Melanie Swan
Philosophy & Economic Theory
New School for Social Research, NY NY
melanie@BlockchainStudies.org
Pfizer, New York NY, March 30, 2017
Slides: http://slideshare.net/LaBlogga
Philosophy of Deep Learning:
Deep Qualia, Statistics, and Blockchain
Image credit: Nvidia
30 Mar 2017
Deep Learning
ASA P value misuse statement
1
Source: http://www.nature.com/news/statisticians-issue-warning-over-misuse-of-p-values-1.19503,
http://amstat.tandfonline.com/doi/abs/10.1080/00031305.2016.1154108
 ASA principles to guide P value use
 The P value alone cannot determine whether a
hypothesis is true or whether results are important
30 Mar 2017
Deep Learning 2
Melanie Swan
 Philosophy and Economic Theory, New School
for Social Research, New York NY
 Founder, Institute for Blockchain Studies
 Singularity University Instructor; Institute for Ethics and
Emerging Technology Affiliate Scholar; EDGE
Essayist; FQXi Advisor
Traditional Markets Background
Economics and Financial
Theory Leadership
New Economies research group
Source: http://www.melanieswan.com, http://blockchainstudies.org/NSNE.pdf, http://blockchainstudies.org/Metaphilosophy_CFP.pdf
https://www.facebook.com/groups/NewEconomies
30 Mar 2017
Deep Learning
Deep Learning vocabulary
What do these terms mean?
 Deep Learning, Machine Learning, Artificial Intelligence
 Deep Belief Net
 Perceptron, Artificial Neuron
 MLP/RELU: Multilayer Perceptron
 Artificial Neural Net
 TensorFlow, Caffe, Theano, Torch, DL4J
 Recurrent Neural Nets
 Boltzmann Machine, Feedforward Neural Net
 Open Source Deep Learning Frameworks
 Google DeepDream, Google Brain, Google DeepMind
3
30 Mar 2017
Deep Learning
Key take-aways
1. What is deep learning?
 Advanced statistical method using logistic regression
 Deep learning is a sub-field of machine learning and
artificial intelligence
2. Why is deep learning important?
 Crucial method of algorithmic data manipulation
3. What do I need to know (as a data scientist)?
 Awareness of new methods like deep learning needed to
keep pace with data growth
4
30 Mar 2017
Deep Learning
Deep Learning and Data Science
5
 Not optional: older algorithms cannot perform to
generate requisite insights
Source: http://blog.algorithmia.com/introduction-to-deep-learning-2016
30 Mar 2017
Deep Learning
Agenda
 Deep Learning Basics
 Definition, operation, drawbacks
 Implications of Deep Learning
 Deep Learning and the Brain
 Deep Learning Blockchain Networks
 Philosophy of Deep Learning
6
Image Source: http://www.opennn.net
30 Mar 2017
Deep Learning
Deep Learning Context
7
Source: Machine Learning Guide, 9. Deep Learning
30 Mar 2017
Deep Learning
Deep Learning Definition
“machines that learn to represent the world” – Yann LeCun
 Deep learning is a class of machine learning algorithms
that use a cascade of layers of processing units to
extract features from data
 Each layer uses the output from the previous layer as input
 Two kinds of learning algorithms
 Supervised (classify labeled data)
 Unsupervised (find patterns in unlabeled data)
 Two phases: training (existing data) and test (new data)
8
Source: Wikiepdia, http://spectrum.ieee.org/automaton/robotics/artificial-intelligence/facebook-ai-director-yann-lecun-on-deep-
learning
30 Mar 2017
Deep Learning
What is Learning? When algorithms detect a
system’s features or rules
9
Single-purpose AI: Deep Blue, 1997
Hard-coded rules
Multi-purpose AI structure: AlphaGo, 2016
Algorithm-detected rules, reusable template
Deep Learning machine
General purpose AI: Deep Qualia, 2xxx?
Novel situation problem-solving,
Algorithm edits/writes rules
Question-answering AI: Watson, 2011
Natural-language processing
Deep Learning prototype
30 Mar 2017
Deep Learning
Deep Learning: what is the problem space?
10
Source: Yann LeCun, CVPR 2015 keynote (Computer Vision ), "What's wrong with Deep Learning" http://t.co/nPFlPZzMEJ
 Level 1 – basic application areas
 Image, text, speech recognition
 Multi-factor recognition (label image with text)
 Sentiment analysis
 Level 2 – complex application areas
 Autonomous driving
 Disease diagnosis, tumor recognition, X-ray/MRI interpretation
 Seismic analysis (earthquake, energy, oil and gas)
30 Mar 2017
Deep Learning
Deep Learning Taxonomy
High-level fundamentals of machine learning
11
Source: Machine Learning Guide, 9. Deep Learning;
AI (artificial intelligence)
Machine learning Other methods
Supervised learning
(labeled data:
classification)
Unsupervised learning
(unlabeled data: pattern
recognition)
Reinforcement learning
Shallow learning
(1-2 layers)
Deep learning
(5-20 layers (expensive))
Recurrent nets (text, speech)
Convolutional nets (images)
Neural Nets (NN) Other methods
Bayesian inference
Support Vector Machines
Decision trees
K-means clustering
K-nearest neighbor
30 Mar 2017
Deep Learning
What is the problem? Computer Vision
(and speech and text recognition)
12
Source: Quoc Le, https://arxiv.org/abs/1112.6209; Yann LeCun, NIPS 2016,
https://drive.google.com/file/d/0BxKBnD5y2M8NREZod0tVdW5FLTQ/view
Marv Minsky, 1966
“summer project”
Jeff Hawkins, 2004,
Hierarchical Temporal
Memory (HTM)
Quoc Le, 2011, Google
Brain cat recognition
Yann LeCun, 2016,
Predictive Learning,
Convolutional net for driving
30 Mar 2017
Deep Learning
Image Recognition: Basic Concept
13
Source: https://developer.clarifai.com/modelshttps://developer.clarifai.com/models
How many orange pixels?
