SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science...
Transcript of SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science...
![Page 1: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/1.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
SIGTACS Seminar Series
Metric Embeddings and Applications in Computer Science
Presented by : Purushottam Kar
January 10, 2009
1 / 23SIGTACS Seminar Series
![Page 2: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/2.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Outline
1 Introduction
2 Embeddings into Normed Spaces
3 Dimensionality Reduction
4 The JL Lemma
5 Discussion
2 / 23SIGTACS Seminar Series
![Page 3: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/3.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Basics
Definition (Metric)
A Metric is a structure (X , ρ) where ρ is a distance measureρ : X × X → R which is non-negative, symmetric and satisfies thetriangle inequality.
Definition (Embedding Distortion)
An embedding f : X → Y from a metric space (X , ρ) to anothermetric space (Y , σ) is said to have a distortion D if
D = supx ,y∈X
σ(f (x),f (y))ρ(x ,y) · sup
x ,y∈X
ρ(x ,y)σ(f (x),f (y)) .
Such embeddings are also called bi-Lipschitz embeddings.
3 / 23SIGTACS Seminar Series
![Page 4: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/4.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Embeddings
Various criterion used to evaluate embeddings
Distortion, Stress, Residual Variance ...
Definition (Embedding Stress)
The stress for an embedding f : X → Y from a metric space (X , ρ) to
another metric space (Y , σ) is defined to be
√√√√ ∑x,y∈X
(σ(f (x),f (y))−ρ(x,y))2∑x,y∈X
ρ(x,y)2 .
Lead to very interesting algorithmic questions
4 / 23SIGTACS Seminar Series
![Page 5: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/5.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Embeddings
Various criterion used to evaluate embeddings
Distortion, Stress, Residual Variance ...
Definition (Embedding Stress)
The stress for an embedding f : X → Y from a metric space (X , ρ) to
another metric space (Y , σ) is defined to be
√√√√ ∑x,y∈X
(σ(f (x),f (y))−ρ(x,y))2∑x,y∈X
ρ(x,y)2 .
Lead to very interesting algorithmic questions
4 / 23SIGTACS Seminar Series
![Page 6: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/6.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Embeddings
Various criterion used to evaluate embeddings
Distortion, Stress, Residual Variance ...
Definition (Embedding Stress)
The stress for an embedding f : X → Y from a metric space (X , ρ) to
another metric space (Y , σ) is defined to be
√√√√ ∑x,y∈X
(σ(f (x),f (y))−ρ(x,y))2∑x,y∈X
ρ(x,y)2 .
Lead to very interesting algorithmic questions
4 / 23SIGTACS Seminar Series
![Page 7: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/7.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Application in Computer Science
Started out as a branch of functional analysis
Algorithmic applications
Metric Embeddings for datasets operating with a non-metricDimensionality reduction to reduce storage space costs, processingtimeFacilitate pruning procedures in database searchesPreserve residual variance (PCA), inter-point similarity (RandomProjections), Stress (MDS)
Streaming Algorithms
5 / 23SIGTACS Seminar Series
![Page 8: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/8.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Application in Computer Science
Started out as a branch of functional analysis
Algorithmic applications
Metric Embeddings for datasets operating with a non-metricDimensionality reduction to reduce storage space costs, processingtimeFacilitate pruning procedures in database searchesPreserve residual variance (PCA), inter-point similarity (RandomProjections), Stress (MDS)
Streaming Algorithms
5 / 23SIGTACS Seminar Series
![Page 9: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/9.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Application in Computer Science
Started out as a branch of functional analysis
Algorithmic applications
Metric Embeddings for datasets operating with a non-metric
Dimensionality reduction to reduce storage space costs, processingtimeFacilitate pruning procedures in database searchesPreserve residual variance (PCA), inter-point similarity (RandomProjections), Stress (MDS)
Streaming Algorithms
5 / 23SIGTACS Seminar Series
![Page 10: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/10.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Application in Computer Science
Started out as a branch of functional analysis
Algorithmic applications
