Friends don t let friends deploy Black-Box models The importance of transparency in Machine Learning. Rich Caruana Microsoft Research
|
|
|
- Tyrone McDonald
- 5 years ago
- Views:
Transcription
1
2 Friends don t let friends deploy Black-Box models The importance of transparency in Machine Learning Rich Caruana Microsoft Research
3 Friends Don t Let Friends Deploy Black-Box Models The Importance of Transparency in Machine Learning Rich Caruana Microsoft Research Joint Work with Ran Gilad-Bachrach, Yin Lou, Sarah Tan, Johannes Gehrke Paul Koch, Marc Sturm, Noemie Elhadad Thanks to Greg Cooper MD PhD, Mike Fine MD MPH, Eric Horvitz MD PhD Nick Craswell, Tom Mitchell, Jacob Bien, Giles Hooker, Noah Snavely Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
4 When is it Safe to Use Machine Learning? data for 1M patients 1000 s great clinical features train state-of-the-art machine learning model on data accuracy looks great on test set: AUC = 0.95 is it safe to deploy this model and use on real patients? is high accuracy on test data enough to trust a model? Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
5 When is it Safe to Use Machine Learning? data for 1M patients 1000 s great clinical features train state-of-the-art machine learning model on data accuracy looks great on test set: AUC = 0.95 is it safe to deploy this model and use on real patients? is high accuracy on test data enough to trust a model? Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
6 When is it Safe to Use Machine Learning? data for 1M patients 1000 s great clinical features train state-of-the-art machine learning model on data accuracy looks great on test set: AUC = 0.95 is it safe to deploy this model and use on real patients? is high accuracy on test data enough to trust a model? Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
7 Motivation: Predicting Pneumonia Risk Study (mid-90 s) LOW Risk: outpatient: antibiotics, call if not feeling better HIGH Risk: admit to hospital ( 10% of pneumonia patients die) One goal was to compare various ML methods: logistic regression rule-based learning k-nearest neighbor neural nets Bayesian methods hierarchical mixtures of experts... Most accurate ML method: multitask neural nets (shallow MTL nets) Safe to use neural nets on patients? No we used logistic regression instead... Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
8 Motivation: Predicting Pneumonia Risk Study (mid-90 s) LOW Risk: outpatient: antibiotics, call if not feeling better HIGH Risk: admit to hospital ( 10% of pneumonia patients die) One goal was to compare various ML methods: logistic regression rule-based learning k-nearest neighbor neural nets Bayesian methods hierarchical mixtures of experts... Most accurate ML method: multitask neural nets (shallow MTL nets) Safe to use neural nets on patients? No we used logistic regression instead... Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
9 Motivation: Predicting Pneumonia Risk Study (mid-90 s) LOW Risk: outpatient: antibiotics, call if not feeling better HIGH Risk: admit to hospital ( 10% of pneumonia patients die) One goal was to compare various ML methods: logistic regression rule-based learning k-nearest neighbor neural nets Bayesian methods hierarchical mixtures of experts... Most accurate ML method: multitask neural nets (shallow MTL nets) Safe to use neural nets on patients? No we used logistic regression instead... Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
10 Motivation: Predicting Pneumonia Risk Study (mid-90 s) RBL learned rule: HasAsthma(x) => LessRisk(x) True pattern in data: asthmatics presenting with pneumonia considered very high risk receive agressive treatment and often admitted to ICU history of asthma also means they often go to healthcare sooner treatment lowers risk of death compared to general population If RBL learned asthma is good for you, NN probably did, too if we use NN for admission decision, could hurt asthmatics Key to discovering HasAsthma(x)... was intelligibility of rules even if we can remove asthma problem from neural net, what other bad patterns don t we know about that RBL missed? Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
11 Motivation: Predicting Pneumonia Risk Study (mid-90 s) RBL learned rule: HasAsthma(x) => LessRisk(x) True pattern in data: asthmatics presenting with pneumonia considered very high risk receive agressive treatment and often admitted to ICU history of asthma also means they often go to healthcare sooner treatment lowers risk of death compared to general population If RBL learned asthma is good for you, NN probably did, too if we use NN for admission decision, could hurt asthmatics Key to discovering HasAsthma(x)... was intelligibility of rules even if we can remove asthma problem from neural net, what other bad patterns don t we know about that RBL missed? Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