Apple or Orange? Melanoma risk or healthy skin?
Degree of contrast in photo colors?
30 Mar 2017
Deep Learning
Regression (review)
 Linear regression
 Predict continuous set of values
(house prices)
 Logistic regression
 Predict binary outcomes (0,1)
14
Logistic regression
(sigmoid function)
Linear regression
30 Mar 2017
Deep Learning
Deep Learning Architecture
15
Source: Michael A. Nielsen, Neural Networks and Deep Learning
30 Mar 2017
Deep Learning
Example: Image recognition
1. Obtain training data set
2. Digitize pixels (convert images to numbers)
 Divide image into 28x28 grid, assign a value (0-255) to each
square based on brightness
3. Read into vector (array) (28x28 = 784 elements per image)
16
Source: Quoc V. Le, A Tutorial on Deep Learning, Part 1: Nonlinear Classifiers and The Backpropagation Algorithm, 2015, Google
Brain, https://cs.stanford.edu/~quocle/tutorial1.pdf
30 Mar 2017
Deep Learning
Deep Learning Architecture
4. Load spreadsheet of vectors into deep learning system
 Each row of spreadsheet is an input
17
Source: http://deeplearning.stanford.edu/tutorial; MNIST dataset: http://yann.lecun.com/exdb/mnist
1. Input 2. Hidden layers 3. Output
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Vector data
30 Mar 2017
Deep Learning
What happens in the Hidden Layers?
18
Source: Michael A. Nielsen, Neural Networks and Deep Learning
 First layer learns primitive features (line, edge, tiniest
unit of sound) by finding combinations of the input vector
data that occur more frequently than by chance
 Logistic regression performed and encoded at each processing
node (Y/N (0,1)), does this example have this feature?
 Feeds these basic features to next layer, which trains
itself to recognize slightly more complicated features
(corner, combination of speech sounds)
 Feeds features to new layers until recognizes full objects
30 Mar 2017
Deep Learning
Feature Recognition in the Hidden Layers
19
Source: Jann LeCun, http://www.pamitc.org/cvpr15/files/lecun-20150610-cvpr-keynote.pdf
30 Mar 2017
Deep Learning
What happens in the Hidden Layers?
20
Source: Nvidia
 First hidden layer extracts all possible low-level features
from data (lines, edges, contours), next layers abstract
into more complex features of possible relevance
30 Mar 2017
Deep Learning
Deep Learning
 Core concept:
 Deep Learning
systems learn
increasingly
complex features
21
Source: Andrew Ng
30 Mar 2017
Deep Learning
Deep Learning
 Google Deep Brain recognizes cats
22
Source: Quoc V. Le et al, Building high-level features using large scale unsupervised learning, 2011, https://arxiv.org/abs/1112.6209
30 Mar 2017
Deep Learning
Deep Learning Architecture
23
Source: Michael A. Nielsen, Neural Networks and Deep Learning
1. Input 2. Hidden layers 3. Output guess
(0,1)
30 Mar 2017
Deep Learning
Deep Learning Math
Test new data after system iterates
24
1. Input 2. Hidden layers 3. Output
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Source: http://deeplearning.stanford.edu/tutorial; MNIST dataset: http://yann.lecun.com/exdb/mnist
 Linear algebra: matrix multiplications of input vectors
 Statistics: logistic regression units (Y/N (0,1)), probability
weighting and updating, inference for outcome prediction
 Calculus: optimization (minimization), gradient descent in
back-propagation to avoid local minima with saddle points
Feed-forward pass
(0,1)
0.5
Back-propagation pass; update probabilities
.5.5
.5.5.5
0
01
.75
.25
Inference
Guess
Actual
30 Mar 2017
Deep Learning
Hidden Layer Unit, Perceptron, Neuron
25
Source: http://deeplearning.stanford.edu/tutorial; MNIST dataset: http://yann.lecun.com/exdb/mnist
1. Input 2. Hidden layers 3. Output
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
 Unit (processing unit, logistic regression
unit), perceptron (“multilayer perceptron”),
artificial neuron
30 Mar 2017
Deep Learning
Kinds of Deep Learning Systems
What Deep Learning net to choose?