Metric Embeddings for datasets operating with a non-metricDimensionality reduction to reduce storage space costs, processingtime
Facilitate pruning procedures in database searchesPreserve residual variance (PCA), inter-point similarity (RandomProjections), Stress (MDS)
Streaming Algorithms
5 / 23SIGTACS Seminar Series
![Page 11: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/11.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Application in Computer Science
Started out as a branch of functional analysis
Algorithmic applications
Metric Embeddings for datasets operating with a non-metricDimensionality reduction to reduce storage space costs, processingtimeFacilitate pruning procedures in database searches
Preserve residual variance (PCA), inter-point similarity (RandomProjections), Stress (MDS)
Streaming Algorithms
5 / 23SIGTACS Seminar Series
![Page 12: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/12.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Application in Computer Science
Started out as a branch of functional analysis
Algorithmic applications
Metric Embeddings for datasets operating with a non-metricDimensionality reduction to reduce storage space costs, processingtimeFacilitate pruning procedures in database searchesPreserve residual variance (PCA), inter-point similarity (RandomProjections), Stress (MDS)
Streaming Algorithms
5 / 23SIGTACS Seminar Series
![Page 13: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/13.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Application in Computer Science
Started out as a branch of functional analysis
Algorithmic applications
Metric Embeddings for datasets operating with a non-metricDimensionality reduction to reduce storage space costs, processingtimeFacilitate pruning procedures in database searchesPreserve residual variance (PCA), inter-point similarity (RandomProjections), Stress (MDS)
Streaming Algorithms
5 / 23SIGTACS Seminar Series
![Page 14: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/14.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Embedding into l∞
Theorem (Fretchet’s Embedding)
Every n-point metric can be isometrically embedded into l∞
Frechet’s Embedding technique - non-expansive
Choose coordinates as projections onto some fixed sets
Triangle inequality ensures contractive embeddings
Choice of “landmark” sets gives other algorithms
Embedding dimension can be reduced to O(qn1q ln n) by tolerating a
distortion of 2q − 1.
6 / 23SIGTACS Seminar Series
![Page 15: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/15.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Embedding into l∞
Theorem (Fretchet’s Embedding)
Every n-point metric can be isometrically embedded into l∞
Frechet’s Embedding technique - non-expansive
Choose coordinates as projections onto some fixed sets
Triangle inequality ensures contractive embeddings
Choice of “landmark” sets gives other algorithms
Embedding dimension can be reduced to O(qn1q ln n) by tolerating a
distortion of 2q − 1.
6 / 23SIGTACS Seminar Series
![Page 16: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/16.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Embedding into l∞
Theorem (Fretchet’s Embedding)
Every n-point metric can be isometrically embedded into l∞
Frechet’s Embedding technique - non-expansive
Choose coordinates as projections onto some fixed sets
Triangle inequality ensures contractive embeddings
Choice of “landmark” sets gives other algorithms
Embedding dimension can be reduced to O(qn1q ln n) by tolerating a
distortion of 2q − 1.
6 / 23SIGTACS Seminar Series
![Page 17: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/17.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Embedding into l∞
Theorem (Fretchet’s Embedding)
Every n-point metric can be isometrically embedded into l∞
Frechet’s Embedding technique - non-expansive
Choose coordinates as projections onto some fixed sets
Triangle inequality ensures contractive embeddings
Choice of “landmark” sets gives other algorithms
Embedding dimension can be reduced to O(qn1q ln n) by tolerating a
distortion of 2q − 1.
6 / 23SIGTACS Seminar Series
![Page 18: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/18.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Embedding into l∞
Theorem (Fretchet’s Embedding)
Every n-point metric can be isometrically embedded into l∞
Frechet’s Embedding technique - non-expansive
Choose coordinates as projections onto some fixed sets
Triangle inequality ensures contractive embeddings
Choice of “landmark” sets gives other algorithms
Embedding dimension can be reduced to O(qn1q ln n) by tolerating a
distortion of 2q − 1.
