HYDROGEN STORAGE IN MAGNESIUM BASED ALLOYS Jasmina Grbović Novaković.
Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer,...
-
Upload
magnus-barton -
Category
Documents
-
view
228 -
download
0
Transcript of Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer,...
![Page 1: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/1.jpg)
Multi-Task Learning for HIV Therapy Screening
Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer
![Page 2: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/2.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
HIV Therapy Screening Usually combinations (3-6 drugs) out of around
17 antiretroviral drugs administered.
Effect of combinations on virus similar but not identical.
Scarce training data available from treatment records.
Challenge: Prediction of therapy outcome from genotypic information.
data for combination 1 data for combination 2
successful treatment failed treatment
data for comb. 3
![Page 3: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/3.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
Multi-Task Learning Several related prediction problems (tasks).
Not necessarily identical conditional p(y|x) of label given input.
Usually, some conditionals are similar.
Challenge: Use all available training data and account for the
difference in distributions accross tasks.
HIV therapy screening: Can be modeled as multi-task learning problem. Drug combinations (tasks) have similar but not
identical effect on the virus.
![Page 4: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/4.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
62
66
70
74
78
82
Acc
ura
cy %
Overview
Motivation.
HIV therapy screening.
Multiple tasks with differing distributions.
Multi-task learning by distribution matching.
Problem Setting.
Density ratio matches pool to target distribution.
Discriminative estimation of matching weights.
Case study:
HIV therapy screening.
![Page 5: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/5.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
Multi-Task Learning – Problem Setting
Target distribution
Labeled target data
![Page 6: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/6.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
Goal: Minimize loss under target distribution.
Multi-Task Learning – Problem Setting
Target distribution
Labeled target data
![Page 7: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/7.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
Multi-Task Learning – Problem Setting
Target distribution
Labeled target data
Goal: Minimize loss under target distribution.
![Page 8: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/8.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
Multi-Task Learning – Problem Setting
Target distribution Auxiliary distributions
Labeled target data
Labeled auxiliary data
Goal: Minimize loss under target distribution.
![Page 9: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/9.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
Multi-Task Learning – Problem Setting
Target distribution Auxiliary distributions
Labeled target data
Labeled auxiliary data
Problem Setting: M
ulti-Task Learning
Goal: Minimize loss under target distribution.
![Page 10: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/10.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
Multi-Task Learning – Problem Setting
Target distribution Auxiliary distributions
Labeled target data
Labeled auxiliary data
Goal: Minimize loss under target distribution.
![Page 11: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/11.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
Multi-Task Learning
Target distribution Pool distribution
≠
Labeled target data
Goal: Minimize loss under target distribution.
![Page 12: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/12.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
Distribution Matching
Target distribution Pool distribution
Labeled target data
Goal: Minimize loss under target distribution.
=
![Page 13: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/13.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
Distribution Matching
Target distribution Pool distribution
Labeled target data
Goal: Minimize loss under target distribution.
=
Expected loss under target distribution
Rescale loss for each pool example
Expectation over training pool
![Page 14: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/14.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
Distribution Matching Goal: Minimize loss under target distribution.
=
xy=−1
xy=+1
Target distribution Pool distribution
![Page 15: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/15.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
Distribution Matching Goal: Minimize loss under target distribution.
=
xy=−1
xy=+1
Target distribution Pool distribution
x1
0.10.22
![Page 16: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/16.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
Distribution Matching Goal: Minimize loss under target distribution.
=
xy=−1
xy=+1
Target distribution Pool distribution
x1
0.100
x2
![Page 17: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/17.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
Estimation of Density Ratio Goal: Minimize loss under target distribution.
=
![Page 18: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/18.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
Estimation of Density Ratio Goal: Minimize loss under target distribution.
Theorem:
Potentially high-dimensional densities
=
One binary conditional density
![Page 19: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/19.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
Estimation of Density Ratio Goal: Minimize loss under target distribution.
Theorem:
Intuition of : how much more likely is to be drawn from target than from auxiliary density.
=
Pool
![Page 20: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/20.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
Estimation of Density Ratio Goal: Minimize loss under target distribution.
Theorem:
Intuition of : how much more likely is to be drawn from target than from auxiliary density.
