Fast Unbalanced Optimal Transport on a Tree

J
1 KYOTO UNIVERSITY
KYOTO UNIVERSITY
Fast Unbalanced Optimal Transport on a Tree
Ryoma Sato
Kyoto University / RIKEN AIP
2 / 13 KYOTO UNIVERSITY
Self Introduction

I am a second year master's student at
Kyoto University

I’m interested in algorithmic aspects of machine
learning and data mining for structured data, including
Graph neural networks:

Ryoma Sato, Makoto Yamada, Hisashi Kashima. Approximation Ratios of
Graph Neural Networks for Combinatorial Problems. NeurIPS 2019.

Ryoma Sato, Makoto Yamada, Hisashi Kashima. Random Features
Strengthen Graph Neural Networks. SDM 2021
Optimal transport:

Ryoma Sato, Makoto Yamada, Hisashi Kashima. Fast Unbalanced Optimal
Transport on a Tree. NeurIPS 2020.
 Today’s topic
3 / 13 KYOTO UNIVERSITY
Background: optimal transport is useful

The optimal transport (OT) distance measures
the discrepancy of two distributions.
We consider discrete distributions in this presentation.

The OT distance is the minimum cumulative
distance that all masses need to travel from
one distribution to another distribution

In generative modeling, a mass is a sample.
discrepancy of model  sample distribution

In text classification, a mass is a word.
 OT does not require the same support  KL divergence
 OT exploits the underground geometry
From Word Embeddings To
Document Distances, ICML 2015
4 / 13 KYOTO UNIVERSITY
Background: sliced OT is computationally cheap

OT is formulated as a linear program  cubic cost

Sliced OT projects distributions to random
1D spaces and computes OT there

Greedy matching solves 1D OT exactly  linear cost
: distance matrix (input), : matching matrix (variable)
: 1st mass vector (input), : 2nd mass vector (input)
The leftmost mass should be matched to
the leftmost mass
The second leftmost mass should be matched
to the second leftmost mass ...
https://www.programmersought.com/article/67174999352/
https://analyticsindiamag.com/how-to-establish-domain-transferability-in-neural-models/
5 / 13 KYOTO UNIVERSITY
Background: unbalanced OT is robust

OT is sensitive to outliers because transporting outliers
becomes the dominating term

Unbalanced OT (UOT) allows to discard and create
masses by paying some penalties
 We can discard outliers  robust to outliers

UOT is also formulated by a linear program
 cubic cost
6 / 13 KYOTO UNIVERSITY
Background: UOT is difficult even in 1D spaces

We want to make a cheap alternative of UOT as 1D OT

But the greedy matching fails to solve 1D UOT

Let’s consider the following instance with discard cost λ

The following plan costs 3λ.

The following plan costs 2λ + 2ε. Thus this is better.
λ λ λ
7 / 13 KYOTO UNIVERSITY
Background: UOT is difficult even in 1D spaces

Let’s consider the following instance with discard cost λ

The following plan costs λ + 2ε.

The following plan costs 2λ + 2ε. Thus this is worse.
λ ε
8 / 13 KYOTO UNIVERSITY
Background: UOT is difficult even in 1D spaces

Although these two instances share the leftmost part,
the leftmost mass in the first instance should be
discarded while that in the second instance should not

The optimal UOT plan cannot be determined locally
 The optimal OT plan is determined locally

Thus the greedy algorithm fails to solve 1D UOT

We proposed how to solve 1D UOT efficiently
λ λ λ
λ ε
9 / 13 KYOTO UNIVERSITY
Algorithm: prune redundant plans

Our proposed method determines assignments from
left to right (as the greedy algorithm)

Although there are exponentially many plans, most of
them are redundant.
We proved that only O(n) plans are non-redundant
 Only one plan is non-redundant (thus greedy is valid) in the standard OT
not yet
not yet
 non redundant
 redundant
10 / 13 KYOTO UNIVERSITY
Algorithm: we solve 1D UOT in O(n log2
n) time

A naive algorithm requires cubic time even with this
(non redundant plan) observation

More algorithmic techniques are required for further
speedup (skipped in this presentation)

Dynamic programming

Fast convex min-sum convolution

Efficient data structure (BBST)

Weighted union heuristics

Finally, we derived a quasi-linear time algorithm
which runs in O(n log2
n) time in the worst case
11 / 13 KYOTO UNIVERSITY
Algorithm: tree UOT generalizes 1D UOT

Our method can be extended to tree spaces
A 1D space (path) is a special case of tree spaces

In text classification, the word
space can be represented by a
word tree. Each mass (word)
travels on the word tree to a
nearby (semantically similar) word.