26
Source: Yann LeCun, CVPR 2015 keynote (Computer Vision ), "What's wrong with Deep Learning" http://t.co/nPFlPZzMEJ
 Supervised algorithms (classify labeled data)
 Image (object) recognition
 Convolutional net (image processing), deep belief
network, recursive neural tensor network
 Text analysis (name recognition, sentiment
analysis)
 Recurrent net (iteration; character level text),
recursive neural tensor network
 Speech recognition
 Recurrent net
 Unsupervised algorithms (find patterns in
unlabeled data)
 Boltzmann machine or autoencoder
30 Mar 2017
Deep Learning
Advanced
Deep Learning Architectures
27
Source: http://prog3.com/sbdm/blog/zouxy09/article/details/8781396
 Deep Belief Network
 Connections between layers not units
 Establish weighting guesses for
processing units before run deep
learning system
 Used to pre-train systems to assign
initial probability weights (more efficient)
 Deep Boltzmann Machine
 Stochastic recurrent neural network
 Runs learning on internal
representations
 Represent and solve combinatoric
problems
Deep
Boltzmann
Machine
Deep
Belief
Network
30 Mar 2017
Deep Learning
Convolutional net: Image Enhancement
 Google DeepDream: Convolutional neural network
enhances (potential) patterns in images; deliberately
over-processing images
28
Source: Georges Seurat, Un dimanche après-midi à l'Île de la Grande Jatte, 1884-1886;
http://web.cs.hacettepe.edu.tr/~aykut/classes/spring2016/bil722; Google DeepDream uses algorithmic pareidolia (seeing an image
when none is present) to create a dream-like hallucinogenic appearance
30 Mar 2017
Deep Learning
How big are Deep Learning systems?
 Google Deep Brain cat recognition, 2011
 1 billion connections, 10 million images (200x200 pixel),
1,000 machines (16,000 cores), 3 days, each instantiation of
the network spanned 170 servers, 20,000 object categories
 State of the art, 2015-2016
 Nvidia facial recognition example, 2016, 100 million images,
10 layers, 18 parameters, 30 exaflops, 30 GPU days
 Google, 11.2-billion parameter system
 Lawrence Livermore Lab, 15-billion parameter system
 Digital Reasoning, 2015, cognitive computing (Nashville TN),
160 billion parameters, trained on three multi-core
computers overnight
29
Source: https://futurism.com/biggest-neural-network-ever-pushes-ai-deep-learning, Digital Reasoning paper:
https://arxiv.org/pdf/1506.02338v3.pdf
30 Mar 2017
Deep Learning
Deep Learning, Deep Flaws?
 Even though now possible, still early days
 Expensive and inefficient, big systems
 Only available to massive data processing
operations (Google, Facebook, Microsoft, Baidu)
 Black box: we don’t know how it works
 Reusable model but still can’t multi-task
 Atari example: cannot learn multiple games
 Drop Asteroids to learn Frogger
 Add common sense to intelligence
 Background information, reasoning, planning
 Memory (update and remember states of the world)
 …Deep Learning is still a Specialty System
30
AlphaGo
applied to
Atari games
Source: http://www.theverge.com/2016/10/10/13224930/ai-deep-learning-limitations-drawbacks
30 Mar 2017
Deep Learning
We had the math, what took so long?
 A) Hardware, software, processing
advances; and B) more data
 Key advances in hardware chips
 GPU chips (graphics processing unit):
3D graphics cards designed to do fast
matrix multiplication
 Google TPU chip (tensor processing
unit): custom ASICs for machine
learning, used in AlphaGo
 Training the amount of data
required was too slow to be useful
 Now can train neural nets quickly, still
expensive
31
Tensor
(Scalar (x,y,z), Vector (x,y,z)3, Tensor (x,y,z)9)
Google TPU chip (Tensor
Processing Unit), 2016
Source: http://www.techradar.com/news/computing-components/processors/google-s-tensor-processing-unit-explained-this-is-what-
the-future-of-computing-looks-like-1326915
30 Mar 2017
Deep Learning
Agenda
 Deep Learning Basics
 Definition, operation, drawbacks
 Implications of Deep Learning
 Deep Learning and the Brain
 Deep Learning Blockchain Networks
 Philosophy of Deep Learning
32
Image Source: http://www.opennn.net
30 Mar 2017
Deep Learning
Deep Learning and the Brain
33
30 Mar 2017
Deep Learning
 Deep learning neural networks are inspired by the
structure of the cerebral cortex
 The processing unit, perceptron, artificial neuron is the
mathematical representation of a biological neuron
 In the cerebral cortex, there can be several layers of
interconnected perceptrons
34
Deep Qualia machine? General purpose AI
Mutual inspiration of neurological and computing research
30 Mar 2017
Deep Learning
Deep Qualia machine?
 Visual cortex is hierarchical with intermediate layers
 The ventral (recognition) pathway in the visual cortex has multiple
stages: Retina - LGN - V1 - V2 - V4 - PIT – AIT
 Human brain simulation projects
 Swiss Blue Brain project, European Human Brain Project
35
Source: Jann LeCun, http://www.pamitc.org/cvpr15/files/lecun-20150610-cvpr-keynote.pdf
30 Mar 2017
Deep Learning
Agenda
 Deep Learning Basics
 Definition, operation, drawbacks
 Implications of Deep Learning
 Deep Learning and the Brain
 Deep Learning Blockchain Networks
 Philosophy of Deep Learning
36
Image Source: http://www.opennn.net
30 Mar 2017
Deep Learning
Deep Learning Blockchain Networks
37
30 Mar 2017
Deep Learning
Blockchain Technology
38
Source: http://www.amazon.com/Bitcoin-Blueprint-New-World-Currency/dp/1491920491
30 Mar 2017
Deep Learning
What is Blockchain Technology?
