SlideShare ist ein Scribd-Unternehmen logo
An Introduction to
Hyperdimensional Computing
Mike Heddes
University of California, Irvine
Department of Computer Science
March 2024
Outline
1. Alphabet
2. Operations
3. Composition
4. Example: The dollar of the Netherlands
5. Application: Graph classification
6. Torchhd: PyTorch library for HD/VSA
1
Origin
Hyperdimensional computing (HD), or vector symbolic architectures (VSA)
Observation
The cerebellum cortex operates on high-dimensional representations [1].
• Forming compositional distributed representations [2, 3]
• The intersection of symbolic and connectionist artificial intelligence
• Symbols → vectors, creating a “concept space”
• Distance metric: angular distance
2
Flavors of HD/VSA
Many variations of HD/VSA [4]:
• Binary Spatter Codes [5]
• Multiply-Add-Permute
• Holographic Reduced Representations [6]
• and more...
3
Alphabet
Within the Multiply-Add-Permute HD/VSA model [7]:
• Symbols x, y ∼ U{−1, 1}d
• Rademacher distribution
〈x, x〉
d
= 1 E

〈x, y〉
d

= 0 Var

〈x, y〉
d

=
1
d
In high-dimensional spaces:
• Inner products are highly concentrated: quasi-orthogonal
• Only d orthogonal symbols
• O(exp(d)) quasi-orthogonal symbols [4]
4
Operations: bundling
• Bundling of symbols: component-wise addition
• Useful for representing a set of symbols
• Result can be normalized to {−1, 1}d
E[〈x, x + y〉] = 〈x, x〉 + E[〈x, y〉] = d
5
Operations: binding
• Binding of symbols: component-wise multiplication (Hadamard product)
• Distributes over bundling
• Self-inverse property
x ◦ y ∈ {−1, 1}d
x ◦ x = 1
• Useful for the association of symbols
E[〈x, x ◦ y〉] = E[〈1, x ◦ x ◦ y〉] = E[〈1, y〉] = 0
E[〈x, (x ◦ y) ◦ y〉] = E[〈x, x〉] = d
6
Operations: permute
• Permute of a symbol: permutation of its components
• Distributes over binding and bundling
• Useful for the assignment of an ordering
〈x, ρ(x)〉 ≈ 0
• We denote ρ(ρ(x)) as ρ2(x)
• Also has an inverse permutation: ρ−1(x)
7
Operations
Operation Notation Domain Codomain
Bundling x + y H × H H or Zd
Binding x ◦ y H × H H
Permute ρ(x) H H
• Where H = {−1, 1}d
• Theses operations enable composition
8
Composition: multiset
• Representation for a set of symbols {x1, x2, . . . , xn}:
s = x1 + x2 + · · · + xn =
n
X
i=1
xi
• Query the frequency of symbol xj:
E