We can “tree-slice” high dimensional
spaces instead of 1D-slicing,
which captures richer structures
http://www.sthda.com/english/articles/31-principal-component-methods-in-r-practical-guide/117-hcpc-hierarchical-clustering-on-principal-components-essentials/
a → b
12 / 13 KYOTO UNIVERSITY
Experiments: our algorithm is empirically fast

We confirmed that our algorithm could compute tree
UOT with one million masses within one second

We also confirmed that tree-slicing high dimensional
spaces could approximate the original UOT problem
13 / 13 KYOTO UNIVERSITY
Conclusion: fast computation of tree UOT

Sliced OT is a fast alternative of OT

UOT is a robust variant of OT

1D UOT is more difficult than 1D OT

We proposed an efficient algorithm for 1D UOT for the
first time

Our method can be extended to tree spaces

Our method is empirically fast (1M masses in 1 sec)
1 of 13

Recommended

Direct feedback alignment provides learning in Deep Neural Networks by
Direct feedback alignment provides learning in Deep Neural NetworksDirect feedback alignment provides learning in Deep Neural Networks
Direct feedback alignment provides learning in Deep Neural NetworksDeep Learning JP
2.4K views21 slides
テンソル多重線形ランクの推定法について(Estimation of Multi-linear Tensor Rank) by
テンソル多重線形ランクの推定法について(Estimation of Multi-linear Tensor Rank)テンソル多重線形ランクの推定法について(Estimation of Multi-linear Tensor Rank)
テンソル多重線形ランクの推定法について(Estimation of Multi-linear Tensor Rank)Tatsuya Yokota
5K views49 slides
行列およびテンソルデータに対する機械学習(数理助教の会 2011/11/28) by
行列およびテンソルデータに対する機械学習(数理助教の会 2011/11/28)行列およびテンソルデータに対する機械学習(数理助教の会 2011/11/28)
行列およびテンソルデータに対する機械学習(数理助教の会 2011/11/28)ryotat
18.9K views51 slides
Deep forest by
Deep forestDeep forest
Deep forestnaoto moriyama
13.1K views18 slides
数学で解き明かす深層学習の原理 by
数学で解き明かす深層学習の原理数学で解き明かす深層学習の原理
数学で解き明かす深層学習の原理Taiji Suzuki
3.8K views42 slides
PRML Chapter 14 by
PRML Chapter 14PRML Chapter 14
PRML Chapter 14Masahito Ohue
7.4K views65 slides

More Related Content

What's hot

MIRU2016 チュートリアル by
MIRU2016 チュートリアルMIRU2016 チュートリアル
MIRU2016 チュートリアルShunsuke Ono
25.5K views46 slides
サポートベクターマシン(SVM)の数学をみんなに説明したいだけの会 by
サポートベクターマシン(SVM)の数学をみんなに説明したいだけの会サポートベクターマシン(SVM)の数学をみんなに説明したいだけの会
サポートベクターマシン(SVM)の数学をみんなに説明したいだけの会Kenyu Uehara
78.9K views37 slides
ようやく分かった!最尤推定とベイズ推定 by
ようやく分かった!最尤推定とベイズ推定ようやく分かった!最尤推定とベイズ推定
ようやく分かった!最尤推定とベイズ推定Akira Masuda
96.8K views31 slides
因果推論の基礎 by
因果推論の基礎因果推論の基礎
因果推論の基礎Hatsuru Morita
13.9K views28 slides
DNNの曖昧性に関する研究動向 by
DNNの曖昧性に関する研究動向DNNの曖昧性に関する研究動向
DNNの曖昧性に関する研究動向Naoki Matsunaga
385 views60 slides
機会学習ハッカソン:ランダムフォレスト by
機会学習ハッカソン:ランダムフォレスト機会学習ハッカソン:ランダムフォレスト
機会学習ハッカソン:ランダムフォレストTeppei Baba
156.9K views46 slides

What's hot(20)

