SlideShare a Scribd company logo
1 of 41
Download to read offline
Takeaways from ICML 2019
Hong Kong Machine Learning Meetup
Season 1 Episode 12 – Season Finale
Gautier Marti
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 1 / 41
Table of contents
1 Day 1 - Tutorials
2 Day 2 - U.S. Census, Time Series, Hawkes Processes, Shapley values,
Topological Data Analysis, Optimal Transport for Graphs
3 Day 3 - Robotics, Gaussian Processes, Learning with noisy labels
4 Day 4 - Interpretability, Natural Language Processing
5 Day 5 - Workshop Time Series
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 2 / 41
Disclaimer
I was sent to the ICML 2019 conference by my employer Shell Street Labs.
However, the opinions expressed in this presentation are my own and do
not reflect in any ways the view of my employer.
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 3 / 41
Section 1
Day 1 - Tutorials
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 4 / 41
Subsection 1
Safe Machine Learning
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 5 / 41
Safe Machine Learning
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 6 / 41
COMPAS - An example of ML biases & misspecification
We show, however, that the widely used commercial risk
assessment software COMPAS is no more accurate or fair than
predictions made by people with little or no criminal justice
expertise.
The accuracy, fairness, and limits of predicting recidivism
https://advances.sciencemag.org/content/4/1/eaao5580
Our analysis of Northpointe’s tool, called COMPAS (which
stands for Correctional Offender Management Profiling for
Alternative Sanctions), found that black defendants were far
more likely than white defendants to be incorrectly judged to be
at a higher risk of recidivism, while white defendants were more
likely than black defendants to be incorrectly flagged as low risk.
How We Analyzed the COMPAS Recidivism Algorithm
https://www.propublica.org/article/
how-we-analyzed-the-compas-recidivism-algorithm
GitHub: https://github.com/propublica/compas-analysis
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 7 / 41
RL in the Wild - Other examples of misspecifications
Reinforcement learning algorithms can break in surprising,
counterintuitive ways.
Faulty Reward Functions in the Wild
https://openai.com/blog/faulty-reward-functions/
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 8 / 41
Robustness - Adversarial attacks
Robust Physical-World Attacks on
Deep Learning Visual Classifica-
tion
https://arxiv.org/pdf/
1707.08945.pdf
Fooling automated surveillance
cameras: adversarial patches to
attack person detection
https://arxiv.org/pdf/
1904.08653.pdf
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 9 / 41
Subsection 2
Active Learning: From Theory to Practice
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 10 / 41
Active Learning: From Theory to Practice
Slides for the tutorial: http://nowak.ece.wisc.edu/ActiveML.html
Active Learning tries to answer the question:
Can we train machines with less labeled data and less human
supervision?
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 11 / 41
Active Learning
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 12 / 41
Rethinking classical model generalization
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 13 / 41
Subsection 3
A Tutorial on Attention in Deep Learning
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 14 / 41
A Tutorial on Attention in Deep Learning
Slides: http://alex.smola.org/talks/ICML19-attention.pdf
https://www.d2l.ai/
https://github.com/d2l-ai
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 15 / 41
Section 2
Day 2 - U.S. Census, Time Series, Hawkes Processes,
Shapley values, Topological Data Analysis, Optimal
Transport for Graphs
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 16 / 41
Subsection 1
Differential privacy
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 17 / 41
U.S. Census & Differential privacy
Good related read: https://www.sciencemag.org/news/2019/01/
can-set-equations-keep-us-census-data-private
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 18 / 41
Subsection 2
Time Series
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 19 / 41
Deep Factors for Forecasting
https://arxiv.org/pdf/1905.12417.pdf
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 20 / 41
Hawkes Processes
ICML 2018 tutorial: http://learning.mpi-sws.org/tpp-icml18/
http://proceedings.mlr.press/v97/trouleau19a/trouleau19a.pdf
cf. tick for practitioners: https://github.com/X-DataInitiative/tick
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 21 / 41