12 Motivation: Predicting Pneumonia Risk Study (mid-90 s) RBL learned rule: HasAsthma(x) => LessRisk(x) True pattern in data: asthmatics presenting with pneumonia considered very high risk receive agressive treatment and often admitted to ICU history of asthma also means they often go to healthcare sooner treatment lowers risk of death compared to general population If RBL learned asthma is good for you, NN probably did, too if we use NN for admission decision, could hurt asthmatics Key to discovering HasAsthma(x)... was intelligibility of rules even if we can remove asthma problem from neural net, what other bad patterns don t we know about that RBL missed? Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
13 Motivation: Predicting Pneumonia Risk Study (mid-90 s) RBL learned rule: HasAsthma(x) => LessRisk(x) True pattern in data: asthmatics presenting with pneumonia considered very high risk receive agressive treatment and often admitted to ICU history of asthma also means they often go to healthcare sooner treatment lowers risk of death compared to general population If RBL learned asthma is good for you, NN probably did, too if we use NN for admission decision, could hurt asthmatics Key to discovering HasAsthma(x)... was intelligibility of rules even if we can remove asthma problem from neural net, what other bad patterns don t we know about that RBL missed? Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
14 Lessons Risky to use data for purposes it was not designed for Most data has unexpected landmines Not ethical to collect correct data for asthma Much too difficult to fully understand the data Our approach is to make the learned models as intelligible as possible Must be able to understand models used in healthcare Also true for race and gender bias where the bias is in the training data Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
15 All we need is an accurate, intelligible model Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
16 Accuracy Problem: The Accuracy vs. Intelligibility Tradeoff Boosted Trees Random Forests Neural Nets Single Decision Tree Intelligibility Logistic Regression Naive Bayes Decision Lists Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
17 Accuracy Problem: The Accuracy vs. Intelligibility Tradeoff Boosted Trees Random Forests Neural Nets??? Single Decision Tree Logistic Regression Naive Bayes Decision Lists Intelligibility Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
18 Model Space from Simple to Complex Linear Model: y = β 0 + β 1 x β n x n Additive Model: y = f 1 (x 1 ) f n (x n ) Additive Model with Interactions: y = i f i(x i ) + ij f ij(x i, x j ) + ijk f ijk(x i, x j, x k ) +... Full Complexity Model: y = f (x 1,..., x n ) Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
19 Model Space from Simple to Complex Linear Model: y = β 0 + β 1 x β n x n Additive Model: y = f 1 (x 1 ) f n (x n ) Additive Model with Interactions: y = i f i(x i ) + ij f ij(x i, x j ) + ijk f ijk(x i, x j, x k ) +... Full Complexity Model: y = f (x 1,..., x n ) Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
20 Model Space from Simple to Complex Linear Model: y = β 0 + β 1 x β n x n Additive Model: y = f 1 (x 1 ) f n (x n ) Additive Model with Interactions: y = i f i(x i ) + ij f ij(x i, x j ) + ijk f ijk(x i, x j, x k ) +... Full Complexity Model: y = f (x 1,..., x n ) Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
21 Model Space from Simple to Complex Linear Model: y = β 0 + β 1 x β n x n Additive Model: y = f 1 (x 1 ) f n (x n ) Additive Model with Interactions: y = i f i(x i ) + ij f ij(x i, x j ) + ijk f ijk(x i, x j, x k ) +... Full Complexity Model: y = f (x 1,..., x n ) Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
22 Add ML-Steroids to old Stats Method: GAMs GA2Ms Generalized Additive Models (GAMs) Developed at Stanford by Hastie and Tibshirani in late 80 s Regression: y = f 1 (x 1 ) f n (x n ) Classification: logit(y) = f 1 (x 1 ) f n (x n ) Each feature is shaped by shape function f i T. Hastie and R. Tibshirani. Generalized additive models. Chapman & Hall/CRC, Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
23 Skip all algorithmic details and jump to one result Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
24 What GA2Ms Learn About Pneumonia Risk (POD) as a Function of Age Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
25 Intelligible model also learned: Has Asthma => lower risk History of chest pain => lower risk History of heart disease => lower risk Good we didn t deploy neural net back in 1995 But can understand, edit and safely deploy intelligible GA2M model Intelligible/transparent model is like having a magic pair of glasses Model correctness depends on how model will be used this is a good model for health insurance providers but needs to be repaired to use for hospital admissions Important: Must keep potentially offending features in model! Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