MIRU2016 チュートリアル by Shunsuke Ono
MIRU2016 チュートリアルMIRU2016 チュートリアル
MIRU2016 チュートリアル
Shunsuke Ono25.5K views
サポートベクターマシン(SVM)の数学をみんなに説明したいだけの会 by Kenyu Uehara
サポートベクターマシン(SVM)の数学をみんなに説明したいだけの会サポートベクターマシン(SVM)の数学をみんなに説明したいだけの会
サポートベクターマシン(SVM)の数学をみんなに説明したいだけの会
Kenyu Uehara78.9K views
ようやく分かった!最尤推定とベイズ推定 by Akira Masuda
ようやく分かった!最尤推定とベイズ推定ようやく分かった!最尤推定とベイズ推定
ようやく分かった!最尤推定とベイズ推定
Akira Masuda96.8K views
DNNの曖昧性に関する研究動向 by Naoki Matsunaga
DNNの曖昧性に関する研究動向DNNの曖昧性に関する研究動向
DNNの曖昧性に関する研究動向
Naoki Matsunaga385 views
機会学習ハッカソン:ランダムフォレスト by Teppei Baba
機会学習ハッカソン:ランダムフォレスト機会学習ハッカソン:ランダムフォレスト
機会学習ハッカソン:ランダムフォレスト
Teppei Baba156.9K views
PCAの最終形態GPLVMの解説 by 弘毅 露崎
PCAの最終形態GPLVMの解説PCAの最終形態GPLVMの解説
PCAの最終形態GPLVMの解説
弘毅 露崎23.9K views
FDRの使い方 (Kashiwa.R #3) by Haruka Ozaki
FDRの使い方 (Kashiwa.R #3)FDRの使い方 (Kashiwa.R #3)
FDRの使い方 (Kashiwa.R #3)
Haruka Ozaki40.6K views
Semi supervised, weakly-supervised, unsupervised, and active learning by Yusuke Uchida
Semi supervised, weakly-supervised, unsupervised, and active learningSemi supervised, weakly-supervised, unsupervised, and active learning
Semi supervised, weakly-supervised, unsupervised, and active learning
Yusuke Uchida6.3K views
Noisy Labels と戦う深層学習 by Plot Hong
Noisy Labels と戦う深層学習Noisy Labels と戦う深層学習
Noisy Labels と戦う深層学習
Plot Hong616 views
信号処理・画像処理における凸最適化 by Shunsuke Ono
信号処理・画像処理における凸最適化信号処理・画像処理における凸最適化
信号処理・画像処理における凸最適化
Shunsuke Ono11.3K views
猫でも分かりたい線形回帰の自由度 by YukinoriKambe
猫でも分かりたい線形回帰の自由度猫でも分かりたい線形回帰の自由度
猫でも分かりたい線形回帰の自由度
YukinoriKambe3.7K views
強化学習その2 by nishio
強化学習その2強化学習その2
強化学習その2
nishio22.7K views
PRML上巻勉強会 at 東京大学 資料 第1章後半 by Ohsawa Goodfellow
PRML上巻勉強会 at 東京大学 資料 第1章後半PRML上巻勉強会 at 東京大学 資料 第1章後半
PRML上巻勉強会 at 東京大学 資料 第1章後半
Ohsawa Goodfellow6.6K views
PRML学習者から入る深層生成モデル入門 by tmtm otm
PRML学習者から入る深層生成モデル入門PRML学習者から入る深層生成モデル入門
PRML学習者から入る深層生成モデル入門
tmtm otm5.7K views
グラフィカルモデル入門 by Kawamoto_Kazuhiko
グラフィカルモデル入門グラフィカルモデル入門
グラフィカルモデル入門
Kawamoto_Kazuhiko117.8K views
モンテカルロサンプリング by Kosei ABE
モンテカルロサンプリングモンテカルロサンプリング
モンテカルロサンプリング
Kosei ABE12.8K views
劣モジュラ最適化と機械学習1章 by Hakky St
劣モジュラ最適化と機械学習1章劣モジュラ最適化と機械学習1章
劣モジュラ最適化と機械学習1章
Hakky St14.9K views