Subsection 3
Shapley values: Explainability & Data Valuation
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 22 / 41
Shapley values
A new trend in ML based on:
Shapley, Lloyd S. A value for n-person games. Contributions to the
Theory of Games 2.28 (1953). 158 citations.
φi (v) =
S⊆N{i}
|S|!(N − |S| − 1)!
N!
(v(S ∪ {i}) − v(S))
Recent applications:
Explainability: A Unified Approach to Interpreting Model Predictions
http://papers.nips.cc/paper/
7062-a-unified-approach-to-interpreting-model-predictions.
pdf
Data valuation: Data Shapley: Equitable Valuation of Data for
Machine Learning
http://proceedings.mlr.press/v97/ghorbani19c/ghorbani19c.pdf
Since computationally intensive, many papers try to approximate these
values fast...
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 23 / 41
Subsection 4
Topological Data Analysis
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 24 / 41
Topological Data Analysis
Already a tutorial at ICML 2013: Topological Data Analysis and Machine
Learning http://www2.stat.duke.edu/~sayan/Primoz/ICML.pdf
A round-up of TDA papers at ICML 2019:
https://bastian.rieck.me/blog/posts/2019/icml_tda_roundup/
(6 TDA-related papers)
Take-home message: TDA can provide robust features to Machine
Learning models.
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 25 / 41
Subsection 5
Optimal Transport
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 26 / 41
Optimal Transport
Trending in the ML community (at least 5 ICML 2019 papers) since
Cuturi’s 2013 NIPS paper: Sinkhorn Distances: Lightspeed Computation
of Optimal Transport, which made Optimal Transport for Machine
Learning possible in practice. Way too slow before!
Theoretical contributions:
On Efficient Optimal Transport: An Analysis of Greedy and
Accelerated Mirror Descent Algorithms
Methodology:
Optimal Transport for structured data with application on graphs
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 27 / 41
Section 3
Day 3 - Robotics, Gaussian Processes, Learning with
noisy labels
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 28 / 41
Subsection 1
Machine Learning with Application to Robotics
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 29 / 41
Machine Learning with Application to Robotics
Solving complex PDEs and other stochastic control problems in real time
is not really feasible as of today. Machine Learning (supervised learning of
trajectories, learning from demonstration, etc.) can help robotics.
Recorded talk:
https://www.facebook.com/icml.imls/videos/2368059266588651/
Lab: http://lasa.epfl.ch/
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 30 / 41
Subsection 2
Gaussian Processes
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 31 / 41
Gaussian Processes
One flagship project: The Automatic Statistician
https://www.automaticstatistician.com/index/
Discovering Latent Covariance Structures for Multiple Time Series
https://arxiv.org/pdf/1703.09528.pdf
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 32 / 41
Subsection 3
Labels. . .
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 33 / 41
Labels. . .
Learning Dependency Structures for Weak Supervision Models
https://arxiv.org/pdf/1903.05844.pdf
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 34 / 41
Section 4
Day 4 - Interpretability, Natural Language Processing
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 35 / 41
Subsection 1
Interpretability
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 36 / 41
Interpretability
Towards a Deep and Unified Understanding of Deep Neural Models in NLP
http://proceedings.mlr.press/v97/guan19a/guan19a.pdf
Explaining Deep Neural Networks with a Polynomial Time Algorithm for
Shapley Values Approximation
http://proceedings.mlr.press/v97/ancona19a/ancona19a.pdf
https://icml.cc/media/Slides/icml/2019/grandball(13-09-00)
-13-09-25-4776-explaining_deep.pdf
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 37 / 41
Subsection 2
Natural Language Processing
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 38 / 41
Natural Language Processing
MeanSum: A Neural Model for Unsupervised Multi-Document
Abstractive Summarization
https://arxiv.org/pdf/1810.05739.pdf
https://icml.cc/media/Slides/icml/2019/104(13-11-00)
-13-12-10-4891-meansum_a_neur.pdf
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 39 / 41
Section 5
Day 5 - Workshop Time Series
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 40 / 41
GluonTS
GluonTS: Probabilistic Time Series Models in Python
https://arxiv.org/pdf/1906.05264.pdf
https://github.com/awslabs/gluon-ts
Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 41 / 41