26 Intelligible model also learned: Has Asthma => lower risk History of chest pain => lower risk History of heart disease => lower risk Good we didn t deploy neural net back in 1995 But can understand, edit and safely deploy intelligible GA2M model Intelligible/transparent model is like having a magic pair of glasses Model correctness depends on how model will be used this is a good model for health insurance providers but needs to be repaired to use for hospital admissions Important: Must keep potentially offending features in model! Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
27 Intelligible model also learned: Has Asthma => lower risk History of chest pain => lower risk History of heart disease => lower risk Good we didn t deploy neural net back in 1995 But can understand, edit and safely deploy intelligible GA2M model Intelligible/transparent model is like having a magic pair of glasses Model correctness depends on how model will be used this is a good model for health insurance providers but needs to be repaired to use for hospital admissions Important: Must keep potentially offending features in model! Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
28 Intelligible model also learned: Has Asthma => lower risk History of chest pain => lower risk History of heart disease => lower risk Good we didn t deploy neural net back in 1995 But can understand, edit and safely deploy intelligible GA2M model Intelligible/transparent model is like having a magic pair of glasses Model correctness depends on how model will be used this is a good model for health insurance providers but needs to be repaired to use for hospital admissions Important: Must keep potentially offending features in model! Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
29 Transparent GAM Models for Private AI Interpretable GAM model class is a good match for homomorphic encryption Interpretable models may help preserve data privacy Potential issue with transparency vs. encryption Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
30 Transparent GAM Models for Private AI Interpretable GAM model class is a good match for homomorphic encryption Interpretable models may help preserve data privacy Potential issue with transparency vs. encryption Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
31 Why GAMs Are Good For Homomorphic Encryption Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
32 Why GAMs Are Good For Homomorphic Encryption Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
33 Why GAMs Are Good For Homomorphic Encryption Original 2nd-degree Polynomial Fit Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
34 Why GAMs Are Good For Homomorphic Encryption Poly-GAMs are competitive models Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
35 Transparent GAM Models for Private AI Interpretable GAM model class is a good match for homomorphic encryption Interpretable models may help preserve data privacy Potential issue with transparency vs. encryption Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
36 Why the Simplicity of GAM Models Might Be Good For Preserving Privacy Complex Black-Box Deep Net Transparent GAM Model Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
37 Why the Simplicity of GAM Models Might Be Good For Preserving Privacy Complex Black-Box Deep Net Transparent GAM Model Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
38 Transparent GAM Models for Private AI Interpretable GAM model class is a good match for homomorphic encryption Interpretable models may help preserve data privacy Potential issue with transparency vs. encryption Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
39 Potential Problem with Encryption if Model Remains Hidden Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
40 Potential Problem with Encryption if Model Remains Hidden Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
41 Transparency for Fairness and Bias Detection/Elimination (FAT/ML) ML trained on data will learn the biases in that data ML for resume processing will learn gender bias ML for recidivism prediction will learn race bias... Remember, the bias is in the data! How to deal with bias using intelligible models: keep bias features in data when model is trained remove what was learned from bias features after training If offending variables are eliminated prior to training often can t tell you have a problem makes it harder to correct the problem EU General Data Protection Regulation (goes into effect 2018): Article 9 makes it more difficult to use personal data revealing racial or ethnic origin and other special categories Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