6 / 23SIGTACS Seminar Series
![Page 19: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/19.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Embedding into l2
Theorem (Bourgain’s Embedding)
Every n-point metric can be O(log n)-embedded into l2
Uses a random selection of the landmark sets
Tight - The graph metric of a constant degree expander hasΩ(log n) distortion into any Euclidean space
Any embedding of the Hamming cube into l2 incurs Ω(√
log n)
distortion
7 / 23SIGTACS Seminar Series
![Page 20: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/20.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Embedding into l2
Theorem (Bourgain’s Embedding)
Every n-point metric can be O(log n)-embedded into l2
Uses a random selection of the landmark sets
Tight - The graph metric of a constant degree expander hasΩ(log n) distortion into any Euclidean space
Any embedding of the Hamming cube into l2 incurs Ω(√
log n)
distortion
7 / 23SIGTACS Seminar Series
![Page 21: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/21.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Embedding into l2
Theorem (Bourgain’s Embedding)
Every n-point metric can be O(log n)-embedded into l2
Uses a random selection of the landmark sets
Tight - The graph metric of a constant degree expander hasΩ(log n) distortion into any Euclidean space
Any embedding of the Hamming cube into l2 incurs Ω(√
log n)
distortion
7 / 23SIGTACS Seminar Series
![Page 22: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/22.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Dimensionality Reduction in l1
Impossible - A D-embedding of n points may require nΩ(1/D2)
dimensions
No “flattening” results known for other lp metrics either ...
Except for p = 2
8 / 23SIGTACS Seminar Series
![Page 23: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/23.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Dimensionality Reduction in l1
Impossible - A D-embedding of n points may require nΩ(1/D2)
dimensions
No “flattening” results known for other lp metrics either ...
Except for p = 2
8 / 23SIGTACS Seminar Series
![Page 24: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/24.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Dimensionality Reduction in l1
Impossible - A D-embedding of n points may require nΩ(1/D2)
dimensions
No “flattening” results known for other lp metrics either ...
Except for p = 2
8 / 23SIGTACS Seminar Series
![Page 25: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/25.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
The Johnson-Lindenstrauss Lemma
Theorem (The JL-Lemma)
Given ε > 0 and integer n, let k ≥ k0 = O(ε−2 log n). For every setP of n points in Rd there exists f : Rd −→ Rk such that for allu, v ∈ P
(1− ε)‖u − v‖2 ≤ ‖f (u)− f (v)‖2 ≤ (1 + ε)‖u − v‖2.
Implementation as a randomized algorithm
Equivalent interpretations - random projection vs. random rotation
Various Proofs known [IM98], [DG99], [AV99], [A01]
Common Technique
Point Drafting −→ Set DraftingUnion Bound−→ Set Embedding
9 / 23SIGTACS Seminar Series
![Page 26: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/26.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
The Johnson-Lindenstrauss Lemma
Theorem (The JL-Lemma)
Given ε > 0 and integer n, let k ≥ k0 = O(ε−2 log n). For every setP of n points in Rd there exists f : Rd −→ Rk such that for allu, v ∈ P
(1− ε)‖u − v‖2 ≤ ‖f (u)− f (v)‖2 ≤ (1 + ε)‖u − v‖2.
Implementation as a randomized algorithm
Equivalent interpretations - random projection vs. random rotation
Various Proofs known [IM98], [DG99], [AV99], [A01]
Common Technique
Point Drafting −→ Set DraftingUnion Bound−→ Set Embedding
9 / 23SIGTACS Seminar Series
![Page 27: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/27.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
The Johnson-Lindenstrauss Lemma
Theorem (The JL-Lemma)
Given ε > 0 and integer n, let k ≥ k0 = O(ε−2 log n). For every setP of n points in Rd there exists f : Rd −→ Rk such that for allu, v ∈ P
(1− ε)‖u − v‖2 ≤ ‖f (u)− f (v)‖2 ≤ (1 + ε)‖u − v‖2.
Implementation as a randomized algorithm
Equivalent interpretations - random projection vs. random rotation
Various Proofs known [IM98], [DG99], [AV99], [A01]
Common Technique
Point Drafting −→ Set DraftingUnion Bound−→ Set Embedding
9 / 23SIGTACS Seminar Series
![Page 28: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/28.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
The Johnson-Lindenstrauss Lemma
Theorem (The JL-Lemma)
Given ε > 0 and integer n, let k ≥ k0 = O(ε−2 log n). For every setP of n points in Rd there exists f : Rd −→ Rk such that for allu, v ∈ P
(1− ε)‖u − v‖2 ≤ ‖f (u)− f (v)‖2 ≤ (1 + ε)‖u − v‖2.