More Related Content

Similar to Takeaways from ICML 2019, Long Beach, California

Processing and understanding geo-social media content
Processing and understanding geo-social media contentProcessing and understanding geo-social media content
Processing and understanding geo-social media content
foostermann
 

Similar to Takeaways from ICML 2019, Long Beach, California (17)

Master Class: GANS with Applications in Synthetic Data Generation
Master Class:   GANS with  Applications in  Synthetic Data GenerationMaster Class:   GANS with  Applications in  Synthetic Data Generation
Master Class: GANS with Applications in Synthetic Data Generation
 
20210224_Kirkby_EMinar.pdf
20210224_Kirkby_EMinar.pdf20210224_Kirkby_EMinar.pdf
20210224_Kirkby_EMinar.pdf
 
Technology roadmap highlights_report 2015
Technology roadmap highlights_report 2015Technology roadmap highlights_report 2015
Technology roadmap highlights_report 2015
 
OSDC 2019 | Ignite: Vox Pupuli – Empowering the Puppet Community by Tim Meusel
OSDC 2019 | Ignite: Vox Pupuli – Empowering the Puppet Community by Tim MeuselOSDC 2019 | Ignite: Vox Pupuli – Empowering the Puppet Community by Tim Meusel
OSDC 2019 | Ignite: Vox Pupuli – Empowering the Puppet Community by Tim Meusel
 
GAIA and Alternative Networks
GAIA and Alternative NetworksGAIA and Alternative Networks
GAIA and Alternative Networks
 
Enriching geo-social media through geographic contextualization
Enriching geo-social media through geographic contextualizationEnriching geo-social media through geographic contextualization
Enriching geo-social media through geographic contextualization
 
Enriching geo-social media content @AGILE 2015
Enriching geo-social media content @AGILE 2015Enriching geo-social media content @AGILE 2015
Enriching geo-social media content @AGILE 2015
 
Introduction to LLMs
Introduction to LLMsIntroduction to LLMs
Introduction to LLMs
 
AML & ALMA: Project Overview
AML & ALMA: Project OverviewAML & ALMA: Project Overview
AML & ALMA: Project Overview
 
Streaming Topic Model Training and Inference with Apache Flink
Streaming Topic Model Training and Inference with Apache FlinkStreaming Topic Model Training and Inference with Apache Flink
Streaming Topic Model Training and Inference with Apache Flink
 
Hybrid geo-information processing
Hybrid geo-information processingHybrid geo-information processing
Hybrid geo-information processing
 
EuroPython 2019: GeoSpatial Analysis using Python and JupyterHub
EuroPython 2019: GeoSpatial Analysis using Python and JupyterHubEuroPython 2019: GeoSpatial Analysis using Python and JupyterHub
EuroPython 2019: GeoSpatial Analysis using Python and JupyterHub
 
Tabnet presentation
Tabnet presentationTabnet presentation
Tabnet presentation
 
Protelis: Practical Aggregate Programming - Symposium on Applied Computing (S...
Protelis: Practical Aggregate Programming - Symposium on Applied Computing (S...Protelis: Practical Aggregate Programming - Symposium on Applied Computing (S...
Protelis: Practical Aggregate Programming - Symposium on Applied Computing (S...
 
Processing and understanding geo-social media content
Processing and understanding geo-social media contentProcessing and understanding geo-social media content
Processing and understanding geo-social media content
 
A Discrete Krill Herd Optimization Algorithm for Community Detection
A Discrete Krill Herd Optimization Algorithm for Community DetectionA Discrete Krill Herd Optimization Algorithm for Community Detection
A Discrete Krill Herd Optimization Algorithm for Community Detection
 
TuSoW: Tuple Spaces for Edge Computing
TuSoW: Tuple Spaces for Edge ComputingTuSoW: Tuple Spaces for Edge Computing
TuSoW: Tuple Spaces for Edge Computing
 

More from Gautier Marti

On clustering financial time series - A need for distances between dependent ...
On clustering financial time series - A need for distances between dependent ...On clustering financial time series - A need for distances between dependent ...
On clustering financial time series - A need for distances between dependent ...
Gautier Marti
 

More from Gautier Marti (15)

Using Large Language Models in 10 Lines of Code
Using Large Language Models in 10 Lines of CodeUsing Large Language Models in 10 Lines of Code
Using Large Language Models in 10 Lines of Code
 
What deep learning can bring to...
What deep learning can bring to...What deep learning can bring to...
What deep learning can bring to...
 
A quick demo of Top2Vec With application on 2020 10-K business descriptions
A quick demo of Top2Vec With application on 2020 10-K business descriptionsA quick demo of Top2Vec With application on 2020 10-K business descriptions
A quick demo of Top2Vec With application on 2020 10-K business descriptions
 
Autoregressive Convolutional Neural Networks for Asynchronous Time Series
Autoregressive Convolutional Neural Networks for Asynchronous Time SeriesAutoregressive Convolutional Neural Networks for Asynchronous Time Series
Autoregressive Convolutional Neural Networks for Asynchronous Time Series
 
Some contributions to the clustering of financial time series - Applications ...
Some contributions to the clustering of financial time series - Applications ...Some contributions to the clustering of financial time series - Applications ...
Some contributions to the clustering of financial time series - Applications ...
 