Similar to Fast Unbalanced Optimal Transport on a Tree

IRJET- A Particle Swarm Optimization Algorithm for Total Cost Minimization in... by
IRJET- A Particle Swarm Optimization Algorithm for Total Cost Minimization in...IRJET- A Particle Swarm Optimization Algorithm for Total Cost Minimization in...
IRJET- A Particle Swarm Optimization Algorithm for Total Cost Minimization in...IRJET Journal
20 views4 slides
The Swarm Based Routing Algorithms by
The Swarm Based Routing AlgorithmsThe Swarm Based Routing Algorithms
The Swarm Based Routing AlgorithmsLakeisha Jones
2 views78 slides
Huge-Scale Molecular Dynamics Simulation of Multi-bubble Nuclei by
Huge-Scale Molecular Dynamics Simulation of Multi-bubble NucleiHuge-Scale Molecular Dynamics Simulation of Multi-bubble Nuclei
Huge-Scale Molecular Dynamics Simulation of Multi-bubble NucleiHiroshi Watanabe
1.4K views20 slides
NON - EUCLIDEAN METRIC AND PATH PLANNING by
NON - EUCLIDEAN METRIC AND PATH PLANNINGNON - EUCLIDEAN METRIC AND PATH PLANNING
NON - EUCLIDEAN METRIC AND PATH PLANNINGIJCSES Journal
132 views11 slides
Nano mos25 by
Nano mos25Nano mos25
Nano mos25Mamoon Ismail Khalid
91 views13 slides
Citython presentation by
Citython presentationCitython presentation
Citython presentationAnkit Tewari
160 views30 slides

Similar to Fast Unbalanced Optimal Transport on a Tree(20)

IRJET- A Particle Swarm Optimization Algorithm for Total Cost Minimization in... by IRJET Journal
IRJET- A Particle Swarm Optimization Algorithm for Total Cost Minimization in...IRJET- A Particle Swarm Optimization Algorithm for Total Cost Minimization in...
IRJET- A Particle Swarm Optimization Algorithm for Total Cost Minimization in...
IRJET Journal20 views
The Swarm Based Routing Algorithms by Lakeisha Jones
The Swarm Based Routing AlgorithmsThe Swarm Based Routing Algorithms
The Swarm Based Routing Algorithms
Lakeisha Jones2 views
Huge-Scale Molecular Dynamics Simulation of Multi-bubble Nuclei by Hiroshi Watanabe
Huge-Scale Molecular Dynamics Simulation of Multi-bubble NucleiHuge-Scale Molecular Dynamics Simulation of Multi-bubble Nuclei
Huge-Scale Molecular Dynamics Simulation of Multi-bubble Nuclei
Hiroshi Watanabe1.4K views
NON - EUCLIDEAN METRIC AND PATH PLANNING by IJCSES Journal
NON - EUCLIDEAN METRIC AND PATH PLANNINGNON - EUCLIDEAN METRIC AND PATH PLANNING
NON - EUCLIDEAN METRIC AND PATH PLANNING
IJCSES Journal132 views
Citython presentation by Ankit Tewari
Citython presentationCitython presentation
Citython presentation
Ankit Tewari160 views
Unsupervised Transcription Of Piano Music by Ashley Jones
Unsupervised Transcription Of Piano MusicUnsupervised Transcription Of Piano Music
Unsupervised Transcription Of Piano Music
Ashley Jones2 views
Impact Of Technology On New Business Models by Victoria Soto
Impact Of Technology On New Business ModelsImpact Of Technology On New Business Models
Impact Of Technology On New Business Models
Victoria Soto3 views
Aggregation computation over distributed data streams(the final version) by Yueshen Xu
Aggregation computation over distributed data streams(the final version)Aggregation computation over distributed data streams(the final version)
Aggregation computation over distributed data streams(the final version)
Yueshen Xu529 views
Algorithms And Optimization Techniques For Solving TSP by Carrie Romero
Algorithms And Optimization Techniques For Solving TSPAlgorithms And Optimization Techniques For Solving TSP
Algorithms And Optimization Techniques For Solving TSP
Carrie Romero4 views
Optimised Calculation of Symmetries for State Space Reduction by Universität Rostock
Optimised Calculation of Symmetries for State Space ReductionOptimised Calculation of Symmetries for State Space Reduction
Optimised Calculation of Symmetries for State Space Reduction
Comparing reinforcement learning and access points with rowel by ijcseit
Comparing reinforcement learning and access points with rowelComparing reinforcement learning and access points with rowel
Comparing reinforcement learning and access points with rowel
ijcseit276 views
International Journal of Computer Science, Engineering and Information Techno... by ijcseit
International Journal of Computer Science, Engineering and Information Techno...International Journal of Computer Science, Engineering and Information Techno...
International Journal of Computer Science, Engineering and Information Techno...
ijcseit9 views
High-Speed Neural Network Controller for Autonomous Robot Navigation using FPGA by iosrjce
High-Speed Neural Network Controller for Autonomous Robot Navigation using FPGAHigh-Speed Neural Network Controller for Autonomous Robot Navigation using FPGA
High-Speed Neural Network Controller for Autonomous Robot Navigation using FPGA
iosrjce279 views
Vol 16 No 2 - July-December 2016 by ijcsbi
Vol 16 No 2 - July-December 2016Vol 16 No 2 - July-December 2016
Vol 16 No 2 - July-December 2016
ijcsbi103 views
Tuple-Based Coordination in Large-Scale Situated Systems by Roberto Casadei
Tuple-Based Coordination in Large-Scale Situated SystemsTuple-Based Coordination in Large-Scale Situated Systems
Tuple-Based Coordination in Large-Scale Situated Systems
Roberto Casadei82 views
THE IMPACT OF MOBILE NODES ARRIVAL PATTERNS IN MANETS USING POISSON MODELS by IJMIT JOURNAL
THE IMPACT OF MOBILE NODES ARRIVAL PATTERNS IN MANETS USING POISSON MODELSTHE IMPACT OF MOBILE NODES ARRIVAL PATTERNS IN MANETS USING POISSON MODELS
THE IMPACT OF MOBILE NODES ARRIVAL PATTERNS IN MANETS USING POISSON MODELS
IJMIT JOURNAL61 views