 Blockchain technology is an Internet-
based ledger system for submitting,
logging, and tracking transactions
 Allows the secure transfer of assets
(like money) and information,
computationally, without a human
intermediary
 Secure asset transfer protocol, like email
 First application is currency (Bitcoin) and
FinTech re-engineering, subsequent
applications in algorithmic data processing
39
Source: Blockchain Smartnetworks, https://www.slideshare.net/lablogga/blockchain-smartnetworks
30 Mar 2017
Deep Learning
Deep Learning Blockchain Networks
Help resolve Deep Learning challenges
40
Source: http://www.melanieswan.com, http://blockchainstudies.org/NSNE.pdf, http://blockchainstudies.org/Metaphilosophy_CFP.pdf
 Deep Learning systems need greater capacity
 Put Deep Learning systems on the Internet in a secure-
trackable, remunerable way; distributed not parallel systems
 Deep Learning systems need more complexity and
side modules
 Instantiate common sense, memory, planning modules
 Deep Learning systems do not reveal what happens
in the hidden layers
 Track arbitrarily-many transactions with smart contracts
 Core blockchain functionality employed
 Automated coordination of massive amounts of operations
via smart contracts (automatically-executing Internet-based
programs)
30 Mar 2017
Deep Learning
Deep Learning systems go online with Blockchain
 Key point is to put Deep
Learning systems on the Internet
 Blockchain is perfect technology
to control secure access, yet
have all of the 24/7 availability,
flexibility, scale, and side
modules needed
 Provide global infrastructure to
work on current problems
 Genomic disease, protein modeling,
financial risk assessment,
astronomical data analysis
41
30 Mar 2017
Deep Learning
 Combine Deep Learning and Blockchain Technology
 Deep learning technology, particularly coupled with blockchain
systems, might create a new kind of global computing platform
 Deep Learning and Blockchains are similar
 Indicative of a shift toward having increasingly sophisticated and
automated computational tools
 Mode of operation of both is making (statistically-supported)
guesses about reality states of the world
 Predictive inference (deep learning) and cryptographic nonce-
guesses (blockchain)
 Current sense-making model of the world, we are guessing at more
complex forms of reality
42
Advanced Computational Infrastructure
Deep Learning Blockchain Networks
30 Mar 2017
Deep Learning
Agenda
 Deep Learning Basics
 Definition, operation, drawbacks
 Implications of Deep Learning
 Deep Learning and the Brain
 Deep Learning Blockchain Networks
 Philosophy of Deep Learning
43
Image Source: http://www.opennn.net
30 Mar 2017
Deep Learning
Philosophy of Deep Learning
44
30 Mar 2017
Deep Learning 45
Human’s Role in the World is Changing
Sparse data we control Abundant data controls us?
 Deep Learning is emphasizing the
presence of Big Data
30 Mar 2017
Deep Learning
Philosophy of Deep Learning - Definition
46
 The Philosophy of Deep Learning is
the branch of philosophy concerned
with the definition, methods, and
implications of Deep Learning
 Internal Industry Practice
 Internal to the field as a generalized
articulation of the concepts, theory, and
systems that comprise the overall use of
deep learning algorithms
 External Social Impact
 External to the field, considering the
impact of deep learning more broadly
on individuals, society, and the world
30 Mar 2017
Deep Learning
3 Kinds of Philosophic Concerns
 Ontology (existence, reality)
 What is it? What is deep learning?
 What does it mean?
 Epistemology (knowledge)
 What knowledge are we gaining from
deep learning?
 What is the proof standard?
 Axiology or Valorization (ethics,
aesthetics)
 What is noticed, overlooked?
 What is ethical practice?
 What is beauty, elegance?
47
Sources: http://www.melanieswan.com/documents/Philosophy_of_Big_Data_SWAN.pdf
30 Mar 2017
Deep Learning
Explanation: does the map fit the territory?
48
1626 map of “the Island of California”
Source: California Is An Island Off the Northerne Part of America; John Speed, "America," 1626, London
 Explanandum
 What is being
explained
 Explanans
 The
explanation
30 Mar 2017
Deep Learning
How do we understand reality?
 Methods, models, and
tools
 Descartes, Optics, 1637
 Deep Learning, 2017
49
30 Mar 2017
Deep Learning
Agenda
 Deep Learning Basics
 Definition, operation, drawbacks
 Implications of Deep Learning
 Deep Learning and the Brain
 Deep Learning Blockchain Networks
 Philosophy of Deep Learning
50
Image Source: http://www.opennn.net
30 Mar 2017
Deep Learning
Key take-aways
 What is deep learning?
 Advanced statistical method using logistic regression
 Deep learning is a sub-field of machine learning and
artificial intelligence
 Why is deep learning important?
 Crucial method of algorithmic data manipulation
 What do I need to know (as a data scientist)?
 Awareness of new methods like deep learning needed to
keep pace with data growth
51
30 Mar 2017
Deep Learning
Conclusion
 Deep learning systems are machine
learning algorithms that learn
increasingly complex feature sets from
data via hidden layers
 Deep qualia systems might be a step
forward in brain simulation in computer
networks and general intelligence
 Next-generation global infrastructure:
Deep Learning Blockchain Networks
merging deep learning systems and
blockchain technology
52
30 Mar 2017
Deep Learning
Resources
53
Distill, a visual,
interactive journal for
machine learning
research
http://distill.pub/
Melanie Swan
Philosophy & Economic Theory
New School for Social Research, NY NY
melanie@BlockchainStudies.org
Philosophy of Deep Learning:
Deep Qualia, Statistics, and Blockchain
Pfizer, New York NY, March 30, 2017
Slides: http://slideshare.net/LaBlogga
Thank You! Questions?