Clustering CDS: algorithms, distances, stability and convergence rates
Clustering CDS: algorithms, distances, stability and convergence ratesClustering CDS: algorithms, distances, stability and convergence rates
Clustering CDS: algorithms, distances, stability and convergence rates
 
Clustering Financial Time Series using their Correlations and their Distribut...
Clustering Financial Time Series using their Correlations and their Distribut...Clustering Financial Time Series using their Correlations and their Distribut...
Clustering Financial Time Series using their Correlations and their Distribut...
 
A closer look at correlations
A closer look at correlationsA closer look at correlations
A closer look at correlations
 
Clustering Financial Time Series: How Long is Enough?
Clustering Financial Time Series: How Long is Enough?Clustering Financial Time Series: How Long is Enough?
Clustering Financial Time Series: How Long is Enough?
 
Optimal Transport vs. Fisher-Rao distance between Copulas
Optimal Transport vs. Fisher-Rao distance between CopulasOptimal Transport vs. Fisher-Rao distance between Copulas
Optimal Transport vs. Fisher-Rao distance between Copulas
 
On Clustering Financial Time Series - Beyond Correlation
On Clustering Financial Time Series - Beyond CorrelationOn Clustering Financial Time Series - Beyond Correlation
On Clustering Financial Time Series - Beyond Correlation
 
Optimal Transport between Copulas for Clustering Time Series
Optimal Transport between Copulas for Clustering Time SeriesOptimal Transport between Copulas for Clustering Time Series
Optimal Transport between Copulas for Clustering Time Series
 
On the stability of clustering financial time series
On the stability of clustering financial time seriesOn the stability of clustering financial time series
On the stability of clustering financial time series
 
Clustering Random Walk Time Series
Clustering Random Walk Time SeriesClustering Random Walk Time Series
Clustering Random Walk Time Series
 
On clustering financial time series - A need for distances between dependent ...
On clustering financial time series - A need for distances between dependent ...On clustering financial time series - A need for distances between dependent ...
On clustering financial time series - A need for distances between dependent ...
 

Recently uploaded

Gartner's Data Analytics Maturity Model.pptx
Gartner's Data Analytics Maturity Model.pptxGartner's Data Analytics Maturity Model.pptx
Gartner's Data Analytics Maturity Model.pptx
chadhar227
 
PLE-statistics document for primary schs
PLE-statistics document for primary schsPLE-statistics document for primary schs
PLE-statistics document for primary schs
cnajjemba
 
Cytotec in Jeddah+966572737505) get unwanted pregnancy kit Riyadh
Cytotec in Jeddah+966572737505) get unwanted pregnancy kit RiyadhCytotec in Jeddah+966572737505) get unwanted pregnancy kit Riyadh
Cytotec in Jeddah+966572737505) get unwanted pregnancy kit Riyadh
Abortion pills in Riyadh +966572737505 get cytotec
 
怎样办理圣地亚哥州立大学毕业证(SDSU毕业证书)成绩单学校原版复制
怎样办理圣地亚哥州立大学毕业证(SDSU毕业证书)成绩单学校原版复制怎样办理圣地亚哥州立大学毕业证(SDSU毕业证书)成绩单学校原版复制
怎样办理圣地亚哥州立大学毕业证(SDSU毕业证书)成绩单学校原版复制
vexqp
 
怎样办理纽约州立大学宾汉姆顿分校毕业证(SUNY-Bin毕业证书)成绩单学校原版复制
怎样办理纽约州立大学宾汉姆顿分校毕业证(SUNY-Bin毕业证书)成绩单学校原版复制怎样办理纽约州立大学宾汉姆顿分校毕业证(SUNY-Bin毕业证书)成绩单学校原版复制
怎样办理纽约州立大学宾汉姆顿分校毕业证(SUNY-Bin毕业证书)成绩单学校原版复制
vexqp
 
+97470301568>>weed for sale in qatar ,weed for sale in dubai,weed for sale in...
+97470301568>>weed for sale in qatar ,weed for sale in dubai,weed for sale in...+97470301568>>weed for sale in qatar ,weed for sale in dubai,weed for sale in...
+97470301568>>weed for sale in qatar ,weed for sale in dubai,weed for sale in...
Health
 