More from joisino

キャッシュオブリビアスアルゴリズム by
キャッシュオブリビアスアルゴリズムキャッシュオブリビアスアルゴリズム
キャッシュオブリビアスアルゴリズムjoisino
14 views76 slides
Metric Recovery from Unweighted k-NN Graphs by
Metric Recovery from Unweighted k-NN GraphsMetric Recovery from Unweighted k-NN Graphs
Metric Recovery from Unweighted k-NN Graphsjoisino
31 views45 slides
Towards Principled User-side Recommender Systems by
Towards Principled User-side Recommender SystemsTowards Principled User-side Recommender Systems
Towards Principled User-side Recommender Systemsjoisino
59 views20 slides
CLEAR: A Fully User-side Image Search System by
CLEAR: A Fully User-side Image Search SystemCLEAR: A Fully User-side Image Search System
CLEAR: A Fully User-side Image Search Systemjoisino
17 views19 slides
Private Recommender Systems: How Can Users Build Their Own Fair Recommender S... by
Private Recommender Systems: How Can Users Build Their Own Fair Recommender S...Private Recommender Systems: How Can Users Build Their Own Fair Recommender S...
Private Recommender Systems: How Can Users Build Their Own Fair Recommender S...joisino
233 views24 slides
An Introduction to Spectral Graph Theory by
An Introduction to Spectral Graph TheoryAn Introduction to Spectral Graph Theory
An Introduction to Spectral Graph Theoryjoisino
463 views47 slides

More from joisino(13)

キャッシュオブリビアスアルゴリズム by joisino
キャッシュオブリビアスアルゴリズムキャッシュオブリビアスアルゴリズム
キャッシュオブリビアスアルゴリズム
joisino14 views
Metric Recovery from Unweighted k-NN Graphs by joisino
Metric Recovery from Unweighted k-NN GraphsMetric Recovery from Unweighted k-NN Graphs
Metric Recovery from Unweighted k-NN Graphs
joisino31 views
Towards Principled User-side Recommender Systems by joisino
Towards Principled User-side Recommender SystemsTowards Principled User-side Recommender Systems
Towards Principled User-side Recommender Systems
joisino59 views
CLEAR: A Fully User-side Image Search System by joisino
CLEAR: A Fully User-side Image Search SystemCLEAR: A Fully User-side Image Search System
CLEAR: A Fully User-side Image Search System
joisino17 views
Private Recommender Systems: How Can Users Build Their Own Fair Recommender S... by joisino
Private Recommender Systems: How Can Users Build Their Own Fair Recommender S...Private Recommender Systems: How Can Users Build Their Own Fair Recommender S...
Private Recommender Systems: How Can Users Build Their Own Fair Recommender S...
joisino233 views
An Introduction to Spectral Graph Theory by joisino
An Introduction to Spectral Graph TheoryAn Introduction to Spectral Graph Theory
An Introduction to Spectral Graph Theory
joisino463 views
Word Tour: One-dimensional Word Embeddings via the Traveling Salesman Problem... by joisino
Word Tour: One-dimensional Word Embeddings via the Traveling Salesman Problem...Word Tour: One-dimensional Word Embeddings via the Traveling Salesman Problem...
Word Tour: One-dimensional Word Embeddings via the Traveling Salesman Problem...
joisino3.4K views
最適輸送入門 by joisino
最適輸送入門最適輸送入門
最適輸送入門
joisino11.6K views
ユーザーサイド情報検索システム by joisino
ユーザーサイド情報検索システムユーザーサイド情報検索システム
ユーザーサイド情報検索システム
joisino1.8K views
最適輸送の解き方 by joisino
最適輸送の解き方最適輸送の解き方
最適輸送の解き方
joisino21.6K views
Random Features Strengthen Graph Neural Networks by joisino
Random Features Strengthen Graph Neural NetworksRandom Features Strengthen Graph Neural Networks
Random Features Strengthen Graph Neural Networks
joisino338 views
グラフニューラルネットワークとグラフ組合せ問題 by joisino
グラフニューラルネットワークとグラフ組合せ問題グラフニューラルネットワークとグラフ組合せ問題
グラフニューラルネットワークとグラフ組合せ問題
joisino4.5K views
死にたくない by joisino
死にたくない死にたくない
死にたくない
joisino5.9K views