〈xj, s〉
d

=
1
d
n
X
i=1
E[〈xj, xi〉] =
n
X
i=1
1[xj = xi]
• No free lunch: bundling capacity O(d) [8]
9
Composition: sequence
• Representation for a sequence of symbols (x1, x2, . . . , xn):
s = x1 + ρ(x2) + ρ2
(x3) + · · · + ρn−1
(xn) =
n
X
i=1
ρi−1
(xi)
• Let X = [x1, x2, . . . , xn] be the codebook
• Retrieve the (noisy) k-th symbol:
x̃k = ρ1−k
(s) = xk + noise with XT
noise ≈ 0
= ρ1−k
(x1 + ρ(x2) + ρ2
(x3) + · · · + ρn−1
(xn))
= ρ1−k
(x1) + ρ2−k
(x2) + ρ3−k
(x3) + · · · + ρn−k
(xn)
10
Composition: key-value store
• Representation for a set of mappings {(k1 ↔ v1), (k2 ↔ v2), . . . , (kn ↔ vn)}:
s = k1 ◦ v1 + k2 ◦ v2 + · · · + kn ◦ vn =
n
X
i=1
ki ◦ vi
• Retrieve the (noisy) value of the j-th key:
ṽj = s ◦ kj = vj + noise with XT
noise ≈ 0
= (k1 ◦ v1 + k2 ◦ v2 + · · · + kn ◦ vn) ◦ kj
= k1 ◦ kj ◦ v1 + k2 ◦ kj ◦ v2 + · · · + kn ◦ kj ◦ vn
11
Associative memory
• Clean up of a noisy symbol x̃:
cleanup(x̃, X) = lookup(x̃, X, X) with lookup(q, K, V) = V σ
‚
KTq
d
Œ
• σ is a nonlinear function, e.g., ReLU, Top-k, Softmax
• Variations include:
• Hopfield network [9]
• sparse distributed memory [10]
• attention mechanism [11]
12
The dollar of the Netherlands
• Classical example of reasoning by analogy [12] (originally the dollar of Mexico)
• Key-value stores containing country (k1), capitol (k2), and currency (k3):
u = k1 ◦ USA + k2 ◦ WDC + k3 ◦ USD
n = k1 ◦ NLD + k2 ◦ AMS + k3 ◦ EUR
What is the dollar of the Netherlands?
ß
EUR = u ◦ n ◦ USD = EUR + noise with XT
noise ≈ 0
= (k1 ◦ k1 ◦ USA ◦ NLD + k2 ◦ k2 ◦ WDC ◦ AMS + k3 ◦ k3 ◦ USD ◦ EUR
+ k1 ◦ k2 ◦ USA ◦ AMS + k2 ◦ k3 ◦ WDC ◦ EUR + k3 ◦ k1 ◦ USD ◦ NLD
+ k1 ◦ k3 ◦ USA ◦ EUR + k2 ◦ k1 ◦ WDC ◦ NLD + k3 ◦ k2 ◦ USD ◦ AMS) ◦ USD
= USA ◦ NLD ◦ USD + WDC ◦ AMS ◦ USD + EUR + . . .
13
Graph classification
Problem
Given a set of labeled graphs, predict the label of a given unlabeled graph.
Approach
1. Split the set of labeled graphs by their label.
2. Create a vector representation of each graph.
3. Create a vector representation of each set.
4. Find the set most similar to the unlabeled graph.
14
Graph encoding
As proposed in GraphHD [13], given an undirected graph G = (V, E).
1. Sample a (global) set of vertex-rank symbols
2. Rank vertices by their PageRank centrality
3. Assign vertex-rank symbols to vertices
4. Bind vertex symbols to represent an edge
5. Bundle all edge symbols
g =
X
(i,j)∈E
vi ◦ vj
15
Classification overview
Test
data
Train
data
Encoder
Training
Model
M1
M2
...
Mk
Query vector
Similarities
16
Comparison with Perceptron
Model
M1
M2
...
Mk
Query vector
Similarities
HD/VSA classifier
y =
MTq
Æ
tr(MTM)
Æ
qTq
Perceptron
y = WT
x + b
17
Torchhd: PyTorch library for HD/VSA
Enable quick development of performant HD/VSA experiments and applications [14].
• A simple programming experience
• High-performance execution
• Support for a variety of HD/VSA applications and research directions
• Includes many reference implementations
• Open source: https://github.com/hyperdimensional-computing/torchhd/
18
References I
1. Kanerva, P. Hyperdimensional computing: An introduction to computing in distributed representation with
high-dimensional random vectors. Cognitive Computation 1, 139–159 (2009).
2. Smolensky, P. Tensor Product Variable Binding and the Representation of Symbolic Structures in Connectionist
Systems. Artificial Intelligence 46, 159–216 (1990).
3. Plate, T. A. Holographic Reduced Representations: Convolution Algebra for Compositional Distributed
Representations. in International Joint Conference on Artificial Intelligence (IJCAI) (1991), 30–35.
4. Kleyko, D., Rachkovskij, D. A., Osipov, E.  Rahimi, A. A survey on hyperdimensional computing aka vector symbolic
architectures, part i: Models and data transformations. ACM Computing Surveys 55, 1–40 (2022).
5. Kanerva, P. Fully distributed representation. in Proceedings of the Real World Computing Symposium (1997),
358–365.
6. Plate, T. A. Holographic reduced representations. IEEE Transactions on Neural networks 6, 623–641 (1995).
7. Gayler, R. Multiplicative binding, representation operators and analogy. Advances in analogy research: integration
of theory and data from the cognitive, computational and neural sciences (1998).
8. Thomas, A., Dasgupta, S.  Rosing, T. Theoretical Foundations of Hyperdimensional Computing. Journal of Artificial
Intelligence Research (JAIR) 72, 215–249 (2021).
9. Hopfield, J. J. Neural networks and physical systems with emergent collective computational abilities. Proceedings
of the national academy of sciences 79, 2554–2558 (1982).
19
References II
10. Kanerva, P. Sparse distributed memory and related models. Tech. rep. (1992).
11. Vaswani, A. et al. Attention is all you need. Advances in neural information processing systems 30 (2017).
12. Kanerva, P. What we mean when we say” What’s the dollar of Mexico?”: Prototypes and mapping in concept space.
in AAAI fall symposium series (2010).
13. Nunes, I., Heddes, M., Givargis, T., Nicolau, A.  Veidenbaum, A. GraphHD: Efficient graph classification using
hyperdimensional computing. in Design, Automation  Test in Europe Conference  Exhibition (DATE) (2022).
14. Heddes, M. et al. Torchhd: An Open Source Python Library to Support Research on Hyperdimensional Computing
and Vector Symbolic Architectures. Journal of Machine Learning Research 24, 1–10.
http://jmlr.org/papers/v24/23-0300.html (2023).
20

Weitere ähnliche Inhalte

Ähnlich wie 2024.03.22 - Mike Heddes - Introduction to Hyperdimensional Computing.pdf