Reconciling Conflicting Data Curation Actions: Transparency Through Argument...
Reconciling Conflicting Data Curation Actions:  Transparency Through Argument...Reconciling Conflicting Data Curation Actions:  Transparency Through Argument...
Reconciling Conflicting Data Curation Actions: Transparency Through Argument...
Bertram Ludäscher
 
Top profile Call Girls In Hapur [ 7014168258 ] Call Me For Genuine Models We ...
Top profile Call Girls In Hapur [ 7014168258 ] Call Me For Genuine Models We ...Top profile Call Girls In Hapur [ 7014168258 ] Call Me For Genuine Models We ...
Top profile Call Girls In Hapur [ 7014168258 ] Call Me For Genuine Models We ...
nirzagarg
 
怎样办理圣路易斯大学毕业证(SLU毕业证书)成绩单学校原版复制
怎样办理圣路易斯大学毕业证(SLU毕业证书)成绩单学校原版复制怎样办理圣路易斯大学毕业证(SLU毕业证书)成绩单学校原版复制
怎样办理圣路易斯大学毕业证(SLU毕业证书)成绩单学校原版复制
vexqp
 
Top profile Call Girls In Satna [ 7014168258 ] Call Me For Genuine Models We ...
Top profile Call Girls In Satna [ 7014168258 ] Call Me For Genuine Models We ...Top profile Call Girls In Satna [ 7014168258 ] Call Me For Genuine Models We ...
Top profile Call Girls In Satna [ 7014168258 ] Call Me For Genuine Models We ...
nirzagarg
 
Top profile Call Girls In bhavnagar [ 7014168258 ] Call Me For Genuine Models...
Top profile Call Girls In bhavnagar [ 7014168258 ] Call Me For Genuine Models...Top profile Call Girls In bhavnagar [ 7014168258 ] Call Me For Genuine Models...
Top profile Call Girls In bhavnagar [ 7014168258 ] Call Me For Genuine Models...
gajnagarg
 
Top profile Call Girls In Tumkur [ 7014168258 ] Call Me For Genuine Models We...
Top profile Call Girls In Tumkur [ 7014168258 ] Call Me For Genuine Models We...Top profile Call Girls In Tumkur [ 7014168258 ] Call Me For Genuine Models We...
Top profile Call Girls In Tumkur [ 7014168258 ] Call Me For Genuine Models We...
nirzagarg
 
一比一原版(UCD毕业证书)加州大学戴维斯分校毕业证成绩单原件一模一样
一比一原版(UCD毕业证书)加州大学戴维斯分校毕业证成绩单原件一模一样一比一原版(UCD毕业证书)加州大学戴维斯分校毕业证成绩单原件一模一样
一比一原版(UCD毕业证书)加州大学戴维斯分校毕业证成绩单原件一模一样
wsppdmt
 

Recently uploaded (20)

Gartner's Data Analytics Maturity Model.pptx
Gartner's Data Analytics Maturity Model.pptxGartner's Data Analytics Maturity Model.pptx
Gartner's Data Analytics Maturity Model.pptx
 
PLE-statistics document for primary schs
PLE-statistics document for primary schsPLE-statistics document for primary schs
PLE-statistics document for primary schs
 
Cytotec in Jeddah+966572737505) get unwanted pregnancy kit Riyadh
Cytotec in Jeddah+966572737505) get unwanted pregnancy kit RiyadhCytotec in Jeddah+966572737505) get unwanted pregnancy kit Riyadh
Cytotec in Jeddah+966572737505) get unwanted pregnancy kit Riyadh
 
怎样办理圣地亚哥州立大学毕业证(SDSU毕业证书)成绩单学校原版复制
怎样办理圣地亚哥州立大学毕业证(SDSU毕业证书)成绩单学校原版复制怎样办理圣地亚哥州立大学毕业证(SDSU毕业证书)成绩单学校原版复制
怎样办理圣地亚哥州立大学毕业证(SDSU毕业证书)成绩单学校原版复制
 
怎样办理纽约州立大学宾汉姆顿分校毕业证(SUNY-Bin毕业证书)成绩单学校原版复制
怎样办理纽约州立大学宾汉姆顿分校毕业证(SUNY-Bin毕业证书)成绩单学校原版复制怎样办理纽约州立大学宾汉姆顿分校毕业证(SUNY-Bin毕业证书)成绩单学校原版复制
怎样办理纽约州立大学宾汉姆顿分校毕业证(SUNY-Bin毕业证书)成绩单学校原版复制
 
+97470301568>>weed for sale in qatar ,weed for sale in dubai,weed for sale in...
+97470301568>>weed for sale in qatar ,weed for sale in dubai,weed for sale in...+97470301568>>weed for sale in qatar ,weed for sale in dubai,weed for sale in...
+97470301568>>weed for sale in qatar ,weed for sale in dubai,weed for sale in...
 
