using meta learning to initialize bayesian optimization
play

Using Meta Learning to Initialize Bayesian Optimization - PowerPoint PPT Presentation

Aug 23, 2022 •365 likes •852 views

Using Meta Learning to Initialize Bayesian Optimization Albert-Ludwigs-Universitt Freiburg Matthias Feurer 1 Jost Tobias Springenberg 2 Frank Hutter 1 1 Research Group on Learning, Optimization, and Automated Algorithm Design 2 Machine Learning

  1. Using Meta Learning to Initialize Bayesian Optimization Albert-Ludwigs-Universität Freiburg Matthias Feurer 1 Jost Tobias Springenberg 2 Frank Hutter 1 1 Research Group on Learning, Optimization, and Automated Algorithm Design 2 Machine Learning Lab Department of Computer Science, University of Freiburg, Germany ECAI-2014 Workshop on Meta-learning & Algorithm Selection, 19 August 2014

  2. Your task: Build an Iris classification system The iris pictures on this slide are from wikimedia commons and used under the following licenses: Top left: Iris Versicolor is public domain; Bottom left: Iris setosa is licensed by Radomil under CC BY-SA 3.0; Top right: Iris Virginica is licensed by C T Johansson under CC BY 3.0. MetaSel ’14 Feurer, Springenberg and Hutter – MI-SMBO 2 / 18

  3. Your task: Build an Iris classification system Choose an algorithm based on dataset characteristics, e.g. for the Iris dataset this could be an SVM The iris pictures on this slide are from wikimedia commons and used under the following licenses: Top left: Iris Versicolor is public domain; Bottom left: Iris setosa is licensed by Radomil under CC BY-SA 3.0; Top right: Iris Virginica is licensed by C T Johansson under CC BY 3.0. MetaSel ’14 Feurer, Springenberg and Hutter – MI-SMBO 2 / 18

  4. Your task: Build an Iris classification system Choose an algorithm based on dataset characteristics, e.g. for the Iris dataset this could be an SVM Manual tuning -> fiddling with hyperparameters. The iris pictures on this slide are from wikimedia commons and used under the following licenses: Top left: Iris Versicolor is public domain; Bottom left: Iris setosa is licensed by Radomil under CC BY-SA 3.0; Top right: Iris Virginica is licensed by C T Johansson under CC BY 3.0. MetaSel ’14 Feurer, Springenberg and Hutter – MI-SMBO 2 / 18

  5. Your task: Build an Iris classification system Choose an algorithm based on dataset characteristics, e.g. for the Iris dataset this could be an SVM Manual tuning -> fiddling with hyperparameters. Better: Use automated methods like PSO, GA or SMBO The iris pictures on this slide are from wikimedia commons and used under the following licenses: Top left: Iris Versicolor is public domain; Bottom left: Iris setosa is licensed by Radomil under CC BY-SA 3.0; Top right: Iris Virginica is licensed by C T Johansson under CC BY 3.0. MetaSel ’14 Feurer, Springenberg and Hutter – MI-SMBO 2 / 18

  6. Your task: Build an Iris classification system Choose an algorithm based on dataset characteristics, e.g. for the Iris dataset this could be an SVM Manual tuning -> fiddling with hyperparameters. Better: Use automated methods like PSO, GA or SMBO Best: AutoWeka The iris pictures on this slide are from wikimedia commons and used under the following licenses: Top left: Iris Versicolor is public domain; Bottom left: Iris setosa is licensed by Radomil under CC BY-SA 3.0; Top right: Iris Virginica is licensed by C T Johansson under CC BY 3.0. MetaSel ’14 Feurer, Springenberg and Hutter – MI-SMBO 2 / 18

  7. Adding the Iris Japonica to the dataset The iris pictures on this slide are from wikimedia commons and used under the following licenses: Top left: Iris Versicolor is public domain; Bottom left: Iris setosa is licensed by Radomil under CC BY-SA 3.0; Top right: Iris Virginica is licensed by C T Johansson under CC BY 3.0; Bottom right: Iris Japonica is licensed by KENPEI under CC BY-SA 3.0 MetaSel ’14 Feurer, Springenberg and Hutter – MI-SMBO 3 / 18