Implementation as a randomized algorithm
Equivalent interpretations - random projection vs. random rotation
Various Proofs known [IM98], [DG99], [AV99], [A01]
Common Technique
Point Drafting −→ Set DraftingUnion Bound−→ Set Embedding
9 / 23SIGTACS Seminar Series
![Page 29: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/29.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Enter Achlioptas
Instead of choosing from an uncountably infinite domain, can wechoose vectors from a finite set of vectors ?
Achlioptas: In fact ‘choosing’ from the d-dimensional HammingCube 1,−1d works.
Consider a random vector R = (X1,X2, . . . ,Xd), where each Xi ischosen from one of the two distributions:
D1 =1√d
−1 with probability 1/21 with probability 1/2
D2 =1√d
−√
3 with probability 1/60 with probability 2/3√
3 with probability 1/6
10 / 23SIGTACS Seminar Series
![Page 30: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/30.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Enter Achlioptas
Instead of choosing from an uncountably infinite domain, can wechoose vectors from a finite set of vectors ?
Achlioptas: In fact ‘choosing’ from the d-dimensional HammingCube 1,−1d works.
Consider a random vector R = (X1,X2, . . . ,Xd), where each Xi ischosen from one of the two distributions:
D1 =1√d
−1 with probability 1/21 with probability 1/2
D2 =1√d
−√
3 with probability 1/60 with probability 2/3√
3 with probability 1/6
10 / 23SIGTACS Seminar Series
![Page 31: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/31.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Enter Achlioptas
Instead of choosing from an uncountably infinite domain, can wechoose vectors from a finite set of vectors ?
Achlioptas: In fact ‘choosing’ from the d-dimensional HammingCube 1,−1d works.
Consider a random vector R = (X1,X2, . . . ,Xd), where each Xi ischosen from one of the two distributions:
D1 =1√d
−1 with probability 1/21 with probability 1/2
D2 =1√d
−√
3 with probability 1/60 with probability 2/3√
3 with probability 1/6
10 / 23SIGTACS Seminar Series
![Page 32: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/32.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Enter Achlioptas
Pick k such random vectors R1,R2, . . .Rk .
For a given unit vector α = (α1, α2, . . . , αd), the low(k-)dimensional vector corresponding to α is
f (α) =√
dk (〈α,R1〉 , 〈α,R2〉 , . . . , 〈α,Rk〉)
Advantage: Simple and can be implemented as SQL queries.
11 / 23SIGTACS Seminar Series
![Page 33: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/33.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Enter Achlioptas
Pick k such random vectors R1,R2, . . .Rk .
For a given unit vector α = (α1, α2, . . . , αd), the low(k-)dimensional vector corresponding to α is
f (α) =√
dk (〈α,R1〉 , 〈α,R2〉 , . . . , 〈α,Rk〉)
Advantage: Simple and can be implemented as SQL queries.
11 / 23SIGTACS Seminar Series
![Page 34: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/34.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Enter Achlioptas
Pick k such random vectors R1,R2, . . .Rk .
For a given unit vector α = (α1, α2, . . . , αd), the low(k-)dimensional vector corresponding to α is
f (α) =√
dk (〈α,R1〉 , 〈α,R2〉 , . . . , 〈α,Rk〉)
Advantage: Simple and can be implemented as SQL queries.
11 / 23SIGTACS Seminar Series
![Page 35: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/35.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Main Theorem
Let S = 〈α,R1〉2 + 〈α,R2〉2 + · · · 〈α,Rk〉2
Theorem (Main Theorem)
For every d-dimensional unit vector α, integer k ≥ 1 and ε > 0
Pr[S ≥ (1± ε) k
d · 1]≤ e
−k2 ( ε
2
2 −ε3
3 )
Hence, if k ≥ 4+2βε2/2−ε3/3 log n, this probability becomes smaller than
2n2+β which is inverse polynomial w.r.t n.
12 / 23SIGTACS Seminar Series
![Page 36: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/36.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Main Theorem
Let S = 〈α,R1〉2 + 〈α,R2〉2 + · · · 〈α,Rk〉2
Theorem (Main Theorem)
For every d-dimensional unit vector α, integer k ≥ 1 and ε > 0
Pr[S ≥ (1± ε) k
d · 1]≤ e
−k2 ( ε
2
2 −ε3
3 )
Hence, if k ≥ 4+2βε2/2−ε3/3 log n, this probability becomes smaller than
2n2+β which is inverse polynomial w.r.t n.