15_wk4_unsupervised-learning_manifold-EM-cs365-2014.pdf
15_wk4_unsupervised-learning_manifold-EM-cs365-2014.pdf15_wk4_unsupervised-learning_manifold-EM-cs365-2014.pdf
15_wk4_unsupervised-learning_manifold-EM-cs365-2014.pdf
McSwathi
 
Declare Your Language (at DLS)
Declare Your Language (at DLS)Declare Your Language (at DLS)
Declare Your Language (at DLS)
Eelco Visser
 
Seminar PSU 09.04.2013 - 10.04.2013 MiFIT, Arbuzov Vyacheslav
Seminar PSU 09.04.2013 - 10.04.2013 MiFIT, Arbuzov VyacheslavSeminar PSU 09.04.2013 - 10.04.2013 MiFIT, Arbuzov Vyacheslav
Seminar PSU 09.04.2013 - 10.04.2013 MiFIT, Arbuzov Vyacheslav
Vyacheslav Arbuzov
 

Ähnlich wie 2024.03.22 - Mike Heddes - Introduction to Hyperdimensional Computing.pdf (20)

15_wk4_unsupervised-learning_manifold-EM-cs365-2014.pdf
15_wk4_unsupervised-learning_manifold-EM-cs365-2014.pdf15_wk4_unsupervised-learning_manifold-EM-cs365-2014.pdf
15_wk4_unsupervised-learning_manifold-EM-cs365-2014.pdf
 
Relaxation methods for the matrix exponential on large networks
Relaxation methods for the matrix exponential on large networksRelaxation methods for the matrix exponential on large networks
Relaxation methods for the matrix exponential on large networks
 
Codes and Isogenies
Codes and IsogeniesCodes and Isogenies
Codes and Isogenies
 
Pattern learning and recognition on statistical manifolds: An information-geo...
Pattern learning and recognition on statistical manifolds: An information-geo...Pattern learning and recognition on statistical manifolds: An information-geo...
Pattern learning and recognition on statistical manifolds: An information-geo...
 
Hierarchical matrices for approximating large covariance matries and computin...
Hierarchical matrices for approximating large covariance matries and computin...Hierarchical matrices for approximating large covariance matries and computin...
Hierarchical matrices for approximating large covariance matries and computin...
 
Declare Your Language (at DLS)
Declare Your Language (at DLS)Declare Your Language (at DLS)
Declare Your Language (at DLS)
 
Geospatial Data in R
Geospatial Data in RGeospatial Data in R
Geospatial Data in R
 
Lec3
Lec3Lec3
Lec3
 
Seminar PSU 09.04.2013 - 10.04.2013 MiFIT, Arbuzov Vyacheslav
Seminar PSU 09.04.2013 - 10.04.2013 MiFIT, Arbuzov VyacheslavSeminar PSU 09.04.2013 - 10.04.2013 MiFIT, Arbuzov Vyacheslav
Seminar PSU 09.04.2013 - 10.04.2013 MiFIT, Arbuzov Vyacheslav
 
R Language Introduction
R Language IntroductionR Language Introduction
R Language Introduction
 
Slides: A glance at information-geometric signal processing
Slides: A glance at information-geometric signal processingSlides: A glance at information-geometric signal processing
Slides: A glance at information-geometric signal processing
 
Lect4
Lect4Lect4
Lect4
 
QMC Error SAMSI Tutorial Aug 2017
QMC Error SAMSI Tutorial Aug 2017QMC Error SAMSI Tutorial Aug 2017
QMC Error SAMSI Tutorial Aug 2017
 
Deep Learning, Keras, and TensorFlow
Deep Learning, Keras, and TensorFlowDeep Learning, Keras, and TensorFlow
Deep Learning, Keras, and TensorFlow
 
Interactive High-Dimensional Visualization of Social Graphs
Interactive High-Dimensional Visualization of Social GraphsInteractive High-Dimensional Visualization of Social Graphs
Interactive High-Dimensional Visualization of Social Graphs
 
Program on Quasi-Monte Carlo and High-Dimensional Sampling Methods for Applie...
Program on Quasi-Monte Carlo and High-Dimensional Sampling Methods for Applie...Program on Quasi-Monte Carlo and High-Dimensional Sampling Methods for Applie...
Program on Quasi-Monte Carlo and High-Dimensional Sampling Methods for Applie...
 
