are possible, more powerful, and not too far off in the future.
Compared with 10nm technology available in the market, a nanosheet-basedfive nm technology gives a 40% performance boost.
Significant boost to meeting the future demands of artificialintelligence systems, virtualreality and mobile devices.
The world's first five-nanometre silicon chip. A finger-nail sized chips with 30 billion transistors.
Source: IBM Research 30B transistors chip https://www.ibm.com/blogs/think/2017/06/5-nanometer-transistors/
www.pieroleo.com
37
http://science.sciencemag.org/content/345/6197/668IBM Research Brain Chip - http://www.research.ibm.com/articles/brain-chip.shtml
In 2014, IBM presented 1M spiking-neuron chip with a scalable communication network and interface. The chip has 5.4 billion transistors, 4096 neuro-synaptic cores and 256 million configurable synapses.
Neuromorphic Computing – IBM True North
Source: Introduction - https://www.youtube.com/watch?time_continue=3&v=jqI0L44yFEo
www.pieroleo.com
38
Source: IBM Research Gesture recognition at Low power devices - https://www.ibm.com/blogs/research/2017/07/brain-inspired-cvpr-2017/
Source Video: https://www.youtube.com/watch?v=g08IW-qRomM
Trained a spiking neural network to recognize 10 hand gestures in real-time at 96.5 percent accuracy within a tenth of a second from the start of each gesture, while consuming under 200 mW – much lower power than frame-based systems, which use traditional processors.
39
• 64 millionneurons
• 16 billionsynapses,
• Processor component willconsume the energy equivalentof a dim light bulb– a mere 10 wattsto power.
Source: IBM Research new TrueNorth project update https://www-03.ibm.com/press/us/en/pressrelease/52657.wss
U.S. Air Force Research Lab Taps IBM to Build Brain-Inspired AI Supercomputing System
www.pieroleo.com
40
In-memory Computing with 1 Million Devices for Applications in AI
Source IBM Research Phase Change Memory: https://www.ibm.com/blogs/research/2017/10/ibm-scientists-demonstrate-memory-computing-1-million-devices-applications-ai/
IBM Researchdemonstrated that an unsupervised machine-learning algorithm, running on one millionphase change memory(PCM) devices, successfully foundtemporal correlations in unknown data streams.
When compared to state-of-the-art classical computers, this prototype technology is expected to yield 200x improvements in both speed and energy efficiency
Source: http://www.nature.com/articles/s41467-017-01481-9
PCM Device Collocated Memory and computing
41
IBM and MIT to Pursue Joint Research in Artificial Intelligence, Establish New MIT–IBM Watson AI Lab
IBM plans to make a 10-Year, $240 MillionInvestment in new lab with MIT to advance AI hardware and software and algorithms
Advancing shared prosperity through AI
• AI & economic and societal benefits to a broader range of people, nations, and enterprises.
• AI & Ethics
AI algorithms
Developing advanced algorithms to expand capabilities in machine learning and reasoning.
Physics of AI
• AI hardware materials, devices, and architectures that will support future analog computational approaches to AI
• Quantum computing & Machine learning.
AI & industries
• Healthcare• Cyber Security
Source: http://mitibmwatsonailab.mit.edu/
www.pieroleo.com
42
Blockchains will do for trusted transactions what the Internet did for Information
The Future of Transactions
Challenges
www.pieroleo.com
43
The Future of Transactions
• Blockchain economies• AI+Blockchain• Crypto anchors
What we are doing?
www.pieroleo.com
44
Welcome to the Blockchain economy!