12 / 23SIGTACS Seminar Series
![Page 37: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/37.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Expected Value of ‖f (α)‖2
On expectation the length of a unit vector α is preserved.
E[‖f (α)‖2
]= E
k∑i=1
d
k
d∑j=1
Xjαj
2
=d
k
k∑i=1
d∑j=1
E [X 2j ]α2
j +d∑
j<l
E [XjXl ]αjαl
=
d
k
k∑i=1
1
d= 1 = ‖α‖2
13 / 23SIGTACS Seminar Series
![Page 38: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/38.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Deviation from Expectation: Proof of MainTheorem
By Markov inequality,
Pr
[S > (1 + ε)
k
d
]< E
[ehS]e−(1+ε) hk
d
Pr
[S < (1− ε)k
d
]< E
[e−hS
]e(1−ε) hk
d
Since the vectors R ′i s are all chosen independently we can rewritethe above as
Pr
[S > (1 + ε)
k
d
]<
(E[ehQ2
1
])k
e−(1+ε) hkd
Pr
[S < (1− ε)k
d
]<
(E[e−hQ2
1
])k
e(1−ε) hkd
where Q1 = 〈α,R1〉
14 / 23SIGTACS Seminar Series
![Page 39: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/39.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Deviation from Expectation: Proof of MainTheorem
By Markov inequality,
Pr
[S > (1 + ε)
k
d
]< E
[ehS]e−(1+ε) hk
d
Pr
[S < (1− ε)k
d
]< E
[e−hS
]e(1−ε) hk
d
Since the vectors R ′i s are all chosen independently we can rewritethe above as
Pr
[S > (1 + ε)
k
d
]<
(E[ehQ2
1
])k
e−(1+ε) hkd
Pr
[S < (1− ε)k
d
]<
(E[e−hQ2
1
])k
e(1−ε) hkd
where Q1 = 〈α,R1〉
14 / 23SIGTACS Seminar Series
![Page 40: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/40.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Proof of Main Theorem
By Taylor’s Expansion,
Pr
[S < (1− ε)k
d
]<
(E
[1− hQ2
1 +hQ4
1
2
])k
e−(1+ε) hkd
=
(1− h
d+
h2E [Q41 ]
2
)k
e(1−ε) hkd
Lemma
For h ∈ [0, d/2) and all d ≥ 1,
E[ehQ2
1
]≤ 1√
1− 2h/d(1)
E[Q4
1
]≤ 3
d2(2)
15 / 23SIGTACS Seminar Series
![Page 41: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/41.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Proof of Main Theorem using Inequalities (1)and (2)
If we take h = dε2(1+ε) , for the upper bound we have the following:
Pr
[S > (1 + ε)
k
d
]<
(1√
1− 2h/d
)k
e−(1+ε) hkd
= ((1 + ε)e−ε)k/2 < e−k
2 ( ε2
2 −ε3
3 ).
For the same value of h, for the lower bound we get:
Pr
[S < (1− ε)k
d
]<
(1− h/d +
3h2
2d2
)k
e(1−ε) hkd
< e−k
2 ( ε2
2 −ε3
3 ).
16 / 23SIGTACS Seminar Series
![Page 42: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/42.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Proof of Main Theorem using Inequalities (1)and (2)
If we take h = dε2(1+ε) , for the upper bound we have the following:
Pr
[S > (1 + ε)
k
d
]<
(1√
1− 2h/d
)k
e−(1+ε) hkd
= ((1 + ε)e−ε)k/2 < e−k
2 ( ε2
2 −ε3
3 ).
For the same value of h, for the lower bound we get:
Pr
[S < (1− ε)k
d
]<
(1− h/d +
3h2
2d2
)k
e(1−ε) hkd
< e−k
2 ( ε2
2 −ε3
3 ).