  8. Adding the Iris Japonica to the dataset Manual tuning: Use experience and start from the parameters found on the Iris dataset The iris pictures on this slide are from wikimedia commons and used under the following licenses: Top left: Iris Versicolor is public domain; Bottom left: Iris setosa is licensed by Radomil under CC BY-SA 3.0; Top right: Iris Virginica is licensed by C T Johansson under CC BY 3.0; Bottom right: Iris Japonica is licensed by KENPEI under CC BY-SA 3.0 MetaSel ’14 Feurer, Springenberg and Hutter – MI-SMBO 3 / 18

  9. Adding the Iris Japonica to the dataset Manual tuning: Use experience and start from the parameters found on the Iris dataset Automated methods -> start from scratch The iris pictures on this slide are from wikimedia commons and used under the following licenses: Top left: Iris Versicolor is public domain; Bottom left: Iris setosa is licensed by Radomil under CC BY-SA 3.0; Top right: Iris Virginica is licensed by C T Johansson under CC BY 3.0; Bottom right: Iris Japonica is licensed by KENPEI under CC BY-SA 3.0 MetaSel ’14 Feurer, Springenberg and Hutter – MI-SMBO 3 / 18

  10. Adding the Iris Japonica to the dataset Manual tuning: Use experience and start from the parameters found on the Iris dataset Automated methods -> start from scratch → Cast use experience into an algorithm. The iris pictures on this slide are from wikimedia commons and used under the following licenses: Top left: Iris Versicolor is public domain; Bottom left: Iris setosa is licensed by Radomil under CC BY-SA 3.0; Top right: Iris Virginica is licensed by C T Johansson under CC BY 3.0; Bottom right: Iris Japonica is licensed by KENPEI under CC BY-SA 3.0 MetaSel ’14 Feurer, Springenberg and Hutter – MI-SMBO 3 / 18

  11. Sequential Model-based Bayesian Optimization (SMBO) ML Algorithm A Configuration Space Λ of A Dataset D MetaSel ’14 Feurer, Springenberg and Hutter – MI-SMBO 4 / 18

  12. Sequential Model-based Bayesian Optimization (SMBO) ML Algorithm A Configuration Space Λ of A Dataset D Configuration Task Configuration λ ∗ MetaSel ’14 Feurer, Springenberg and Hutter – MI-SMBO 4 / 18

  13. Sequential Model-based Bayesian Optimization (SMBO) Fit regression Select promising ML Algorithm A model on configuration ( λ , A λ ( D )) pairs λ ∈ Λ Configuration Space Λ of A Evaluate A λ ( D ) Dataset D Configuration Task Configuration λ ∗ MetaSel ’14 Feurer, Springenberg and Hutter – MI-SMBO 4 / 18

  14. Metalearning-Initialized SMBO (MI-SMBO) Fit regression Select promising ML Algorithm A model on pairs of configuration ( λ , A λ ( D new )) λ ∈ Λ Configuration Space Λ of A Evaluate Dataset D new A λ ( D new ) Configuration Task Configuration λ ∗ MetaSel ’14 Feurer, Springenberg and Hutter – MI-SMBO 5 / 18

  15. Metalearning-Initialized SMBO (MI-SMBO) Initialize Search Find Datasets D i with λ ∗ similar to D new D i Fit regression Select promising ML Algorithm A model on pairs of configuration ( λ , A λ ( D new )) λ ∈ Λ Configuration Space Λ of A Evaluate Dataset D new A λ ( D new ) Configuration Task Configuration λ ∗ MetaSel ’14 Feurer, Springenberg and Hutter – MI-SMBO 5 / 18

  16. Metafeatures # training examples: 150 # classes: 3 # features: 4 # numerical features: 4 # categorical features: 0 missing values? No The iris pictures on this slide are from wikimedia commons and used under the following licenses: Top left: Iris Versicolor is public domain; Bottom left: Iris setosa is licensed by Radomil under CC BY-SA 3.0; Top right: Iris Virginica is licensed by C T Johansson under CC BY 3.0. MetaSel ’14 Feurer, Springenberg and Hutter – MI-SMBO 6 / 18

  17. Metalearning-Initialized Bayesian Optimization For a new dataset D new : Sort known datasets D 1 : N by distance to D new . For each of these datasets, extract the best known hyperparameter configuration λ ∗ D i . Initialize SMBO with the first k hyperparameter configurations from the sorted list. MetaSel ’14 Feurer, Springenberg and Hutter – MI-SMBO 7 / 18