Image credit: Nvidia

Philosophy of Deep Learning

  • 1.
    Melanie Swan Philosophy &Economic Theory New School for Social Research, NY NY melanie@BlockchainStudies.org Pfizer, New York NY, March 30, 2017 Slides: http://slideshare.net/LaBlogga Philosophy of Deep Learning: Deep Qualia, Statistics, and Blockchain Image credit: Nvidia
  • 2.
    30 Mar 2017 DeepLearning ASA P value misuse statement 1 Source: http://www.nature.com/news/statisticians-issue-warning-over-misuse-of-p-values-1.19503, http://amstat.tandfonline.com/doi/abs/10.1080/00031305.2016.1154108  ASA principles to guide P value use  The P value alone cannot determine whether a hypothesis is true or whether results are important
  • 3.
    30 Mar 2017 DeepLearning 2 Melanie Swan  Philosophy and Economic Theory, New School for Social Research, New York NY  Founder, Institute for Blockchain Studies  Singularity University Instructor; Institute for Ethics and Emerging Technology Affiliate Scholar; EDGE Essayist; FQXi Advisor Traditional Markets Background Economics and Financial Theory Leadership New Economies research group Source: http://www.melanieswan.com, http://blockchainstudies.org/NSNE.pdf, http://blockchainstudies.org/Metaphilosophy_CFP.pdf https://www.facebook.com/groups/NewEconomies
  • 4.
    30 Mar 2017 DeepLearning Deep Learning vocabulary What do these terms mean?  Deep Learning, Machine Learning, Artificial Intelligence  Deep Belief Net  Perceptron, Artificial Neuron  MLP/RELU: Multilayer Perceptron  Artificial Neural Net  TensorFlow, Caffe, Theano, Torch, DL4J  Recurrent Neural Nets  Boltzmann Machine, Feedforward Neural Net  Open Source Deep Learning Frameworks  Google DeepDream, Google Brain, Google DeepMind 3
  • 5.
    30 Mar 2017 DeepLearning Key take-aways 1. What is deep learning?  Advanced statistical method using logistic regression  Deep learning is a sub-field of machine learning and artificial intelligence 2. Why is deep learning important?  Crucial method of algorithmic data manipulation 3. What do I need to know (as a data scientist)?  Awareness of new methods like deep learning needed to keep pace with data growth 4
  • 6.
    30 Mar 2017 DeepLearning Deep Learning and Data Science 5  Not optional: older algorithms cannot perform to generate requisite insights Source: http://blog.algorithmia.com/introduction-to-deep-learning-2016
  • 7.
    30 Mar 2017 DeepLearning Agenda  Deep Learning Basics  Definition, operation, drawbacks  Implications of Deep Learning  Deep Learning and the Brain  Deep Learning Blockchain Networks  Philosophy of Deep Learning 6 Image Source: http://www.opennn.net
  • 8.
    30 Mar 2017 DeepLearning Deep Learning Context 7 Source: Machine Learning Guide, 9. Deep Learning
  • 9.
    30 Mar 2017 DeepLearning Deep Learning Definition “machines that learn to represent the world” – Yann LeCun  Deep learning is a class of machine learning algorithms that use a cascade of layers of processing units to extract features from data  Each layer uses the output from the previous layer as input  Two kinds of learning algorithms  Supervised (classify labeled data)  Unsupervised (find patterns in unlabeled data)  Two phases: training (existing data) and test (new data) 8 Source: Wikiepdia, http://spectrum.ieee.org/automaton/robotics/artificial-intelligence/facebook-ai-director-yann-lecun-on-deep- learning
  • 10.
    30 Mar 2017 DeepLearning What is Learning? When algorithms detect a system’s features or rules 9 Single-purpose AI: Deep Blue, 1997 Hard-coded rules Multi-purpose AI structure: AlphaGo, 2016 Algorithm-detected rules, reusable template Deep Learning machine General purpose AI: Deep Qualia, 2xxx? Novel situation problem-solving, Algorithm edits/writes rules Question-answering AI: Watson, 2011 Natural-language processing Deep Learning prototype
  • 11.
    30 Mar 2017 DeepLearning Deep Learning: what is the problem space? 10 Source: Yann LeCun, CVPR 2015 keynote (Computer Vision ), "What's wrong with Deep Learning" http://t.co/nPFlPZzMEJ  Level 1 – basic application areas  Image, text, speech recognition  Multi-factor recognition (label image with text)  Sentiment analysis  Level 2 – complex application areas  Autonomous driving  Disease diagnosis, tumor recognition, X-ray/MRI interpretation  Seismic analysis (earthquake, energy, oil and gas)
  • 12.
    30 Mar 2017 DeepLearning Deep Learning Taxonomy High-level fundamentals of machine learning 11 Source: Machine Learning Guide, 9. Deep Learning; AI (artificial intelligence) Machine learning Other methods Supervised learning (labeled data: classification) Unsupervised learning (unlabeled data: pattern recognition) Reinforcement learning Shallow learning (1-2 layers) Deep learning (5-20 layers (expensive)) Recurrent nets (text, speech) Convolutional nets (images) Neural Nets (NN) Other methods Bayesian inference Support Vector Machines Decision trees K-means clustering K-nearest neighbor
  • 13.