Source IBM IBV Studies https://www-935.ibm.com/services/us/gbs/thoughtleadership/blockchainlibrary.html
2017
2016
Already Active in Blockchain
Investigating
Source: IBM IBV Surveyof 3,000 global C-suite executives
www.pieroleo.com
45
Welcome to the Blockchain economy: the Charm effect
Already Active in Blockchain
Investigating
Source: https://en.wikipedia.org/wiki/Crossing_the_Chasm
Crossing the Chasm: Marketing and Selling High-Tech Products to Mainstream Customers
According to Moore, the marketer should focus on one group of customers at a time, using each group as a base for marketing to the next group. The most difficult step is making the transition between visionaries (early adopters) and pragmatists (early majority). This is the chasm that he refers to. If a successful firm can create a bandwagon effect in which enough momentum builds, then the product becomes a de facto standard.
Chasm
www.pieroleo.com
46
Blockchain & AI
Artificial Intelligence-driven analysis are
performed directly on the blockchain, cross-checking a myriad of
regulations and records as well as supply chain and IoT data, including time, date stamps and geospatial, as well as
visual information.
Source: IBM Research combines AI and Blockchain: https://www.ibm.com/blogs/research/2017/05/power-blockchain-watson/
Interconnect 2017 – Everledger Diamonds Tracking on Blockchaiins - https://www.youtube.com/watch?v=-zYnYXpmtoc
Advanced computer vision AI identifies a diamond from its natural properties, which can’t be changed
Automatic generation of smart contracts from the Kimberly Process certification scheme
40 metadata points
Everledger Diamonds
www.pieroleo.com
47
Crypto Anchors
IBM Research crypto Anchors for Diagnostic devices: https://www.zurich.ibm.com/st/precision_diagnostics/cryptoanchors.html
Crypto anchors extend blockchain’sValue Into the physical realm. Tamper-proof signatures will authenticate products, from medicin to diamonds and make counterfeit nearly impossible
Crypto anchors come in many forms. They can even be embedded into an editable shade of a magnetic ink, which can be used to dye a malaria pill, for example
IBM Research Crypto Anchor for diagnostic devices
www.pieroleo.com
48
The Future of Problem Solving
• Quantum computing• Edge computing• Nanotechnology
What we are doing?
www.pieroleo.com
49
New methods of computing are emerging to solve problems that classic machines cannot even attempt or consider more effective ways to solve a problem
The Future of Problem Solving
Challenges
www.pieroleo.com
50
The future is quantum
• A 20 qubits available on cloud to our clients at the end of this year (already made available 5-16 and 17 qubits)
• A prototype with 50 qubits has been successful tested with a double of coherence time with respect previous systems.
• An open source Quantum Information software (Qikit)
• Quantum Computing: Breaking Through the 49 Qubit Simulation Barrier
.Source: https://www.ibm.com/blogs/research/2017/11/the-future-is-quantumSource: https://www.ibm.com/blogs/research/2017/10/quantum-computing-barrier/
www.pieroleo.com
51
Edge Computing - Mesh Networks
IBM Research Edge Computing - https://www.ibm.com/blogs/research/2017/02/bringing-edge-computing-to-life/Mesh Network Alerts from IBM and The Weather Company - https://www.youtube.com/watch?time_continue=5&v=IITqVRWvDAw
The idea behind edge computing is to process the data right where it isgenerated.
A stadium has potentially more rawcapacity than some of the most powerfulsupercomputers in the world when it’s full of twenty thousand people carryingsmartphones that sport a powerful CPU, array of sensors, storage and multiple radios for communications – and can be connected to one another.
Add tablets, drones, connected vehicles, and other smart devices into the mix, and you have beginnings of the next paradigmof computing.Available n 54 countries
www.pieroleo.com
52
Separating bioparticles at the nanoscale for early disease detection - NanoDLD
New chip technology from IBM Research scientists could enable physicians to perform pre-symptomatic disease detection with a diagnostic device that separate interesting nanoscale-sized biological targets, readily available in saliva and urine.