Sketching and locality sensitive hashing for alignment
Sketching and locality sensitive hashing for alignmentSketching and locality sensitive hashing for alignment
Sketching and locality sensitive hashing for alignment
 
Introduction to Neural Networks and Deep Learning from Scratch
Introduction to Neural Networks and Deep Learning from ScratchIntroduction to Neural Networks and Deep Learning from Scratch
Introduction to Neural Networks and Deep Learning from Scratch
 
Ggplot2 v3
Ggplot2 v3Ggplot2 v3
Ggplot2 v3
 
theory of computation lecture 01
theory of computation lecture 01theory of computation lecture 01
theory of computation lecture 01
 

Mehr von Advanced-Concepts-Team

Pablo Gomez - Solving Large-scale Challenges with ESA Datalabs
Pablo Gomez - Solving Large-scale Challenges with ESA DatalabsPablo Gomez - Solving Large-scale Challenges with ESA Datalabs
Pablo Gomez - Solving Large-scale Challenges with ESA Datalabs
Advanced-Concepts-Team
 
Jonathan Sauder - Miniaturizing Mechanical Systems for CubeSats: Design Princ...
Jonathan Sauder - Miniaturizing Mechanical Systems for CubeSats: Design Princ...Jonathan Sauder - Miniaturizing Mechanical Systems for CubeSats: Design Princ...
Jonathan Sauder - Miniaturizing Mechanical Systems for CubeSats: Design Princ...
Advanced-Concepts-Team
 
EDEN ISS - A space greenhouse analogue in Antarctica
EDEN ISS - A space greenhouse analogue in AntarcticaEDEN ISS - A space greenhouse analogue in Antarctica
EDEN ISS - A space greenhouse analogue in Antarctica
Advanced-Concepts-Team
 
How to give a robot a soul
How to give a robot a soulHow to give a robot a soul
How to give a robot a soul
Advanced-Concepts-Team
 
Vernal pools a new ecosystem for astrobiology studies
Vernal pools a new ecosystem for astrobiology studiesVernal pools a new ecosystem for astrobiology studies
Vernal pools a new ecosystem for astrobiology studies
Advanced-Concepts-Team
 

Mehr von Advanced-Concepts-Team (20)

Isabelle Diacaire - From Ariadnas to Industry R&D in optics and photonics
Isabelle Diacaire - From Ariadnas to Industry R&D in optics and photonicsIsabelle Diacaire - From Ariadnas to Industry R&D in optics and photonics
Isabelle Diacaire - From Ariadnas to Industry R&D in optics and photonics
 
The ExoGRAVITY project - observations of exoplanets from the ground with opti...
The ExoGRAVITY project - observations of exoplanets from the ground with opti...The ExoGRAVITY project - observations of exoplanets from the ground with opti...
The ExoGRAVITY project - observations of exoplanets from the ground with opti...
 
MOND_famaey.pdf
MOND_famaey.pdfMOND_famaey.pdf
MOND_famaey.pdf
 
Pablo Gomez - Solving Large-scale Challenges with ESA Datalabs
Pablo Gomez - Solving Large-scale Challenges with ESA DatalabsPablo Gomez - Solving Large-scale Challenges with ESA Datalabs
Pablo Gomez - Solving Large-scale Challenges with ESA Datalabs
 
Jonathan Sauder - Miniaturizing Mechanical Systems for CubeSats: Design Princ...
Jonathan Sauder - Miniaturizing Mechanical Systems for CubeSats: Design Princ...Jonathan Sauder - Miniaturizing Mechanical Systems for CubeSats: Design Princ...
Jonathan Sauder - Miniaturizing Mechanical Systems for CubeSats: Design Princ...
 
Towards an Artificial Muse for new Ideas in Quantum Physics
Towards an Artificial Muse for new Ideas in Quantum PhysicsTowards an Artificial Muse for new Ideas in Quantum Physics
Towards an Artificial Muse for new Ideas in Quantum Physics
 
EDEN ISS - A space greenhouse analogue in Antarctica
EDEN ISS - A space greenhouse analogue in AntarcticaEDEN ISS - A space greenhouse analogue in Antarctica
EDEN ISS - A space greenhouse analogue in Antarctica
 
How to give a robot a soul
How to give a robot a soulHow to give a robot a soul
How to give a robot a soul
 
Information processing with artificial spiking neural networks
Information processing with artificial spiking neural networksInformation processing with artificial spiking neural networks
Information processing with artificial spiking neural networks
 
Exploring Architected Materials Using Machine Learning
Exploring Architected Materials Using Machine LearningExploring Architected Materials Using Machine Learning
Exploring Architected Materials Using Machine Learning
 
Electromagnetically Actuated Systems for Modular, Self-Assembling and Self-Re...
Electromagnetically Actuated Systems for Modular, Self-Assembling and Self-Re...Electromagnetically Actuated Systems for Modular, Self-Assembling and Self-Re...
Electromagnetically Actuated Systems for Modular, Self-Assembling and Self-Re...
 
HORUS: Peering into Lunar Shadowed Regions with AI
HORUS: Peering into Lunar Shadowed Regions with AIHORUS: Peering into Lunar Shadowed Regions with AI
HORUS: Peering into Lunar Shadowed Regions with AI
 
META-SPACE: Psycho-physiologically Adaptive and Personalized Virtual Reality ...
META-SPACE: Psycho-physiologically Adaptive and Personalized Virtual Reality ...META-SPACE: Psycho-physiologically Adaptive and Personalized Virtual Reality ...
META-SPACE: Psycho-physiologically Adaptive and Personalized Virtual Reality ...
 