  18. Similarity of Datasets a s y i u g m t n p a o h e l t y G o y s l b i h o a R I y d t c o o u r u u o O s r . s a a a e z l d a n e a r o n b m n - a r u e d - t y n i a r y l m c i y r p r o g b o s l a i a o l t m t t i t a a a s d m p r i - e u t r h n e o e d t e l i l t i z a x a p i p b r - - h e t - v a t i e r t f a a m e r e h p s r o o t s t o c a p f h - t p a m e f y m l r e i e r u o t o f v r - e t r c a a n e n a m f o c e s - t k s n i a r e e r z b c - t t - a r e e a r e m f e e h h n p s o l o a n b o a a i - t i d e l r e c g w - g t o i o d l v e a t r c t s - r t a n n e a e h m n r u e h b r a w a h k - - s t t a a e e l e a r o m f c b t s - r g i c e a d o t l - r o c o h i s p t i d r - o r m s e t v s - i e t g l i a i t e d e f t b m p a o d i p k s - m r v e - o t r t k o e r l h s u m s t t e s y e i c 0 e e s r s e t s n g s r o l m 0 l k a i s i a e a e c l a a c c r g a i c c 0 t c i a i e m m g l s h e 5 o b d y r e s l m n g i u m - - b e e t n v e - a c e a p s s r n g p o a s f a e a l p v a b w MetaSel ’14 Feurer, Springenberg and Hutter – MI-SMBO 8 / 18

Recommend

Regret bounds for meta Bayesian optimization with an unknown Gaussian process prior Zi Wang*

Regret bounds for meta Bayesian optimization with an unknown Gaussian process prior Zi Wang*

Regret bounds for meta Bayesian optimization with an unknown Gaussian process prior Zi Wang* Beomjoon Kim* Leslie Pack Kaelbling Dec 5 @ NeurIPS 18 Poster #22 Bayesian optimization x * = argmax Goal: f ( x ) x

346 views • 20 slides
Bayesian Model-Agnostic Meta-Learning Taesup Kim* (presenter), Jaesik Yoon* Ousmane Dia,

Bayesian Model-Agnostic Meta-Learning Taesup Kim* (presenter), Jaesik Yoon* Ousmane Dia,

Bayesian Model-Agnostic Meta-Learning Taesup Kim* (presenter), Jaesik Yoon* Ousmane Dia, Sungwoong Kim, Yoshua Bengio, Sungjin Ahn Model-Agnostic Meta-learning (MAML) gradient-based meta-learning framework meta-update task adaptation

886 views • 22 slides
Bayesian Meta-Learning CS 330 1 Logistics Homework 2 due next Wednesday. Project proposal due in

Bayesian Meta-Learning CS 330 1 Logistics Homework 2 due next Wednesday. Project proposal due in

Bayesian Meta-Learning CS 330 1 Logistics Homework 2 due next Wednesday. Project proposal due in two weeks . Poster presentation: Tues 12/3 at 1:30 pm . 2 Disclaimers Bayesian meta-learning is an ac#ve area of research (like most of the

622 views • 33 slides
Optimization-Based Meta-Learning CS 330 1 Course Reminders HW1 due next Weds (9/30). Project

Optimization-Based Meta-Learning CS 330 1 Course Reminders HW1 due next Weds (9/30). Project

Optimization-Based Meta-Learning CS 330 1 Course Reminders HW1 due next Weds (9/30). Project guidelines posted start forming groups & formulating ideas. Guest lecture by Matt Johnson on Monday! 2 Plan for Today Recap - Meta-learning

1.43k views • 33 slides
Bayesian Optimization CSC2541 - Topics in Machine Learning Scalable and Flexible Models of

Bayesian Optimization CSC2541 - Topics in Machine Learning Scalable and Flexible Models of

Bayesian Optimization CSC2541 - Topics in Machine Learning Scalable and Flexible Models of Uncertainty University of Toronto - Fall 2017 Overview 1. Bayesian Optimization of Machine Learning Algorithms 2. Gaussian Process Optimization in the

1.53k views • 94 slides
Outline Intro to RL and Bayesian Learning History of Bayesian RL Model-based Bayesian

Outline Intro to RL and Bayesian Learning History of Bayesian RL Model-based Bayesian

Outline Intro to RL and Bayesian Learning History of Bayesian RL Model-based Bayesian RL Prior knowledge, policy optimization, discussion, Bayesian approaches for other RL variants Model-free Bayesian RL Gaussian

1.23k views • 63 slides
Meta-Bayesian Analysis A Bayesian decision-theoretic analysis of Bayesian inference under model

Meta-Bayesian Analysis A Bayesian decision-theoretic analysis of Bayesian inference under model