Predicting HDB Resale Prices - Conducting Linear Regression Analysis With Orange
Predicting HDB Resale Prices - Conducting Linear Regression Analysis With OrangePredicting HDB Resale Prices - Conducting Linear Regression Analysis With Orange
Predicting HDB Resale Prices - Conducting Linear Regression Analysis With Orange
 
SR-101-01012024-EN.docx Federal Constitution of the Swiss Confederation
SR-101-01012024-EN.docx  Federal Constitution  of the Swiss ConfederationSR-101-01012024-EN.docx  Federal Constitution  of the Swiss Confederation
SR-101-01012024-EN.docx Federal Constitution of the Swiss Confederation
 
Reconciling Conflicting Data Curation Actions: Transparency Through Argument...
Reconciling Conflicting Data Curation Actions:  Transparency Through Argument...Reconciling Conflicting Data Curation Actions:  Transparency Through Argument...
Reconciling Conflicting Data Curation Actions: Transparency Through Argument...
 
Top profile Call Girls In Hapur [ 7014168258 ] Call Me For Genuine Models We ...
Top profile Call Girls In Hapur [ 7014168258 ] Call Me For Genuine Models We ...Top profile Call Girls In Hapur [ 7014168258 ] Call Me For Genuine Models We ...
Top profile Call Girls In Hapur [ 7014168258 ] Call Me For Genuine Models We ...
 
Discover Why Less is More in B2B Research
Discover Why Less is More in B2B ResearchDiscover Why Less is More in B2B Research
Discover Why Less is More in B2B Research
 
Switzerland Constitution 2002.pdf.........
Switzerland Constitution 2002.pdf.........Switzerland Constitution 2002.pdf.........
Switzerland Constitution 2002.pdf.........
 
Ranking and Scoring Exercises for Research
Ranking and Scoring Exercises for ResearchRanking and Scoring Exercises for Research
Ranking and Scoring Exercises for Research
 
Aspirational Block Program Block Syaldey District - Almora
Aspirational Block Program Block Syaldey District - AlmoraAspirational Block Program Block Syaldey District - Almora
Aspirational Block Program Block Syaldey District - Almora
 
怎样办理圣路易斯大学毕业证(SLU毕业证书)成绩单学校原版复制
怎样办理圣路易斯大学毕业证(SLU毕业证书)成绩单学校原版复制怎样办理圣路易斯大学毕业证(SLU毕业证书)成绩单学校原版复制
怎样办理圣路易斯大学毕业证(SLU毕业证书)成绩单学校原版复制
 
Top profile Call Girls In Satna [ 7014168258 ] Call Me For Genuine Models We ...
Top profile Call Girls In Satna [ 7014168258 ] Call Me For Genuine Models We ...Top profile Call Girls In Satna [ 7014168258 ] Call Me For Genuine Models We ...
Top profile Call Girls In Satna [ 7014168258 ] Call Me For Genuine Models We ...
 
Top profile Call Girls In bhavnagar [ 7014168258 ] Call Me For Genuine Models...
Top profile Call Girls In bhavnagar [ 7014168258 ] Call Me For Genuine Models...Top profile Call Girls In bhavnagar [ 7014168258 ] Call Me For Genuine Models...
Top profile Call Girls In bhavnagar [ 7014168258 ] Call Me For Genuine Models...
 
Data Analyst Tasks to do the internship.pdf
Data Analyst Tasks to do the internship.pdfData Analyst Tasks to do the internship.pdf
Data Analyst Tasks to do the internship.pdf
 
Top profile Call Girls In Tumkur [ 7014168258 ] Call Me For Genuine Models We...
Top profile Call Girls In Tumkur [ 7014168258 ] Call Me For Genuine Models We...Top profile Call Girls In Tumkur [ 7014168258 ] Call Me For Genuine Models We...
Top profile Call Girls In Tumkur [ 7014168258 ] Call Me For Genuine Models We...
 