    30 Mar 2017 DeepLearning What is the problem? Computer Vision (and speech and text recognition) 12 Source: Quoc Le, https://arxiv.org/abs/1112.6209; Yann LeCun, NIPS 2016, https://drive.google.com/file/d/0BxKBnD5y2M8NREZod0tVdW5FLTQ/view Marv Minsky, 1966 “summer project” Jeff Hawkins, 2004, Hierarchical Temporal Memory (HTM) Quoc Le, 2011, Google Brain cat recognition Yann LeCun, 2016, Predictive Learning, Convolutional net for driving
  • 14.
    30 Mar 2017 DeepLearning Image Recognition: Basic Concept 13 Source: https://developer.clarifai.com/modelshttps://developer.clarifai.com/models How many orange pixels? Apple or Orange? Melanoma risk or healthy skin? Degree of contrast in photo colors?
  • 15.
    30 Mar 2017 DeepLearning Regression (review)  Linear regression  Predict continuous set of values (house prices)  Logistic regression  Predict binary outcomes (0,1) 14 Logistic regression (sigmoid function) Linear regression
  • 16.
    30 Mar 2017 DeepLearning Deep Learning Architecture 15 Source: Michael A. Nielsen, Neural Networks and Deep Learning
  • 17.
    30 Mar 2017 DeepLearning Example: Image recognition 1. Obtain training data set 2. Digitize pixels (convert images to numbers)  Divide image into 28x28 grid, assign a value (0-255) to each square based on brightness 3. Read into vector (array) (28x28 = 784 elements per image) 16 Source: Quoc V. Le, A Tutorial on Deep Learning, Part 1: Nonlinear Classifiers and The Backpropagation Algorithm, 2015, Google Brain, https://cs.stanford.edu/~quocle/tutorial1.pdf
  • 18.
    30 Mar 2017 DeepLearning Deep Learning Architecture 4. Load spreadsheet of vectors into deep learning system  Each row of spreadsheet is an input 17 Source: http://deeplearning.stanford.edu/tutorial; MNIST dataset: http://yann.lecun.com/exdb/mnist 1. Input 2. Hidden layers 3. Output X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X Vector data
  • 19.
    30 Mar 2017 DeepLearning What happens in the Hidden Layers? 18 Source: Michael A. Nielsen, Neural Networks and Deep Learning  First layer learns primitive features (line, edge, tiniest unit of sound) by finding combinations of the input vector data that occur more frequently than by chance  Logistic regression performed and encoded at each processing node (Y/N (0,1)), does this example have this feature?  Feeds these basic features to next layer, which trains itself to recognize slightly more complicated features (corner, combination of speech sounds)  Feeds features to new layers until recognizes full objects
  • 20.
    30 Mar 2017 DeepLearning Feature Recognition in the Hidden Layers 19 Source: Jann LeCun, http://www.pamitc.org/cvpr15/files/lecun-20150610-cvpr-keynote.pdf
  • 21.
    30 Mar 2017 DeepLearning What happens in the Hidden Layers? 20 Source: Nvidia  First hidden layer extracts all possible low-level features from data (lines, edges, contours), next layers abstract into more complex features of possible relevance
  • 22.
    30 Mar 2017 DeepLearning Deep Learning  Core concept:  Deep Learning systems learn increasingly complex features 21 Source: Andrew Ng
  • 23.
    30 Mar 2017 DeepLearning Deep Learning  Google Deep Brain recognizes cats 22 Source: Quoc V. Le et al, Building high-level features using large scale unsupervised learning, 2011, https://arxiv.org/abs/1112.6209
  • 24.
    30 Mar 2017 DeepLearning Deep Learning Architecture 23 Source: Michael A. Nielsen, Neural Networks and Deep Learning 1. Input 2. Hidden layers 3. Output guess (0,1)
  • 25.
    30 Mar 2017 DeepLearning Deep Learning Math Test new data after system iterates 24 1. Input 2. Hidden layers 3. Output X X X X X X X X X X X X X X X Source: http://deeplearning.stanford.edu/tutorial; MNIST dataset: http://yann.lecun.com/exdb/mnist  Linear algebra: matrix multiplications of input vectors  Statistics: logistic regression units (Y/N (0,1)), probability weighting and updating, inference for outcome prediction  Calculus: optimization (minimization), gradient descent in back-propagation to avoid local minima with saddle points Feed-forward pass (0,1) 0.5 Back-propagation pass; update probabilities .5.5 .5.5.5 0 01 .75 .25 Inference Guess Actual
  • 26.
    30 Mar 2017 DeepLearning Hidden Layer Unit, Perceptron, Neuron 25 Source: http://deeplearning.stanford.edu/tutorial; MNIST dataset: http://yann.lecun.com/exdb/mnist 1. Input 2. Hidden layers 3. Output X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X  Unit (processing unit, logistic regression unit), perceptron (“multilayer perceptron”), artificial neuron
  • 27.
    30 Mar 2017 DeepLearning Kinds of Deep Learning Systems What Deep Learning net to choose? 26 Source: Yann LeCun, CVPR 2015 keynote (Computer Vision ), "What's wrong with Deep Learning" http://t.co/nPFlPZzMEJ  Supervised algorithms (classify labeled data)  Image (object) recognition  Convolutional net (image processing), deep belief network, recursive neural tensor network  Text analysis (name recognition, sentiment analysis)  Recurrent net (iteration; character level text), recursive neural tensor network  Speech recognition  Recurrent net  Unsupervised algorithms (find patterns in unlabeled data)  Boltzmann machine or autoencoder
  • 28.