Meta-Bayesian Analysis A Bayesian decision-theoretic analysis of Bayesian inference under model misspecification Jun Yang joint work with Daniel Roy Department of Statistical Sciences University of Toronto World Congress in Probability and

190 views • 17 slides
Bayesian Optimization in Reduced Eigenbases David Gaudrie 1 , Rodolphe Le Riche 2 , Victor Picheny

Bayesian Optimization in Reduced Eigenbases David Gaudrie 1 , Rodolphe Le Riche 2 , Victor Picheny

Parametric shape optimization Intrinsic dimensionality Meta-modeling in eigenbasis Optimization in eigenbasis Conclusions Bayesian Optimization in Reduced Eigenbases David Gaudrie 1 , Rodolphe Le Riche 2 , Victor Picheny 3 , t Enaux 1 ,

581 views • 55 slides
Me Meta Lear Learnin ing A Bri Brief Introduct ction Xiachong Feng Ou Outline

Me Meta Lear Learnin ing A Bri Brief Introduct ction Xiachong Feng Ou Outline

Me Meta Lear Learnin ing A Bri Brief Introduct ction Xiachong Feng Ou Outline Introduction to Meta Learning Types of Meta-Learning Models Papers: Optimization as a model for few-shot learning ICLR2017

913 views • 60 slides
BAYESIAN GLOBAL OPTIMIZATION Using Optimal Learning to Tune Deep Learning Pipelines Scott Clark

BAYESIAN GLOBAL OPTIMIZATION Using Optimal Learning to Tune Deep Learning Pipelines Scott Clark

BAYESIAN GLOBAL OPTIMIZATION Using Optimal Learning to Tune Deep Learning Pipelines Scott Clark scott@sigopt.com OUTLINE 1. Why is Tuning AI Models Hard? 2. Comparison of Tuning Methods 3. Bayesian Global Optimization 4. Deep Learning

983 views • 67 slides
The Peculiar Optimization and Regularization Challenges in Multi-Task Learning and Meta-Learning

The Peculiar Optimization and Regularization Challenges in Multi-Task Learning and Meta-Learning

The Peculiar Optimization and Regularization Challenges in Multi-Task Learning and Meta-Learning Chelsea Finn Stanford training data test datapoint Braque Cezanne By Braque or Cezanne? How did you accomplish this? Through previous

851 views • 43 slides
Optimization-Based Meta-Learning ( fi nishing from last time) and Non-Parametric Few-Shot

Optimization-Based Meta-Learning ( fi nishing from last time) and Non-Parametric Few-Shot

Optimization-Based Meta-Learning ( fi nishing from last time) and Non-Parametric Few-Shot Learning CS 330 1 Logistics Homework 1 due, Homework 2 out this Wednesday Fill out poster presentation preferences ! (Tues 12/3 or Weds 12/4) Course

554 views • 30 slides
Learning Bayesian networks viewed as an optimization problem Milan Studen y Institute of

Learning Bayesian networks viewed as an optimization problem Milan Studen y Institute of

Learning Bayesian networks viewed as an optimization problem Milan Studen y Institute of Information Theory and Automation of the AS CR Prague COSA Workshop Combinatorial Optimization, Statistics, and Applications Munich, Germany, March

458 views • 29 slides
Meta Reinforcement Learning as Task Inference Jan Humplik, Alexandre Galashov, Leonard

Meta Reinforcement Learning as Task Inference Jan Humplik, Alexandre Galashov, Leonard

Meta Reinforcement Learning as Task Inference Jan Humplik, Alexandre Galashov, Leonard Hasenclever, Pedro A.Ortega, Yee Whye Teh, Nicholas Heess Topic: Bayesian RL Presenter: Ram Ananth Why meta Reinforcement Learning? First Wave of Deep

912 views • 63 slides
Me Meta Lear Learnin ing A Bri Brief Introduct ction Xiachong Feng TG Ph.D. Student

Me Meta Lear Learnin ing A Bri Brief Introduct ction Xiachong Feng TG Ph.D. Student

Me Meta Lear Learnin ing A Bri Brief Introduct ction Xiachong Feng TG Ph.D. Student 2018-12-01 Ou Outline Introduction to Meta Learning Types of Meta-Learning Models Papers: Optimization as a model for few-shot

745 views • 60 slides
Q Learning Forall s, a Initialize Q(s, a) = 0 Repeat Forever Where are you? s. Choose some

Q Learning Forall s, a Initialize Q(s, a) = 0 Repeat Forever Where are you? s. Choose some