16 / 23SIGTACS Seminar Series
![Page 43: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/43.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Proof of Inequality (2)
For inequality (2)
E [Q41 ] = (
∑di=1 Xiαi )
4 =∑
i
E [X 4i ]α4
i +(4
1, 3
)∑i<j
E [X 3i ]E [Xj ]α
3i αi +
(4
2, 2
)∑i<j
E [X 2i ]E [X 2
j ]α2i α
2j +(
4
2, 1, 1
) ∑i<j<k
E [X 2i ]E [Xj ]E [Xk ]α2
i αjαk +(4
1, 1, 1, 1
) ∑i<j<k<l
E [Xi ]E [Xj ]E [Xk ]E [Xl ]αiαjαkαl
=1
d2(α4 + 6
∑i<j
α2i α
2j ) ≤ 3
d2.
17 / 23SIGTACS Seminar Series
![Page 44: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/44.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Proof of Inequality (1)
The idea is to first make the random variable Q1 independent of αand then compare the even moments of Q1 with a properly scalednormal distribution.
Lemma (Worst Vector Lemma)
For all unit vectors α, E [Q2k1 (α)] ≤ E [Q2k
1 (w)], wherew = 1√
d(1, 1, . . . , 1) for k = 1, 2, . . . .
Lemma (Normal Bound Lemma)
If T ∼ N(0, 1/d), then E [Q2k1 (w)] ≤ E [T 2k ], where w = 1√
d(1, 1, . . . , 1)
for k = 1, 2, . . . .
18 / 23SIGTACS Seminar Series
![Page 45: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/45.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Proof of Inequality (1)
E[ehT 2
]=
∫ ∞−∞
1√2π
eλ2/2ehλ2/ddλ
=1√
1− 2h/d
= E
[ ∞∑k=0
hkT 2k
k!
](using MCT)
=∞∑
k=0
hkE[T 2k
]k!
≥∞∑
k=0
hkE[Q2k
1 (w)]
k!= E
[ehQ1(w)2
]≥ E
[ehQ1(α)2
]
19 / 23SIGTACS Seminar Series
![Page 46: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/46.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Proving the Worst Vector Lemma
Let r1 and r2 be i.i.d. r.v. distributed as −1,+1 with equalprobability. Furthermore let a, b,T be any reals andc =
√(a2 + b2)/2 and k > 0 be any integer, then
E[(T + ar1 + br2)2k
]≤ E
[(T + cr1 + cr2)2k
]
Let R1 = 1√d
(r1, r2, . . . , rd). Thus we have
E[Q1(α)2k
]=
1
dk
∑R
E[(R + α1r1 + α2r2)2k
]Pr
[d∑
i=3
αi ri =R√d
]
≤ 1
dk
∑R
E[(R + cr1 + cr2)2k
]Pr
[d∑
i=3
αi ri =R√d
]= E
[Q1(θ)2k
]where c =
√(α2
1 + α22)/2
θ is a more “uniform” unit vector than α.
20 / 23SIGTACS Seminar Series
![Page 47: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/47.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Proving the Worst Vector Lemma
Let r1 and r2 be i.i.d. r.v. distributed as −1,+1 with equalprobability. Furthermore let a, b,T be any reals andc =
√(a2 + b2)/2 and k > 0 be any integer, then
E[(T + ar1 + br2)2k
]≤ E
[(T + cr1 + cr2)2k
]Let R1 = 1√
d(r1, r2, . . . , rd). Thus we have
E[Q1(α)2k
]=
1
dk
∑R
E[(R + α1r1 + α2r2)2k
]Pr
[d∑
i=3
αi ri =R√d
]
≤ 1
dk
∑R
E[(R + cr1 + cr2)2k
]Pr
[d∑
i=3
αi ri =R√d
]= E
[Q1(θ)2k
]where c =
√(α2
1 + α22)/2
θ is a more “uniform” unit vector than α.
20 / 23SIGTACS Seminar Series
![Page 48: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/48.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Proving the Worst Vector Lemma
Let r1 and r2 be i.i.d. r.v. distributed as −1,+1 with equalprobability. Furthermore let a, b,T be any reals andc =
√(a2 + b2)/2 and k > 0 be any integer, then
E[(T + ar1 + br2)2k
]≤ E
[(T + cr1 + cr2)2k
]Let R1 = 1√
d(r1, r2, . . . , rd). Thus we have
E[Q1(α)2k
]=
1
dk
∑R
E[(R + α1r1 + α2r2)2k
]Pr
[d∑
i=3
αi ri =R√d
]
≤ 1
dk
∑R
E[(R + cr1 + cr2)2k
]Pr
[d∑
i=3
αi ri =R√d
]= E
[Q1(θ)2k
]where c =
√(α2
1 + α22)/2
θ is a more “uniform” unit vector than α.