The Large Interferometer For Exoplanets (LIFE) II: Key Methods and Technologies
The Large Interferometer For Exoplanets (LIFE) II: Key Methods and TechnologiesThe Large Interferometer For Exoplanets (LIFE) II: Key Methods and Technologies
The Large Interferometer For Exoplanets (LIFE) II: Key Methods and Technologies
 
Black Holes and Bright Quasars
Black Holes and Bright QuasarsBlack Holes and Bright Quasars
Black Holes and Bright Quasars
 
In vitro simulation of spaceflight environment to elucidate combined effect o...
In vitro simulation of spaceflight environment to elucidate combined effect o...In vitro simulation of spaceflight environment to elucidate combined effect o...
In vitro simulation of spaceflight environment to elucidate combined effect o...
 
The Large Interferometer For Exoplanets (LIFE): the science of characterising...
The Large Interferometer For Exoplanets (LIFE): the science of characterising...The Large Interferometer For Exoplanets (LIFE): the science of characterising...
The Large Interferometer For Exoplanets (LIFE): the science of characterising...
 
Vernal pools a new ecosystem for astrobiology studies
Vernal pools a new ecosystem for astrobiology studiesVernal pools a new ecosystem for astrobiology studies
Vernal pools a new ecosystem for astrobiology studies
 
Keeping a Sentinel Eye on the Volcanoes – from Space!
Keeping a Sentinel Eye on the Volcanoes – from Space!Keeping a Sentinel Eye on the Volcanoes – from Space!
Keeping a Sentinel Eye on the Volcanoes – from Space!
 
AI4Space –Artificial Intelligence at ISTA - Hülsmann & Haser
AI4Space –Artificial Intelligence at ISTA - Hülsmann & HaserAI4Space –Artificial Intelligence at ISTA - Hülsmann & Haser
AI4Space –Artificial Intelligence at ISTA - Hülsmann & Haser
 

Kürzlich hochgeladen

extra-chromosomal-inheritance[1].pptx.pdfpdf
extra-chromosomal-inheritance[1].pptx.pdfpdfextra-chromosomal-inheritance[1].pptx.pdfpdf
extra-chromosomal-inheritance[1].pptx.pdfpdf
DiyaBiswas10
 
The importance of continents, oceans and plate tectonics for the evolution of...
The importance of continents, oceans and plate tectonics for the evolution of...The importance of continents, oceans and plate tectonics for the evolution of...
The importance of continents, oceans and plate tectonics for the evolution of...
Sérgio Sacani
 
Cancer cell metabolism: special Reference to Lactate Pathway
Cancer cell metabolism: special Reference to Lactate PathwayCancer cell metabolism: special Reference to Lactate Pathway
Cancer cell metabolism: special Reference to Lactate Pathway
AADYARAJPANDEY1
 
Pests of Green Manures_Bionomics_IPM_Dr.UPR.pdf
Pests of Green Manures_Bionomics_IPM_Dr.UPR.pdfPests of Green Manures_Bionomics_IPM_Dr.UPR.pdf
Pests of Green Manures_Bionomics_IPM_Dr.UPR.pdf
PirithiRaju
 
Mammalian Pineal Body Structure and Also Functions
Mammalian Pineal Body Structure and Also FunctionsMammalian Pineal Body Structure and Also Functions
Mammalian Pineal Body Structure and Also Functions
YOGESH DOGRA
 
Earliest Galaxies in the JADES Origins Field: Luminosity Function and Cosmic ...
Earliest Galaxies in the JADES Origins Field: Luminosity Function and Cosmic ...Earliest Galaxies in the JADES Origins Field: Luminosity Function and Cosmic ...
Earliest Galaxies in the JADES Origins Field: Luminosity Function and Cosmic ...
Sérgio Sacani
 
THYROID-PARATHYROID medical surgical nursing
THYROID-PARATHYROID medical surgical nursingTHYROID-PARATHYROID medical surgical nursing
THYROID-PARATHYROID medical surgical nursing
Jocelyn Atis
 
Pests of sugarcane_Binomics_IPM_Dr.UPR.pdf
Pests of sugarcane_Binomics_IPM_Dr.UPR.pdfPests of sugarcane_Binomics_IPM_Dr.UPR.pdf
Pests of sugarcane_Binomics_IPM_Dr.UPR.pdf
PirithiRaju
 

Kürzlich hochgeladen (20)

SAMPLING.pptx for analystical chemistry sample techniques
SAMPLING.pptx for analystical chemistry sample techniquesSAMPLING.pptx for analystical chemistry sample techniques
SAMPLING.pptx for analystical chemistry sample techniques
 