11/9/16 Approximate Q-Learning Dan Weld / University of Washington [Many slides taken from Dan Klein and Pieter Abbeel / CS188 Intro to AI at UC Berkeley materials available at http://ai.berkeley.edu.] Q Learning Forall s, a Initialize

523 views • 20 slides
Meta Learning Shengchao Liu Background Meta Learning (AKA Learning to Learn) A

Meta Learning Shengchao Liu Background Meta Learning (AKA Learning to Learn) A

Meta Learning Shengchao Liu Background Meta Learning (AKA Learning to Learn) A fast-learning algorithm: quickly adapted from the source tasks to the target tasks Key terminologies Support Set & Query Set C-Way K-Shot

2k views • 41 slides
Safe Bayesian Optimization for Optimal Control Schillinger, M., Hartmann, B., Skalecki, P.,

Safe Bayesian Optimization for Optimal Control Schillinger, M., Hartmann, B., Skalecki, P.,

Safe Bayesian Optimization for Optimal Control Schillinger, M., Hartmann, B., Skalecki, P., Meister, M., Nguyen-Tuong, D., & Nelles, O. (2017). Safe Active Learning and Safe Bayesian Optimization for Tuning a PI-Controller. IFAC-PapersOnLine ,

203 views • 6 slides
Bayesian Learning 1 Outline MLE, MAP vs. Bayesian Learning Bayesian Linear Regression

Bayesian Learning 1 Outline MLE, MAP vs. Bayesian Learning Bayesian Linear Regression

Bayesian Learning 1 Outline MLE, MAP vs. Bayesian Learning Bayesian Linear Regression Bayesian Gaussian Mixture Models Non-parametric Bayes 2 Take Away ... 1. Maximum Likelihood Estimate (MLE) = arg max p ( D| )

609 views • 27 slides
Bayesian Optimization and Automated Machine Learning Jungtaek Kim (jtkim@postech.ac.kr) Machine

Bayesian Optimization and Automated Machine Learning Jungtaek Kim (jtkim@postech.ac.kr) Machine

Bayesian Optimization and Automated Machine Learning Jungtaek Kim (jtkim@postech.ac.kr) Machine Learning Group, Department of Computer Science and Engineering, POSTECH, 77 Cheongam-ro, Nam-gu, Pohang 37673, Gyeongsangbuk-do, Republic of Korea

961 views • 28 slides
R2-B2: Recursive Reasoning-Based Bayesian Optimization for No-Regret Learning in Games Zhongxiang

R2-B2: Recursive Reasoning-Based Bayesian Optimization for No-Regret Learning in Games Zhongxiang

R2-B2: Recursive Reasoning-Based Bayesian Optimization for No-Regret Learning in Games Zhongxiang Dai 1 Yizhou Chen 1 Bryan Kian Hsiang Low 1 Patrick Jaillet 2 Teck-Hua Ho 3 1 Department of Computer Science, National University of Singapore 2

534 views • 18 slides
Distributed Meta Optimization of Reinforcement Learning Agents Greg Heinrich, Iuri Frosio - GTC

Distributed Meta Optimization of Reinforcement Learning Agents Greg Heinrich, Iuri Frosio - GTC

Distributed Meta Optimization of Reinforcement Learning Agents Greg Heinrich, Iuri Frosio - GTC San Jose, March 2019 Contents Introduction to Reinforcement Learning Introduction to Metaoptimization (on distributed systems) / Maglev

596 views • 45 slides
The he Right Right Tool ool for the or the Job: ob: A Bayesian Meta-Regression of Employment

The he Right Right Tool ool for the or the Job: ob: A Bayesian Meta-Regression of Employment

The he Right Right Tool ool for the or the Job: ob: A Bayesian Meta-Regression of Employment and Training Studies Presentation at the OPRE Methods Inquiries Meeting October 20, 2017 Lauren Vollmer Emily Sama - Miller Alyssa

419 views • 26 slides
Meta-Bayesian Analysis Jun Yang joint work with Daniel M. Roy Department of Statistical Sciences

Meta-Bayesian Analysis Jun Yang joint work with Daniel M. Roy Department of Statistical Sciences

Meta-Bayesian Analysis Jun Yang joint work with Daniel M. Roy Department of Statistical Sciences University of Toronto ISBA 2016 June 16, 2016 Meta-Bayesian Analysis (Yang) 1 Motivation All models are wrong, truth [...] is much too

529 views • 16 slides

More recommend