    30 Mar 2017 DeepLearning Advanced Deep Learning Architectures 27 Source: http://prog3.com/sbdm/blog/zouxy09/article/details/8781396  Deep Belief Network  Connections between layers not units  Establish weighting guesses for processing units before run deep learning system  Used to pre-train systems to assign initial probability weights (more efficient)  Deep Boltzmann Machine  Stochastic recurrent neural network  Runs learning on internal representations  Represent and solve combinatoric problems Deep Boltzmann Machine Deep Belief Network
  • 29.
    30 Mar 2017 DeepLearning Convolutional net: Image Enhancement  Google DeepDream: Convolutional neural network enhances (potential) patterns in images; deliberately over-processing images 28 Source: Georges Seurat, Un dimanche après-midi à l'Île de la Grande Jatte, 1884-1886; http://web.cs.hacettepe.edu.tr/~aykut/classes/spring2016/bil722; Google DeepDream uses algorithmic pareidolia (seeing an image when none is present) to create a dream-like hallucinogenic appearance
  • 30.
    30 Mar 2017 DeepLearning How big are Deep Learning systems?  Google Deep Brain cat recognition, 2011  1 billion connections, 10 million images (200x200 pixel), 1,000 machines (16,000 cores), 3 days, each instantiation of the network spanned 170 servers, 20,000 object categories  State of the art, 2015-2016  Nvidia facial recognition example, 2016, 100 million images, 10 layers, 18 parameters, 30 exaflops, 30 GPU days  Google, 11.2-billion parameter system  Lawrence Livermore Lab, 15-billion parameter system  Digital Reasoning, 2015, cognitive computing (Nashville TN), 160 billion parameters, trained on three multi-core computers overnight 29 Source: https://futurism.com/biggest-neural-network-ever-pushes-ai-deep-learning, Digital Reasoning paper: https://arxiv.org/pdf/1506.02338v3.pdf
  • 31.
    30 Mar 2017 DeepLearning Deep Learning, Deep Flaws?  Even though now possible, still early days  Expensive and inefficient, big systems  Only available to massive data processing operations (Google, Facebook, Microsoft, Baidu)  Black box: we don’t know how it works  Reusable model but still can’t multi-task  Atari example: cannot learn multiple games  Drop Asteroids to learn Frogger  Add common sense to intelligence  Background information, reasoning, planning  Memory (update and remember states of the world)  …Deep Learning is still a Specialty System 30 AlphaGo applied to Atari games Source: http://www.theverge.com/2016/10/10/13224930/ai-deep-learning-limitations-drawbacks
  • 32.
    30 Mar 2017 DeepLearning We had the math, what took so long?  A) Hardware, software, processing advances; and B) more data  Key advances in hardware chips  GPU chips (graphics processing unit): 3D graphics cards designed to do fast matrix multiplication  Google TPU chip (tensor processing unit): custom ASICs for machine learning, used in AlphaGo  Training the amount of data required was too slow to be useful  Now can train neural nets quickly, still expensive 31 Tensor (Scalar (x,y,z), Vector (x,y,z)3, Tensor (x,y,z)9) Google TPU chip (Tensor Processing Unit), 2016 Source: http://www.techradar.com/news/computing-components/processors/google-s-tensor-processing-unit-explained-this-is-what- the-future-of-computing-looks-like-1326915
  • 33.
    30 Mar 2017 DeepLearning Agenda  Deep Learning Basics  Definition, operation, drawbacks  Implications of Deep Learning  Deep Learning and the Brain  Deep Learning Blockchain Networks  Philosophy of Deep Learning 32 Image Source: http://www.opennn.net
  • 34.
    30 Mar 2017 DeepLearning Deep Learning and the Brain 33
  • 35.
    30 Mar 2017 DeepLearning  Deep learning neural networks are inspired by the structure of the cerebral cortex  The processing unit, perceptron, artificial neuron is the mathematical representation of a biological neuron  In the cerebral cortex, there can be several layers of interconnected perceptrons 34 Deep Qualia machine? General purpose AI Mutual inspiration of neurological and computing research
  • 36.
    30 Mar 2017 DeepLearning Deep Qualia machine?  Visual cortex is hierarchical with intermediate layers  The ventral (recognition) pathway in the visual cortex has multiple stages: Retina - LGN - V1 - V2 - V4 - PIT – AIT  Human brain simulation projects  Swiss Blue Brain project, European Human Brain Project 35 Source: Jann LeCun, http://www.pamitc.org/cvpr15/files/lecun-20150610-cvpr-keynote.pdf
  • 37.
    30 Mar 2017 DeepLearning Agenda  Deep Learning Basics  Definition, operation, drawbacks  Implications of Deep Learning  Deep Learning and the Brain  Deep Learning Blockchain Networks  Philosophy of Deep Learning 36 Image Source: http://www.opennn.net
  • 38.
    30 Mar 2017 DeepLearning Deep Learning Blockchain Networks 37
  • 39.
    30 Mar 2017 DeepLearning Blockchain Technology 38 Source: http://www.amazon.com/Bitcoin-Blueprint-New-World-Currency/dp/1491920491
  • 40.
    30 Mar 2017 DeepLearning What is Blockchain Technology?  Blockchain technology is an Internet- based ledger system for submitting, logging, and tracking transactions  Allows the secure transfer of assets (like money) and information, computationally, without a human intermediary  Secure asset transfer protocol, like email  First application is currency (Bitcoin) and FinTech re-engineering, subsequent applications in algorithmic data processing 39 Source: Blockchain Smartnetworks, https://www.slideshare.net/lablogga/blockchain-smartnetworks
  • 41.