20 / 23SIGTACS Seminar Series
![Page 49: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/49.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Proving the Normal Bound Lemma
Let Tidi=1 be i.i.d. normal r.v.. By stability of normal distribution
T = 1d
d∑i=1
Ti ∼ N(0, 1/d)
We also have Q1(w) = 1d
d∑i=1
r1
E [Q2k1 (w)] =
1
d2k
d∑i1=1
. . .
d∑i2k =1
E [ri1 . . . ri2k]
E [T 2k ] =1
d2k
d∑i1=1
. . .
d∑i2k =1
E [Ti1 . . .Ti2k]
For each index assignment we have
E [ri1 . . . ri2k] ≤ E [Ti1 . . .Ti2k
]
21 / 23SIGTACS Seminar Series
![Page 50: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/50.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Proving the Normal Bound Lemma
Let Tidi=1 be i.i.d. normal r.v.. By stability of normal distribution
T = 1d
d∑i=1
Ti ∼ N(0, 1/d)
We also have Q1(w) = 1d
d∑i=1
r1
E [Q2k1 (w)] =
1
d2k
d∑i1=1
. . .
d∑i2k =1
E [ri1 . . . ri2k]
E [T 2k ] =1
d2k
d∑i1=1
. . .
d∑i2k =1
E [Ti1 . . .Ti2k]
For each index assignment we have
E [ri1 . . . ri2k] ≤ E [Ti1 . . .Ti2k
]
21 / 23SIGTACS Seminar Series
![Page 51: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/51.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Proving the Normal Bound Lemma
Let Tidi=1 be i.i.d. normal r.v.. By stability of normal distribution
T = 1d
d∑i=1
Ti ∼ N(0, 1/d)
We also have Q1(w) = 1d
d∑i=1
r1
E [Q2k1 (w)] =
1
d2k
d∑i1=1
. . .
d∑i2k =1
E [ri1 . . . ri2k]
E [T 2k ] =1
d2k
d∑i1=1
. . .
d∑i2k =1
E [Ti1 . . .Ti2k]
For each index assignment we have
E [ri1 . . . ri2k] ≤ E [Ti1 . . .Ti2k
]
21 / 23SIGTACS Seminar Series
![Page 52: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/52.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Open questions
Plenty !
No-flattening results for other lp metrics, non metrics
Embeddability of non-metrics into metric spaces - useful indatabases, learning
Information Theoretic Metrics - KL, Bhattacharyya, Mahalanobis -widely used
22 / 23SIGTACS Seminar Series
![Page 53: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/53.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Open questions
Plenty !
No-flattening results for other lp metrics, non metrics
Embeddability of non-metrics into metric spaces - useful indatabases, learning
Information Theoretic Metrics - KL, Bhattacharyya, Mahalanobis -widely used
22 / 23SIGTACS Seminar Series
![Page 54: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/54.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Open questions
Plenty !
No-flattening results for other lp metrics, non metrics
Embeddability of non-metrics into metric spaces - useful indatabases, learning
Information Theoretic Metrics - KL, Bhattacharyya, Mahalanobis -widely used
22 / 23SIGTACS Seminar Series
![Page 55: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/55.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
Open questions
Plenty !
No-flattening results for other lp metrics, non metrics
Embeddability of non-metrics into metric spaces - useful indatabases, learning
Information Theoretic Metrics - KL, Bhattacharyya, Mahalanobis -widely used
22 / 23SIGTACS Seminar Series
![Page 56: SIGTACS Seminar SeriesSIGTACS Seminar Series Metric Embeddings and Applications in Computer Science Presented by : Purushottam Kar January 10, 2009 1 / 23 ... (1 + )ku vk2. Implementation](https://reader033.fdocuments.us/reader033/viewer/2022042319/5f08192a7e708231d42053ca/html5/thumbnails/56.jpg)
Introduction Embeddings into Normed Spaces Dimensionality Reduction The JL Lemma Discussion
THANK YOU
23 / 23SIGTACS Seminar Series