Lab report on liquid viscosity of glycerin
Lab report on liquid viscosity of glycerinLab report on liquid viscosity of glycerin
Lab report on liquid viscosity of glycerin
 
extra-chromosomal-inheritance[1].pptx.pdfpdf
extra-chromosomal-inheritance[1].pptx.pdfpdfextra-chromosomal-inheritance[1].pptx.pdfpdf
extra-chromosomal-inheritance[1].pptx.pdfpdf
 
GEOLOGICAL FIELD REPORT On Kaptai Rangamati Road-Cut Section.pdf
GEOLOGICAL FIELD REPORT  On  Kaptai Rangamati Road-Cut Section.pdfGEOLOGICAL FIELD REPORT  On  Kaptai Rangamati Road-Cut Section.pdf
GEOLOGICAL FIELD REPORT On Kaptai Rangamati Road-Cut Section.pdf
 
The importance of continents, oceans and plate tectonics for the evolution of...
The importance of continents, oceans and plate tectonics for the evolution of...The importance of continents, oceans and plate tectonics for the evolution of...
The importance of continents, oceans and plate tectonics for the evolution of...
 
Astronomy Update- Curiosity’s exploration of Mars _ Local Briefs _ leadertele...
Astronomy Update- Curiosity’s exploration of Mars _ Local Briefs _ leadertele...Astronomy Update- Curiosity’s exploration of Mars _ Local Briefs _ leadertele...
Astronomy Update- Curiosity’s exploration of Mars _ Local Briefs _ leadertele...
 
GLOBAL AND LOCAL SCENARIO OF FOOD AND NUTRITION.pptx
GLOBAL AND LOCAL SCENARIO OF FOOD AND NUTRITION.pptxGLOBAL AND LOCAL SCENARIO OF FOOD AND NUTRITION.pptx
GLOBAL AND LOCAL SCENARIO OF FOOD AND NUTRITION.pptx
 
word2vec, node2vec, graph2vec, X2vec: Towards a Theory of Vector Embeddings o...
word2vec, node2vec, graph2vec, X2vec: Towards a Theory of Vector Embeddings o...word2vec, node2vec, graph2vec, X2vec: Towards a Theory of Vector Embeddings o...
word2vec, node2vec, graph2vec, X2vec: Towards a Theory of Vector Embeddings o...
 
Cancer cell metabolism: special Reference to Lactate Pathway
Cancer cell metabolism: special Reference to Lactate PathwayCancer cell metabolism: special Reference to Lactate Pathway
Cancer cell metabolism: special Reference to Lactate Pathway
 
Pests of Green Manures_Bionomics_IPM_Dr.UPR.pdf
Pests of Green Manures_Bionomics_IPM_Dr.UPR.pdfPests of Green Manures_Bionomics_IPM_Dr.UPR.pdf
Pests of Green Manures_Bionomics_IPM_Dr.UPR.pdf
 
FAIRSpectra - Towards a common data file format for SIMS images
FAIRSpectra - Towards a common data file format for SIMS imagesFAIRSpectra - Towards a common data file format for SIMS images
FAIRSpectra - Towards a common data file format for SIMS images
 
Mammalian Pineal Body Structure and Also Functions
Mammalian Pineal Body Structure and Also FunctionsMammalian Pineal Body Structure and Also Functions
Mammalian Pineal Body Structure and Also Functions
 
insect taxonomy importance systematics and classification
insect taxonomy importance systematics and classificationinsect taxonomy importance systematics and classification
insect taxonomy importance systematics and classification
 
Earliest Galaxies in the JADES Origins Field: Luminosity Function and Cosmic ...
Earliest Galaxies in the JADES Origins Field: Luminosity Function and Cosmic ...Earliest Galaxies in the JADES Origins Field: Luminosity Function and Cosmic ...
Earliest Galaxies in the JADES Origins Field: Luminosity Function and Cosmic ...
 
SCHIZOPHRENIA Disorder/ Brain Disorder.pdf
SCHIZOPHRENIA Disorder/ Brain Disorder.pdfSCHIZOPHRENIA Disorder/ Brain Disorder.pdf
SCHIZOPHRENIA Disorder/ Brain Disorder.pdf
 
THYROID-PARATHYROID medical surgical nursing
THYROID-PARATHYROID medical surgical nursingTHYROID-PARATHYROID medical surgical nursing
THYROID-PARATHYROID medical surgical nursing
 
Topography and sediments of the floor of the Bay of Bengal
Topography and sediments of the floor of the Bay of BengalTopography and sediments of the floor of the Bay of Bengal
Topography and sediments of the floor of the Bay of Bengal
 
Pests of sugarcane_Binomics_IPM_Dr.UPR.pdf
Pests of sugarcane_Binomics_IPM_Dr.UPR.pdfPests of sugarcane_Binomics_IPM_Dr.UPR.pdf
Pests of sugarcane_Binomics_IPM_Dr.UPR.pdf
 
Comparative structure of adrenal gland in vertebrates
Comparative structure of adrenal gland in vertebratesComparative structure of adrenal gland in vertebrates
Comparative structure of adrenal gland in vertebrates
 