Recently uploaded

Evaluation and Standardization of the Marketed Polyherbal drug Patanjali Divy... by
Evaluation and Standardization of the Marketed Polyherbal drug Patanjali Divy...Evaluation and Standardization of the Marketed Polyherbal drug Patanjali Divy...
Evaluation and Standardization of the Marketed Polyherbal drug Patanjali Divy...Anmol Vishnu Gupta
7 views10 slides
Conventional and non-conventional methods for improvement of cucurbits.pptx by
Conventional and non-conventional methods for improvement of cucurbits.pptxConventional and non-conventional methods for improvement of cucurbits.pptx
Conventional and non-conventional methods for improvement of cucurbits.pptxgandhi976
20 views35 slides
Discovery of therapeutic agents targeting PKLR for NAFLD using drug repositio... by
Discovery of therapeutic agents targeting PKLR for NAFLD using drug repositio...Discovery of therapeutic agents targeting PKLR for NAFLD using drug repositio...
Discovery of therapeutic agents targeting PKLR for NAFLD using drug repositio...Trustlife
127 views17 slides
Applications of Large Language Models in Materials Discovery and Design by
Applications of Large Language Models in Materials Discovery and DesignApplications of Large Language Models in Materials Discovery and Design
Applications of Large Language Models in Materials Discovery and DesignAnubhav Jain
13 views17 slides
Ellagic Acid and Its Metabolites as Potent and Selective Allosteric Inhibitor... by
Ellagic Acid and Its Metabolites as Potent and Selective Allosteric Inhibitor...Ellagic Acid and Its Metabolites as Potent and Selective Allosteric Inhibitor...
Ellagic Acid and Its Metabolites as Potent and Selective Allosteric Inhibitor...Trustlife
100 views17 slides
Distinct distributions of elliptical and disk galaxies across the Local Super... by
Distinct distributions of elliptical and disk galaxies across the Local Super...Distinct distributions of elliptical and disk galaxies across the Local Super...
Distinct distributions of elliptical and disk galaxies across the Local Super...Sérgio Sacani
33 views12 slides

Recently uploaded(20)