=
Pool
Targetexamples
auxiliarytask examples
![Page 21: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/21.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
Estimation of Density Ratio Goal: Minimize loss under target distribution.
Theorem:
Intuition of : how much more likely is to be drawn from target than from auxiliary density.
=
Pool
auxiliarytask examples
Estimation of
with probabilistic
classifier (e.g., logreg)Target
examples
![Page 22: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/22.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
Estimation of Density Ratio Goal: Minimize loss under target distribution.
Theorem:
Intuition of : how much more likely is to be drawn from target than from auxiliary density.
=
Pool
Targetexamples
auxiliarytask examples
towards blue
larger
large resampling weights
![Page 23: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/23.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
Prior Knowledge on Task Similarity Prior knowledge in task similarity kernel .
Encoding of prior knowledge in Gaussian prior
on parameters v of a multi-class logistic regression model for the resampling weights.
Main diagonal entries of set to (standard regularizer),
Diagonals of sub-matrices set to .
2v
![Page 24: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/24.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
1. Weight Model:Train Logreg of target vs. auxiliary data with task similarity in .
2. Target Model:
Minimize regularized empirical loss on pool weighted by .
Distribution Matching – Algorithm
Result of step 1: weight model
![Page 25: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/25.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
62
66
70
74
78
82
Acc
ura
cy %
Overview
Motivation.
HIV therapy screening.
Multiple tasks with differing distributions.
Multi-task learning by distribution matching.
Problem Setting.
Density ratio matches pool to target distribution.
Discriminative estimation of matching weights.
Case study:
HIV therapy screening.
![Page 26: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/26.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
HIV Therapy Screening – Prediction Problem
Information about each patient x, binary vector of resistance-relevant virus mutations and of previously given drugs.
Drug combination selected out of 17 drugs. Drug combinations correspond to tasks z.
Target label y (success or failure of therapy). 2 different labelings (virus load and multi-conditional).
viru
s lo
ad
time conditions
![Page 27: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/27.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
HIV Therapy Screening – Data
Patients from hospitals in Italy, Germany, and Sweden. 3260 labeled treatments. 545 different drug combinations (tasks). 50% of combinations with only one labeled treatment.
Similarity of drug combinations: task kernel. Drug feature kernel: product of drug indicator vectors. Mutation table kernel: similarity of mutations that render
drug ineffective.
80/20 training/test split, consistent with time stamps.
training data
time
test data
![Page 28: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/28.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
Reference Methods Independent models (separately trained). One-size-fits-all, product of task and feature kernel,
Bonilla, Agakov, and Williams (2007). Hierarchical Bayesian Kernel,
Evgeniou & Pontil (2004). Hierarchical Bayesian Gaussian Process
Yu, Tresp, and Schwaighofer (2005).
Logistic regression is target model (except for Gaussian process model). RBF kernels.
![Page 29: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/29.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
62
66
70
74
78
82
Acc
ura
cy %
Results – Distribution Matching vs. Other
Distribution matching always best (17 of 20 cases stat. significant) or as good as best reference method.
Improvement over separately trained models 10-14%.
separate one-size-fits-all
hier. Bayeskernel
hier. BayesGauss. Proc.
distributionmatching
virus load
multi-condition
![Page 30: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/30.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
77
78
79
80
Ac
cu
rac
y %
Results – Benefit of Prior Knowledge
The common prior knowledge on similarity of drug combinations does not improve accuracy of distribution matching.
virus load
multi-condition
no priorknowledge
drug. feat.kernel
Mut. tablekernel
![Page 31: Multi-Task Learning for HIV Therapy Screening Steffen Bickel, Jasmina Bogojeska, Thomas Lengauer, Tobias Scheffer.](https://reader035.fdocuments.us/reader035/viewer/2022062322/56649e895503460f94b8ea27/html5/thumbnails/31.jpg)
Bickel, B
ogo
jeska, Len
gau
er, S
cheffer
Conclusions
Multi-task Learning: Multiple problems with different distributions.
Distribution matching: Weighted pool distribution matches target distribution.
Discriminative estimation of weights with Logreg.
Training of target model with weighted loss terms.
Case study: HIV therapy screening. Distribution matching beats iid learning and hier. Bayes.
Benefit over separately trained models 10-14%.