一比一原版(UCD毕业证书)加州大学戴维斯分校毕业证成绩单原件一模一样
一比一原版(UCD毕业证书)加州大学戴维斯分校毕业证成绩单原件一模一样一比一原版(UCD毕业证书)加州大学戴维斯分校毕业证成绩单原件一模一样
一比一原版(UCD毕业证书)加州大学戴维斯分校毕业证成绩单原件一模一样
 

Takeaways from ICML 2019, Long Beach, California

  • 1. Takeaways from ICML 2019 Hong Kong Machine Learning Meetup Season 1 Episode 12 – Season Finale Gautier Marti Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 1 / 41
  • 2. Table of contents 1 Day 1 - Tutorials 2 Day 2 - U.S. Census, Time Series, Hawkes Processes, Shapley values, Topological Data Analysis, Optimal Transport for Graphs 3 Day 3 - Robotics, Gaussian Processes, Learning with noisy labels 4 Day 4 - Interpretability, Natural Language Processing 5 Day 5 - Workshop Time Series Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 2 / 41
  • 3. Disclaimer I was sent to the ICML 2019 conference by my employer Shell Street Labs. However, the opinions expressed in this presentation are my own and do not reflect in any ways the view of my employer. Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 3 / 41
  • 4. Section 1 Day 1 - Tutorials Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 4 / 41
  • 5. Subsection 1 Safe Machine Learning Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 5 / 41
  • 6. Safe Machine Learning Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 6 / 41
  • 7. COMPAS - An example of ML biases & misspecification We show, however, that the widely used commercial risk assessment software COMPAS is no more accurate or fair than predictions made by people with little or no criminal justice expertise. The accuracy, fairness, and limits of predicting recidivism https://advances.sciencemag.org/content/4/1/eaao5580 Our analysis of Northpointe’s tool, called COMPAS (which stands for Correctional Offender Management Profiling for Alternative Sanctions), found that black defendants were far more likely than white defendants to be incorrectly judged to be at a higher risk of recidivism, while white defendants were more likely than black defendants to be incorrectly flagged as low risk. How We Analyzed the COMPAS Recidivism Algorithm https://www.propublica.org/article/ how-we-analyzed-the-compas-recidivism-algorithm GitHub: https://github.com/propublica/compas-analysis Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 7 / 41
  • 8. RL in the Wild - Other examples of misspecifications Reinforcement learning algorithms can break in surprising, counterintuitive ways. Faulty Reward Functions in the Wild https://openai.com/blog/faulty-reward-functions/ Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 8 / 41
  • 9. Robustness - Adversarial attacks Robust Physical-World Attacks on Deep Learning Visual Classifica- tion https://arxiv.org/pdf/ 1707.08945.pdf Fooling automated surveillance cameras: adversarial patches to attack person detection https://arxiv.org/pdf/ 1904.08653.pdf Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 9 / 41
  • 10. Subsection 2 Active Learning: From Theory to Practice Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 10 / 41
  • 11. Active Learning: From Theory to Practice Slides for the tutorial: http://nowak.ece.wisc.edu/ActiveML.html Active Learning tries to answer the question: Can we train machines with less labeled data and less human supervision? Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 11 / 41
  • 12. Active Learning Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 12 / 41
  • 13. Rethinking classical model generalization Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 13 / 41
  • 14. Subsection 3 A Tutorial on Attention in Deep Learning Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 14 / 41
  • 15. A Tutorial on Attention in Deep Learning Slides: http://alex.smola.org/talks/ICML19-attention.pdf https://www.d2l.ai/ https://github.com/d2l-ai Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 15 / 41
  • 16. Section 2 Day 2 - U.S. Census, Time Series, Hawkes Processes, Shapley values, Topological Data Analysis, Optimal Transport for Graphs Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 16 / 41
  • 17. Subsection 1 Differential privacy Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 17 / 41
  • 18. U.S. Census & Differential privacy Good related read: https://www.sciencemag.org/news/2019/01/ can-set-equations-keep-us-census-data-private Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 18 / 41
  • 19. Subsection 2 Time Series Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 19 / 41
  • 20. Deep Factors for Forecasting https://arxiv.org/pdf/1905.12417.pdf Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 20 / 41
  • 21. Hawkes Processes ICML 2018 tutorial: http://learning.mpi-sws.org/tpp-icml18/ http://proceedings.mlr.press/v97/trouleau19a/trouleau19a.pdf cf. tick for practitioners: https://github.com/X-DataInitiative/tick Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 21 / 41