Evaluation and Standardization of the Marketed Polyherbal drug Patanjali Divy... by Anmol Vishnu Gupta
Evaluation and Standardization of the Marketed Polyherbal drug Patanjali Divy...Evaluation and Standardization of the Marketed Polyherbal drug Patanjali Divy...
Evaluation and Standardization of the Marketed Polyherbal drug Patanjali Divy...
Conventional and non-conventional methods for improvement of cucurbits.pptx by gandhi976
Conventional and non-conventional methods for improvement of cucurbits.pptxConventional and non-conventional methods for improvement of cucurbits.pptx
Conventional and non-conventional methods for improvement of cucurbits.pptx
gandhi97620 views
Discovery of therapeutic agents targeting PKLR for NAFLD using drug repositio... by Trustlife
Discovery of therapeutic agents targeting PKLR for NAFLD using drug repositio...Discovery of therapeutic agents targeting PKLR for NAFLD using drug repositio...
Discovery of therapeutic agents targeting PKLR for NAFLD using drug repositio...
Trustlife127 views
Applications of Large Language Models in Materials Discovery and Design by Anubhav Jain
Applications of Large Language Models in Materials Discovery and DesignApplications of Large Language Models in Materials Discovery and Design
Applications of Large Language Models in Materials Discovery and Design
Anubhav Jain13 views
Ellagic Acid and Its Metabolites as Potent and Selective Allosteric Inhibitor... by Trustlife
Ellagic Acid and Its Metabolites as Potent and Selective Allosteric Inhibitor...Ellagic Acid and Its Metabolites as Potent and Selective Allosteric Inhibitor...
Ellagic Acid and Its Metabolites as Potent and Selective Allosteric Inhibitor...
Trustlife100 views
Distinct distributions of elliptical and disk galaxies across the Local Super... by Sérgio Sacani
Distinct distributions of elliptical and disk galaxies across the Local Super...Distinct distributions of elliptical and disk galaxies across the Local Super...
Distinct distributions of elliptical and disk galaxies across the Local Super...
Sérgio Sacani33 views
How to be(come) a successful PhD student by Tom Mens
How to be(come) a successful PhD studentHow to be(come) a successful PhD student
How to be(come) a successful PhD student
Tom Mens524 views
Small ruminant keepers’ knowledge, attitudes and practices towards peste des ... by ILRI
Small ruminant keepers’ knowledge, attitudes and practices towards peste des ...Small ruminant keepers’ knowledge, attitudes and practices towards peste des ...
Small ruminant keepers’ knowledge, attitudes and practices towards peste des ...
ILRI5 views
Small ruminant keepers’ knowledge, attitudes and practices towards peste des ... by ILRI
Small ruminant keepers’ knowledge, attitudes and practices towards peste des ...Small ruminant keepers’ knowledge, attitudes and practices towards peste des ...
Small ruminant keepers’ knowledge, attitudes and practices towards peste des ...
ILRI7 views
CSF -SHEEBA.D presentation.pptx by SheebaD7
CSF -SHEEBA.D presentation.pptxCSF -SHEEBA.D presentation.pptx
CSF -SHEEBA.D presentation.pptx
SheebaD715 views
Open Access Publishing in Astrophysics by Peter Coles
Open Access Publishing in AstrophysicsOpen Access Publishing in Astrophysics
Open Access Publishing in Astrophysics
Peter Coles1.2K views
RemeOs science and clinical evidence by PetrusViitanen1
RemeOs science and clinical evidenceRemeOs science and clinical evidence
RemeOs science and clinical evidence
PetrusViitanen147 views
Effect of Integrated Nutrient Management on Growth and Yield of Solanaceous F... by SwagatBehera9
Effect of Integrated Nutrient Management on Growth and Yield of Solanaceous F...Effect of Integrated Nutrient Management on Growth and Yield of Solanaceous F...
Effect of Integrated Nutrient Management on Growth and Yield of Solanaceous F...
SwagatBehera95 views
Structure of purines and pyrimidines - Jahnvi arora (11228108), mmdu ,mullana... by jahnviarora989
Structure of purines and pyrimidines - Jahnvi arora (11228108), mmdu ,mullana...Structure of purines and pyrimidines - Jahnvi arora (11228108), mmdu ,mullana...
Structure of purines and pyrimidines - Jahnvi arora (11228108), mmdu ,mullana...
jahnviarora9896 views
ELECTRON TRANSPORT CHAIN by DEEKSHA RANI
ELECTRON TRANSPORT CHAINELECTRON TRANSPORT CHAIN
ELECTRON TRANSPORT CHAIN
DEEKSHA RANI10 views