PRESENTATION ABOUT PRINCIPLE OF COSMATIC EVALUATION
PRESENTATION ABOUT PRINCIPLE OF COSMATIC EVALUATIONPRESENTATION ABOUT PRINCIPLE OF COSMATIC EVALUATION
PRESENTATION ABOUT PRINCIPLE OF COSMATIC EVALUATION
 

2024.03.22 - Mike Heddes - Introduction to Hyperdimensional Computing.pdf

  • 1. An Introduction to Hyperdimensional Computing Mike Heddes University of California, Irvine Department of Computer Science March 2024
  • 2. Outline 1. Alphabet 2. Operations 3. Composition 4. Example: The dollar of the Netherlands 5. Application: Graph classification 6. Torchhd: PyTorch library for HD/VSA 1
  • 3. Origin Hyperdimensional computing (HD), or vector symbolic architectures (VSA) Observation The cerebellum cortex operates on high-dimensional representations [1]. • Forming compositional distributed representations [2, 3] • The intersection of symbolic and connectionist artificial intelligence • Symbols → vectors, creating a “concept space” • Distance metric: angular distance 2
  • 4. Flavors of HD/VSA Many variations of HD/VSA [4]: • Binary Spatter Codes [5] • Multiply-Add-Permute • Holographic Reduced Representations [6] • and more... 3
  • 5. Alphabet Within the Multiply-Add-Permute HD/VSA model [7]: • Symbols x, y ∼ U{−1, 1}d • Rademacher distribution 〈x, x〉 d = 1 E 〈x, y〉 d = 0 Var 〈x, y〉 d = 1 d In high-dimensional spaces: • Inner products are highly concentrated: quasi-orthogonal • Only d orthogonal symbols • O(exp(d)) quasi-orthogonal symbols [4] 4
  • 6. Operations: bundling • Bundling of symbols: component-wise addition • Useful for representing a set of symbols • Result can be normalized to {−1, 1}d E[〈x, x + y〉] = 〈x, x〉 + E[〈x, y〉] = d 5
  • 7. Operations: binding • Binding of symbols: component-wise multiplication (Hadamard product) • Distributes over bundling • Self-inverse property x ◦ y ∈ {−1, 1}d x ◦ x = 1 • Useful for the association of symbols E[〈x, x ◦ y〉] = E[〈1, x ◦ x ◦ y〉] = E[〈1, y〉] = 0 E[〈x, (x ◦ y) ◦ y〉] = E[〈x, x〉] = d 6
  • 8. Operations: permute • Permute of a symbol: permutation of its components • Distributes over binding and bundling • Useful for the assignment of an ordering 〈x, ρ(x)〉 ≈ 0 • We denote ρ(ρ(x)) as ρ2(x) • Also has an inverse permutation: ρ−1(x) 7
  • 9. Operations Operation Notation Domain Codomain Bundling x + y H × H H or Zd Binding x ◦ y H × H H Permute ρ(x) H H • Where H = {−1, 1}d • Theses operations enable composition 8
  • 10. Composition: multiset • Representation for a set of symbols {x1, x2, . . . , xn}: s = x1 + x2 + · · · + xn = n X i=1 xi • Query the frequency of symbol xj: E 〈xj, s〉 d = 1 d n X i=1 E[〈xj, xi〉] = n X i=1 1[xj = xi] • No free lunch: bundling capacity O(d) [8] 9
  • 11. Composition: sequence • Representation for a sequence of symbols (x1, x2, . . . , xn): s = x1 + ρ(x2) + ρ2 (x3) + · · · + ρn−1 (xn) = n X i=1 ρi−1 (xi) • Let X = [x1, x2, . . . , xn] be the codebook • Retrieve the (noisy) k-th symbol: x̃k = ρ1−k (s) = xk + noise with XT noise ≈ 0 = ρ1−k (x1 + ρ(x2) + ρ2 (x3) + · · · + ρn−1 (xn)) = ρ1−k (x1) + ρ2−k (x2) + ρ3−k (x3) + · · · + ρn−k (xn) 10
  • 12. Composition: key-value store • Representation for a set of mappings {(k1 ↔ v1), (k2 ↔ v2), . . . , (kn ↔ vn)}: s = k1 ◦ v1 + k2 ◦ v2 + · · · + kn ◦ vn = n X i=1 ki ◦ vi • Retrieve the (noisy) value of the j-th key: ṽj = s ◦ kj = vj + noise with XT noise ≈ 0 = (k1 ◦ v1 + k2 ◦ v2 + · · · + kn ◦ vn) ◦ kj = k1 ◦ kj ◦ v1 + k2 ◦ kj ◦ v2 + · · · + kn ◦ kj ◦ vn 11
  • 13. Associative memory • Clean up of a noisy symbol x̃: cleanup(x̃, X) = lookup(x̃, X, X) with lookup(q, K, V) = V σ ‚ KTq d Œ • σ is a nonlinear function, e.g., ReLU, Top-k, Softmax • Variations include: • Hopfield network [9] • sparse distributed memory [10] • attention mechanism [11] 12