    30 Mar 2017 DeepLearning Deep Learning Blockchain Networks Help resolve Deep Learning challenges 40 Source: http://www.melanieswan.com, http://blockchainstudies.org/NSNE.pdf, http://blockchainstudies.org/Metaphilosophy_CFP.pdf  Deep Learning systems need greater capacity  Put Deep Learning systems on the Internet in a secure- trackable, remunerable way; distributed not parallel systems  Deep Learning systems need more complexity and side modules  Instantiate common sense, memory, planning modules  Deep Learning systems do not reveal what happens in the hidden layers  Track arbitrarily-many transactions with smart contracts  Core blockchain functionality employed  Automated coordination of massive amounts of operations via smart contracts (automatically-executing Internet-based programs)
  • 42.
    30 Mar 2017 DeepLearning Deep Learning systems go online with Blockchain  Key point is to put Deep Learning systems on the Internet  Blockchain is perfect technology to control secure access, yet have all of the 24/7 availability, flexibility, scale, and side modules needed  Provide global infrastructure to work on current problems  Genomic disease, protein modeling, financial risk assessment, astronomical data analysis 41
  • 43.
    30 Mar 2017 DeepLearning  Combine Deep Learning and Blockchain Technology  Deep learning technology, particularly coupled with blockchain systems, might create a new kind of global computing platform  Deep Learning and Blockchains are similar  Indicative of a shift toward having increasingly sophisticated and automated computational tools  Mode of operation of both is making (statistically-supported) guesses about reality states of the world  Predictive inference (deep learning) and cryptographic nonce- guesses (blockchain)  Current sense-making model of the world, we are guessing at more complex forms of reality 42 Advanced Computational Infrastructure Deep Learning Blockchain Networks
  • 44.
    30 Mar 2017 DeepLearning Agenda  Deep Learning Basics  Definition, operation, drawbacks  Implications of Deep Learning  Deep Learning and the Brain  Deep Learning Blockchain Networks  Philosophy of Deep Learning 43 Image Source: http://www.opennn.net
  • 45.
    30 Mar 2017 DeepLearning Philosophy of Deep Learning 44
  • 46.
    30 Mar 2017 DeepLearning 45 Human’s Role in the World is Changing Sparse data we control Abundant data controls us?  Deep Learning is emphasizing the presence of Big Data
  • 47.
    30 Mar 2017 DeepLearning Philosophy of Deep Learning - Definition 46  The Philosophy of Deep Learning is the branch of philosophy concerned with the definition, methods, and implications of Deep Learning  Internal Industry Practice  Internal to the field as a generalized articulation of the concepts, theory, and systems that comprise the overall use of deep learning algorithms  External Social Impact  External to the field, considering the impact of deep learning more broadly on individuals, society, and the world
  • 48.
    30 Mar 2017 DeepLearning 3 Kinds of Philosophic Concerns  Ontology (existence, reality)  What is it? What is deep learning?  What does it mean?  Epistemology (knowledge)  What knowledge are we gaining from deep learning?  What is the proof standard?  Axiology or Valorization (ethics, aesthetics)  What is noticed, overlooked?  What is ethical practice?  What is beauty, elegance? 47 Sources: http://www.melanieswan.com/documents/Philosophy_of_Big_Data_SWAN.pdf
  • 49.
    30 Mar 2017 DeepLearning Explanation: does the map fit the territory? 48 1626 map of “the Island of California” Source: California Is An Island Off the Northerne Part of America; John Speed, "America," 1626, London  Explanandum  What is being explained  Explanans  The explanation
  • 50.
    30 Mar 2017 DeepLearning How do we understand reality?  Methods, models, and tools  Descartes, Optics, 1637  Deep Learning, 2017 49
  • 51.
    30 Mar 2017 DeepLearning Agenda  Deep Learning Basics  Definition, operation, drawbacks  Implications of Deep Learning  Deep Learning and the Brain  Deep Learning Blockchain Networks  Philosophy of Deep Learning 50 Image Source: http://www.opennn.net
  • 52.
    30 Mar 2017 DeepLearning Key take-aways  What is deep learning?  Advanced statistical method using logistic regression  Deep learning is a sub-field of machine learning and artificial intelligence  Why is deep learning important?  Crucial method of algorithmic data manipulation  What do I need to know (as a data scientist)?  Awareness of new methods like deep learning needed to keep pace with data growth 51
  • 53.
    30 Mar 2017 DeepLearning Conclusion  Deep learning systems are machine learning algorithms that learn increasingly complex feature sets from data via hidden layers  Deep qualia systems might be a step forward in brain simulation in computer networks and general intelligence  Next-generation global infrastructure: Deep Learning Blockchain Networks merging deep learning systems and blockchain technology 52
  • 54.
    30 Mar 2017 DeepLearning Resources 53 Distill, a visual, interactive journal for machine learning research http://distill.pub/
  • 55.
    Melanie Swan Philosophy &Economic Theory New School for Social Research, NY NY melanie@BlockchainStudies.org Philosophy of Deep Learning: Deep Qualia, Statistics, and Blockchain Pfizer, New York NY, March 30, 2017 Slides: http://slideshare.net/LaBlogga Thank You! Questions? Image credit: Nvidia