Fast Unbalanced Optimal Transport on a Tree

  • 1. 1 KYOTO UNIVERSITY KYOTO UNIVERSITY Fast Unbalanced Optimal Transport on a Tree Ryoma Sato Kyoto University / RIKEN AIP
  • 2. 2 / 13 KYOTO UNIVERSITY Self Introduction  I am a second year master's student at Kyoto University  I’m interested in algorithmic aspects of machine learning and data mining for structured data, including Graph neural networks:  Ryoma Sato, Makoto Yamada, Hisashi Kashima. Approximation Ratios of Graph Neural Networks for Combinatorial Problems. NeurIPS 2019.  Ryoma Sato, Makoto Yamada, Hisashi Kashima. Random Features Strengthen Graph Neural Networks. SDM 2021 Optimal transport:  Ryoma Sato, Makoto Yamada, Hisashi Kashima. Fast Unbalanced Optimal Transport on a Tree. NeurIPS 2020.  Today’s topic
  • 3. 3 / 13 KYOTO UNIVERSITY Background: optimal transport is useful  The optimal transport (OT) distance measures the discrepancy of two distributions. We consider discrete distributions in this presentation.  The OT distance is the minimum cumulative distance that all masses need to travel from one distribution to another distribution  In generative modeling, a mass is a sample. discrepancy of model  sample distribution  In text classification, a mass is a word.  OT does not require the same support  KL divergence  OT exploits the underground geometry From Word Embeddings To Document Distances, ICML 2015
  • 4. 4 / 13 KYOTO UNIVERSITY Background: sliced OT is computationally cheap  OT is formulated as a linear program  cubic cost  Sliced OT projects distributions to random 1D spaces and computes OT there  Greedy matching solves 1D OT exactly  linear cost : distance matrix (input), : matching matrix (variable) : 1st mass vector (input), : 2nd mass vector (input) The leftmost mass should be matched to the leftmost mass The second leftmost mass should be matched to the second leftmost mass ... https://www.programmersought.com/article/67174999352/ https://analyticsindiamag.com/how-to-establish-domain-transferability-in-neural-models/
  • 5. 5 / 13 KYOTO UNIVERSITY Background: unbalanced OT is robust  OT is sensitive to outliers because transporting outliers becomes the dominating term  Unbalanced OT (UOT) allows to discard and create masses by paying some penalties  We can discard outliers  robust to outliers  UOT is also formulated by a linear program  cubic cost
  • 6. 6 / 13 KYOTO UNIVERSITY Background: UOT is difficult even in 1D spaces  We want to make a cheap alternative of UOT as 1D OT  But the greedy matching fails to solve 1D UOT  Let’s consider the following instance with discard cost λ  The following plan costs 3λ.  The following plan costs 2λ + 2ε. Thus this is better. λ λ λ
  • 7. 7 / 13 KYOTO UNIVERSITY Background: UOT is difficult even in 1D spaces  Let’s consider the following instance with discard cost λ  The following plan costs λ + 2ε.  The following plan costs 2λ + 2ε. Thus this is worse. λ ε
  • 8. 8 / 13 KYOTO UNIVERSITY Background: UOT is difficult even in 1D spaces  Although these two instances share the leftmost part, the leftmost mass in the first instance should be discarded while that in the second instance should not  The optimal UOT plan cannot be determined locally  The optimal OT plan is determined locally  Thus the greedy algorithm fails to solve 1D UOT  We proposed how to solve 1D UOT efficiently λ λ λ λ ε
  • 9. 9 / 13 KYOTO UNIVERSITY Algorithm: prune redundant plans  Our proposed method determines assignments from left to right (as the greedy algorithm)  Although there are exponentially many plans, most of them are redundant. We proved that only O(n) plans are non-redundant  Only one plan is non-redundant (thus greedy is valid) in the standard OT not yet not yet  non redundant  redundant
  • 10. 10 / 13 KYOTO UNIVERSITY Algorithm: we solve 1D UOT in O(n log2 n) time  A naive algorithm requires cubic time even with this (non redundant plan) observation  More algorithmic techniques are required for further speedup (skipped in this presentation)  Dynamic programming  Fast convex min-sum convolution  Efficient data structure (BBST)  Weighted union heuristics  Finally, we derived a quasi-linear time algorithm which runs in O(n log2 n) time in the worst case
  • 11. 11 / 13 KYOTO UNIVERSITY Algorithm: tree UOT generalizes 1D UOT  Our method can be extended to tree spaces A 1D space (path) is a special case of tree spaces  In text classification, the word space can be represented by a word tree. Each mass (word) travels on the word tree to a nearby (semantically similar) word.  We can “tree-slice” high dimensional spaces instead of 1D-slicing, which captures richer structures http://www.sthda.com/english/articles/31-principal-component-methods-in-r-practical-guide/117-hcpc-hierarchical-clustering-on-principal-components-essentials/ a → b
  • 12. 12 / 13 KYOTO UNIVERSITY Experiments: our algorithm is empirically fast  We confirmed that our algorithm could compute tree UOT with one million masses within one second  We also confirmed that tree-slicing high dimensional spaces could approximate the original UOT problem
  • 13. 13 / 13 KYOTO UNIVERSITY Conclusion: fast computation of tree UOT  Sliced OT is a fast alternative of OT  UOT is a robust variant of OT  1D UOT is more difficult than 1D OT  We proposed an efficient algorithm for 1D UOT for the first time  Our method can be extended to tree spaces  Our method is empirically fast (1M masses in 1 sec)