  • 14. The dollar of the Netherlands • Classical example of reasoning by analogy [12] (originally the dollar of Mexico) • Key-value stores containing country (k1), capitol (k2), and currency (k3): u = k1 ◦ USA + k2 ◦ WDC + k3 ◦ USD n = k1 ◦ NLD + k2 ◦ AMS + k3 ◦ EUR What is the dollar of the Netherlands? ß EUR = u ◦ n ◦ USD = EUR + noise with XT noise ≈ 0 = (k1 ◦ k1 ◦ USA ◦ NLD + k2 ◦ k2 ◦ WDC ◦ AMS + k3 ◦ k3 ◦ USD ◦ EUR + k1 ◦ k2 ◦ USA ◦ AMS + k2 ◦ k3 ◦ WDC ◦ EUR + k3 ◦ k1 ◦ USD ◦ NLD + k1 ◦ k3 ◦ USA ◦ EUR + k2 ◦ k1 ◦ WDC ◦ NLD + k3 ◦ k2 ◦ USD ◦ AMS) ◦ USD = USA ◦ NLD ◦ USD + WDC ◦ AMS ◦ USD + EUR + . . . 13
  • 15. Graph classification Problem Given a set of labeled graphs, predict the label of a given unlabeled graph. Approach 1. Split the set of labeled graphs by their label. 2. Create a vector representation of each graph. 3. Create a vector representation of each set. 4. Find the set most similar to the unlabeled graph. 14
  • 16. Graph encoding As proposed in GraphHD [13], given an undirected graph G = (V, E). 1. Sample a (global) set of vertex-rank symbols 2. Rank vertices by their PageRank centrality 3. Assign vertex-rank symbols to vertices 4. Bind vertex symbols to represent an edge 5. Bundle all edge symbols g = X (i,j)∈E vi ◦ vj 15
  • 18. Comparison with Perceptron Model M1 M2 ... Mk Query vector Similarities HD/VSA classifier y = MTq Æ tr(MTM) Æ qTq Perceptron y = WT x + b 17
  • 19. Torchhd: PyTorch library for HD/VSA Enable quick development of performant HD/VSA experiments and applications [14]. • A simple programming experience • High-performance execution • Support for a variety of HD/VSA applications and research directions • Includes many reference implementations • Open source: https://github.com/hyperdimensional-computing/torchhd/ 18
  • 20. References I 1. Kanerva, P. Hyperdimensional computing: An introduction to computing in distributed representation with high-dimensional random vectors. Cognitive Computation 1, 139–159 (2009). 2. Smolensky, P. Tensor Product Variable Binding and the Representation of Symbolic Structures in Connectionist Systems. Artificial Intelligence 46, 159–216 (1990). 3. Plate, T. A. Holographic Reduced Representations: Convolution Algebra for Compositional Distributed Representations. in International Joint Conference on Artificial Intelligence (IJCAI) (1991), 30–35. 4. Kleyko, D., Rachkovskij, D. A., Osipov, E. Rahimi, A. A survey on hyperdimensional computing aka vector symbolic architectures, part i: Models and data transformations. ACM Computing Surveys 55, 1–40 (2022). 5. Kanerva, P. Fully distributed representation. in Proceedings of the Real World Computing Symposium (1997), 358–365. 6. Plate, T. A. Holographic reduced representations. IEEE Transactions on Neural networks 6, 623–641 (1995). 7. Gayler, R. Multiplicative binding, representation operators and analogy. Advances in analogy research: integration of theory and data from the cognitive, computational and neural sciences (1998). 8. Thomas, A., Dasgupta, S. Rosing, T. Theoretical Foundations of Hyperdimensional Computing. Journal of Artificial Intelligence Research (JAIR) 72, 215–249 (2021). 9. Hopfield, J. J. Neural networks and physical systems with emergent collective computational abilities. Proceedings of the national academy of sciences 79, 2554–2558 (1982). 19
  • 21. References II 10. Kanerva, P. Sparse distributed memory and related models. Tech. rep. (1992). 11. Vaswani, A. et al. Attention is all you need. Advances in neural information processing systems 30 (2017). 12. Kanerva, P. What we mean when we say” What’s the dollar of Mexico?”: Prototypes and mapping in concept space. in AAAI fall symposium series (2010). 13. Nunes, I., Heddes, M., Givargis, T., Nicolau, A. Veidenbaum, A. GraphHD: Efficient graph classification using hyperdimensional computing. in Design, Automation Test in Europe Conference Exhibition (DATE) (2022). 14. Heddes, M. et al. Torchhd: An Open Source Python Library to Support Research on Hyperdimensional Computing and Vector Symbolic Architectures. Journal of Machine Learning Research 24, 1–10. http://jmlr.org/papers/v24/23-0300.html (2023). 20