.Access to bioparticicles such as DNA, viruses and exosomes
IBM Research NanoDLD device - https://www-03.ibm.com/press/us/en/pressrelease/50275.wss
www.pieroleo.com
53
Within next few years most of the today's cybersecurity methods and protocols will be breached
The Future of Cyber Security
Challenge
54
The Future of Cyber Security
• Cognitive cybersecurity• Homomorphic encryption• Lattice cryptography
What we are doing?
www.pieroleo.com
55
Cognitive Threats Maps
• Gain local context leading to the incident
• Perform threat research and develop expertise
• Formulate a threat research strategy
• Apply intelligence to understand the threat
Sogeti Gets 50% FasterAnalysis Times
Source Identify and Understand threats with Watson for Cyber Security: https://www.youtube.com/watch?v=O4qftNyqiQA
www.pieroleo.com
56
Source: G. Gentry, IBM Research, Computing Arbitrary Functions of Encrypted Data – ACM http://ece.gmu.edu/coursewebpages/ECE/ECE646/F10/project/F10_Project_Resources/Computing_Arbitrary_Functions_of_Encrypted_Data.pdf
Usable Full Homomorphic Encryption
2010 2011 2012 2013 2014
Estimated amortized timefor computing a single bitoperation on encrypted data
Moore’s law
2009-10: Plausibility[GH’11] A single bit operation takes 30 minutes
IBM Patent 2013
2011-2012: “Real Circuits”
[GHS’12] A 30,000-gate in 36 hours
2013-today: UsabilityHElib [HS’14]: IBM’s open-source FHE implementationBasis for “generic” FHE computation
An “assembly language” for FHEImplements Brakerski-Gentry-Vaikuntanathan (BGV) scheme
Security based on ring LWE (RLWE)The same 30,000-gate circuit in 4-15 minutes
www.pieroleo.com
1. How to share genomic data in a way that preserves the privacy of the data donors, without undermining the utility of the data or impeding its convenient dissemination?
2. How to perform a LARGE--SCALE, PRVIACY--PRESERVING analysis on genomic data, in an untrusted cloud environment or across multiple users?
• Fast drop in the cost of genome--sequencing 2000: $3 billion
• Mar. 2014: $1,000 • Genotyping 1M variations: below
$200
Unleashing the potenDal of the technology• Healthcare: e.g., disease risk
detection, personalized• medicine • Biomedical research: e.g., • geno-phono• association• Legal and forensic• DTC: e.g., ancestry test, paternity
test
Full Homomorphic EncryptionExamples – Genome Analysis
Source: The 2nd Comparison on Critical Assessment of Data Privacy and Protect Secure Genome Analysis http://www.humangenomeprivacy.org/2015/slides/003_iDASH%20workshop%202015_setStage.pdf
Computing the Hamming distance between two Human Genomes under a Full Homomorphic Encryption schema
https://arxiv.org/pdf/1512.04965.pdf
“One of our main findings is that the number of logical qubits required to implement a Grover attack on AES is relatively low, namely between around 3000 and 7000 logical qubits.”
“When realizing AES, only SubBytes Involves T-gates. Moreover, SubBytes is called a minimum of 296 times as in AES-128 and up to 420 times in AES-256. As shown above, for all three standardized key lengths, this results in quantum circuits of quite moderate complexity. So it seems prudent to move away from 128-bit keys when expecting the availability of at least a moderate size quantum computer.”
Source: https://www.iad.gov/iad/customcf/openAttachment.cfm?FilePath=/iad/library/ia-guidance/ia-solutions-for-classified/algorithm-guidance/assets/public/upload/CNSA-Suite-and-Quantum-Computing-FAQ.pdf
Commercial National Security Algorithm Suite
05 January 2016
Source: IBM Researcher comments https://securityintelligence.com/preparing-next-era-computing-quantum-safe-cryptography/
www.pieroleo.com
Assumptions for Quantum-Resilient Public-Key Cryptography
N=pqgx=y mod p
Factoring is hard
Finding short vectors in lattices is hard
Computing discrete logs is hard
(f,g) in Z[x]/(xn+1)
64
Lattice-BasedCryptography
www.pieroleo.com