42 Transparency for Fairness and Bias Detection/Elimination (FAT/ML) ML trained on data will learn the biases in that data ML for resume processing will learn gender bias ML for recidivism prediction will learn race bias... Remember, the bias is in the data! How to deal with bias using intelligible models: keep bias features in data when model is trained remove what was learned from bias features after training If offending variables are eliminated prior to training often can t tell you have a problem makes it harder to correct the problem EU General Data Protection Regulation (goes into effect 2018): Article 9 makes it more difficult to use personal data revealing racial or ethnic origin and other special categories Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
43 Transparency for Fairness and Bias Detection/Elimination (FAT/ML) ML trained on data will learn the biases in that data ML for resume processing will learn gender bias ML for recidivism prediction will learn race bias... Remember, the bias is in the data! How to deal with bias using intelligible models: keep bias features in data when model is trained remove what was learned from bias features after training If offending variables are eliminated prior to training often can t tell you have a problem makes it harder to correct the problem EU General Data Protection Regulation (goes into effect 2018): Article 9 makes it more difficult to use personal data revealing racial or ethnic origin and other special categories Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
44 Transparency for Fairness and Bias Detection/Elimination (FAT/ML) ML trained on data will learn the biases in that data ML for resume processing will learn gender bias ML for recidivism prediction will learn race bias... Remember, the bias is in the data! How to deal with bias using intelligible models: keep bias features in data when model is trained remove what was learned from bias features after training If offending variables are eliminated prior to training often can t tell you have a problem makes it harder to correct the problem EU General Data Protection Regulation (goes into effect 2018): Article 9 makes it more difficult to use personal data revealing racial or ethnic origin and other special categories Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
45 Transparency for Fairness and Bias Detection/Elimination (FAT/ML) ML trained on data will learn the biases in that data ML for resume processing will learn gender bias ML for recidivism prediction will learn race bias... Remember, the bias is in the data! How to deal with bias using intelligible models: keep bias features in data when model is trained remove what was learned from bias features after training If offending variables are eliminated prior to training often can t tell you have a problem makes it harder to correct the problem EU General Data Protection Regulation (goes into effect 2018): Article 9 makes it more difficult to use personal data revealing racial or ethnic origin and other special categories Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
46 Summary High accuracy on test set is not always enough can be very misleading There are land mines hidden in most real data need magic glasses to see landmines In some domains (e.g., healthcare) it s critical to understand model before deploying it Correctness depends on how model will be used data/model not inherently right/wrong GA2Ms give us accuracy and intelligibility at same time Important to keep potentially offending variables in model so bias can be detected and then removed after training Deep Learning is great but sometimes we have to understand what s in the black box GA2Ms can help insure privacy protection because models are so simple Poly-GAMs can be good for encryption, but the model needs to be visible to someone Rich Caruana (Microsoft Research) Faculty Summit: Intelligible Models July 18, / 28
47
48
15: Ethics in Machine Learning, plus Artificial General Intelligence and some old Science Fiction
15: Ethics in Machine Learning, plus Artificial General Intelligence and some old Science Fiction Machine Learning and Real-world Data Ann Copestake and Simone Teufel Computer Laboratory University of
An Introduction to Machine Learning for Social Scientists
An Introduction to Machine Learning for Social Scientists Tyler Ransom University of Oklahoma, Dept. of Economics November 10, 2017 Outline 1. Intro 2. Examples 3. Conclusion Tyler Ransom (OU Econ) An
How Machine Learning and AI Are Disrupting the Current Healthcare System. Session #30, March 6, 2018 Cris Ross, CIO Mayo Clinic, Jim Golden, PwC
How Machine Learning and AI Are Disrupting the Current Healthcare System Session #30, March 6, 2018 Cris Ross, CIO Mayo Clinic, Jim Golden, PwC 1 Conflicts of Interest: Christopher Ross, MBA Has no real