  • 22. Subsection 3 Shapley values: Explainability & Data Valuation Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 22 / 41
  • 23. Shapley values A new trend in ML based on: Shapley, Lloyd S. A value for n-person games. Contributions to the Theory of Games 2.28 (1953). 158 citations. φi (v) = S⊆N{i} |S|!(N − |S| − 1)! N! (v(S ∪ {i}) − v(S)) Recent applications: Explainability: A Unified Approach to Interpreting Model Predictions http://papers.nips.cc/paper/ 7062-a-unified-approach-to-interpreting-model-predictions. pdf Data valuation: Data Shapley: Equitable Valuation of Data for Machine Learning http://proceedings.mlr.press/v97/ghorbani19c/ghorbani19c.pdf Since computationally intensive, many papers try to approximate these values fast... Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 23 / 41
  • 24. Subsection 4 Topological Data Analysis Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 24 / 41
  • 25. Topological Data Analysis Already a tutorial at ICML 2013: Topological Data Analysis and Machine Learning http://www2.stat.duke.edu/~sayan/Primoz/ICML.pdf A round-up of TDA papers at ICML 2019: https://bastian.rieck.me/blog/posts/2019/icml_tda_roundup/ (6 TDA-related papers) Take-home message: TDA can provide robust features to Machine Learning models. Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 25 / 41
  • 26. Subsection 5 Optimal Transport Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 26 / 41
  • 27. Optimal Transport Trending in the ML community (at least 5 ICML 2019 papers) since Cuturi’s 2013 NIPS paper: Sinkhorn Distances: Lightspeed Computation of Optimal Transport, which made Optimal Transport for Machine Learning possible in practice. Way too slow before! Theoretical contributions: On Efficient Optimal Transport: An Analysis of Greedy and Accelerated Mirror Descent Algorithms Methodology: Optimal Transport for structured data with application on graphs Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 27 / 41
  • 28. Section 3 Day 3 - Robotics, Gaussian Processes, Learning with noisy labels Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 28 / 41
  • 29. Subsection 1 Machine Learning with Application to Robotics Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 29 / 41
  • 30. Machine Learning with Application to Robotics Solving complex PDEs and other stochastic control problems in real time is not really feasible as of today. Machine Learning (supervised learning of trajectories, learning from demonstration, etc.) can help robotics. Recorded talk: https://www.facebook.com/icml.imls/videos/2368059266588651/ Lab: http://lasa.epfl.ch/ Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 30 / 41
  • 31. Subsection 2 Gaussian Processes Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 31 / 41
  • 32. Gaussian Processes One flagship project: The Automatic Statistician https://www.automaticstatistician.com/index/ Discovering Latent Covariance Structures for Multiple Time Series https://arxiv.org/pdf/1703.09528.pdf Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 32 / 41
  • 33. Subsection 3 Labels. . . Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 33 / 41
  • 34. Labels. . . Learning Dependency Structures for Weak Supervision Models https://arxiv.org/pdf/1903.05844.pdf Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 34 / 41
  • 35. Section 4 Day 4 - Interpretability, Natural Language Processing Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 35 / 41
  • 36. Subsection 1 Interpretability Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 36 / 41
  • 37. Interpretability Towards a Deep and Unified Understanding of Deep Neural Models in NLP http://proceedings.mlr.press/v97/guan19a/guan19a.pdf Explaining Deep Neural Networks with a Polynomial Time Algorithm for Shapley Values Approximation http://proceedings.mlr.press/v97/ancona19a/ancona19a.pdf https://icml.cc/media/Slides/icml/2019/grandball(13-09-00) -13-09-25-4776-explaining_deep.pdf Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 37 / 41
  • 38. Subsection 2 Natural Language Processing Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 38 / 41
  • 39. Natural Language Processing MeanSum: A Neural Model for Unsupervised Multi-Document Abstractive Summarization https://arxiv.org/pdf/1810.05739.pdf https://icml.cc/media/Slides/icml/2019/104(13-11-00) -13-12-10-4891-meansum_a_neur.pdf Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 39 / 41
  • 40. Section 5 Day 5 - Workshop Time Series Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 40 / 41
  • 41. GluonTS GluonTS: Probabilistic Time Series Models in Python https://arxiv.org/pdf/1906.05264.pdf https://github.com/awslabs/gluon-ts Gautier Marti (Shell Street Labs) Takeaways from ICML 2019 41 / 41