Human-Centric Trusted AI for Data-Driven Economy
Human-Centric Trusted AI for Data-Driven Economy Masugi Inoue 1 and Hideyuki Tokuda 2 National Institute of Information and Communications Technology inoue@nict.go.jp 1, Director, International Research
Ethics of Data Science
Ethics of Data Science Lawrence Hunter, Ph.D. Director, Computational Bioscience Program University of Colorado School of Medicine Larry.Hunter@ucdenver.edu http://compbio.ucdenver.edu/hunter Data Science
Contents. List of Figures List of Tables. Structure of the Book How to Use this Book Online Resources Acknowledgements
Contents List of Figures List of Tables Preface Notation Structure of the Book How to Use this Book Online Resources Acknowledgements Notational Conventions Notational Conventions for Probabilities xiii
The Automatic Classification Problem. Perceptrons, SVMs, and Friends: Some Discriminative Models for Classification
Perceptrons, SVMs, and Friends: Some Discriminative Models for Classification Parallel to AIMA 8., 8., 8.6.3, 8.9 The Automatic Classification Problem Assign object/event or sequence of objects/events
SSB Debate: Model-based Inference vs. Machine Learning
SSB Debate: Model-based nference vs. Machine Learning June 3, 2018 SSB 2018 June 3, 2018 1 / 20 Machine learning in the biological sciences SSB 2018 June 3, 2018 2 / 20 Machine learning in the biological
Session 124TS, A Practical Guide to Machine Learning for Actuaries. Presenters: Dave M. Liner, FSA, MAAA, CERA
Session 124TS, A Practical Guide to Machine Learning for Actuaries Presenters: Dave M. Liner, FSA, MAAA, CERA SOA Antitrust Disclaimer SOA Presentation Disclaimer A practical guide to machine learning
OECD WORK ON ARTIFICIAL INTELLIGENCE
OECD Global Parliamentary Network October 10, 2018 OECD WORK ON ARTIFICIAL INTELLIGENCE Karine Perset, Nobu Nishigata, Directorate for Science, Technology and Innovation ai@oecd.org http://oe.cd/ai OECD
AI Fairness 360. Kush R. Varshney
IBM Research AI AI Fairness 360 Kush R. Varshney krvarshn@us.ibm.com http://krvarshney.github.io @krvarshney http://aif360.mybluemix.net https://github.com/ibm/aif360 https://pypi.org/project/aif360 2018
Our position. ICDPPC declaration on ethics and data protection in artificial intelligence
ICDPPC declaration on ethics and data protection in artificial intelligence AmCham EU speaks for American companies committed to Europe on trade, investment and competitiveness issues. It aims to ensure
On Feature Selection, Bias-Variance, and Bagging
On Feature Selection, Bias-Variance, and Bagging Art Munson 1 Rich Caruana 2 1 Department of Computer Science Cornell University 2 Microsoft Corporation ECML-PKDD 2009 Munson; Caruana (Cornell; Microsoft)
Roadmap for machine learning
Roadmap f machine learning Description and state of the art Definition Machine learning is a term that refers to a set of technologies that evolved from the study of pattern recognition and computational
Limitations of Data Mining in Healthcare.
Limitations of Data Mining in Healthcare. Vitaly Herasevich, MD, PhD, FCCM Associate Professor of Anesthesiology and Medicine, Department of Anesthesiology and Perioperative Medicine, Division of Critical
Artificial Intelligence Machine learning and Deep Learning: Trends and Tools. Dr. Shaona
Artificial Intelligence Machine learning and Deep Learning: Trends and Tools Dr. Shaona Ghosh @shaonaghosh What is Machine Learning? Computer algorithms that learn patterns in data automatically from large
FROM AI TO IA AI: Artificial Intelligence IA: Intelligence Amplification Mieke De Ketelaere, SAS NEMEA
FROM AI TO IA AI: Artificial Intelligence IA: Intelligence Amplification Mieke De Ketelaere, AI/CI @ SAS NEMEA About myself G.M. De Ketelaere University of Stuttgart, DE G.M. De Ketelaere and H.W. Guesgen
& Medical Tourism. DIHTF - Dubai 20 th -21 st Feb 2018 V S Venkatesh -India
& Medical Tourism DIHTF - Dubai 20 th -21 st Feb 2018 V S Venkatesh -India The human brain is an amazing work of art, it has very complex neural circuits and the way it registers, stores, processes and
The Use of Commercial Databases for National Security: Privacy, Evaluation, and Accuracy
The Use of Commercial Databases for National Security: Privacy, Evaluation, and Accuracy Rebecca Wright Computer Science Department Stevens Institute of Technology www.cs.stevens.edu/~rwright National
Statistical Tests: More Complicated Discriminants
03/07/07 PHY310: Statistical Data Analysis 1 PHY310: Lecture 14 Statistical Tests: More Complicated Discriminants Road Map When the likelihood discriminant will fail The Multi Layer Perceptron discriminant
Disclosure: Within the past 12 months, I have had no financial relationships with proprietary entities that produce health care goods and services.
Disclosure: Within the past 12 months, I have had no financial relationships with proprietary entities that produce health care goods and services. Artificial Intelligence (AI): definition John McCarthy,
Application of AI Technology to Industrial Revolution
Application of AI Technology to Industrial Revolution By Dr. Suchai Thanawastien 1. What is AI? Artificial Intelligence or AI is a branch of computer science that tries to emulate the capabilities of learning,
Transparency and Accountability of Algorithmic Systems vs. GDPR?
Transparency and Accountability of Algorithmic Systems vs. GDPR? Nozha Boujemaa Directrice de L Institut DATAIA Directrice de Recherche Inria nozha.boujemaa@inria.fr March 2018 Data & Algorithms «2 sides
Predictive Subsea Integrity Management: Effective Tools and Techniques
Predictive Subsea Integrity Management: Effective Tools and Techniques The Leading Edge of Value-Based Subsea Inspection 1 st November Aberdeen 2017 www.astrimar.com Background Low oil price having major
A Review of Related Work on Machine Learning in Semiconductor Manufacturing and Assembly Lines
A Review of Related Work on Machine Learning in Semiconductor Manufacturing and Assembly Lines DI Darko Stanisavljevic VIRTUAL VEHICLE DI Michael Spitzer VIRTUAL VEHICLE i-know 16 18.-19.10.2016, Graz
Why AI Goes Wrong And How To Avoid It Brandon Purcell
Why AI Goes Wrong And How To Avoid It Brandon Purcell June 18, 2018 2018 FORRESTER. REPRODUCTION PROHIBITED. We probably don t need to worry about this in the near future Source: https://twitter.com/jackyalcine/status/615329515909156865
Adjusting your IWA for Global Perspectives
Adjusting your IWA for Global Perspectives Removing Stimulus Component: 1. When you use any of the articles from the Stimulus packet as evidence in your essay, you may keep this as evidence in the essay.
Outcome Forecasting in Sports. Ondřej Hubáček
Outcome Forecasting in Sports Ondřej Hubáček Motivation & Challenges Motivation exploiting betting markets performance optimization Challenges no available datasets difficulties with establishing the state-of-the-art
Carnegie Mellon University, University of Pittsburgh
Carnegie Mellon University, University of Pittsburgh Carnegie Mellon University, University of Pittsburgh Artificial Intelligence (AI) and Deep Learning (DL) Overview Paola Buitrago Leader AI and BD Pittsburgh
Toward AI Network Society
Toward AI Network Society AI Evolution and Human Evolution Refer to Social, Economic, Educational Issue Paris, October 26, 2017 Osamu SUDOH Chair, the Conference toward AI Network Society, MIC, Gov. of
Human + Machine How AI is Radically Transforming and Augmenting Lives and Businesses Are You Ready?
Human + Machine How AI is Radically Transforming and Augmenting Lives and Businesses Are You Ready? Xavier Anglada Managing Director Accenture Digital Lead in MENA and Turkey @xavianglada TM Forum 1 Meet
Medical Intelligence:
Medical Intelligence: Big Data, Predictive Analytics, Machine Learning, and Artificial Intelligence Anthony C. Chang, MD, MBA, MPH Chief Intelligence and Innovation Officer Children s Hospital of Orange
Shuhua Liu Senior Research Fellow, Docent Arcada Universitty of Applied Sciences. KaTuMetro Kickoff Seminar, University of Helsinki
Intelligent Methods and Models for Mining Community Knowledge: Enabling enriched Understanding of Urban Development in Helsinki Metropolitan Region with Social Intelligence Shuhua Liu Senior Research Fellow,
Demystifying Machine Learning
Demystifying Machine Learning By Simon Agius Muscat Software Engineer with RightBrain PyMalta, 19/07/18 http://www.rightbrain.com.mt 0. Talk outline 1. Explain the reasoning behind my talk 2. Defining
Proposers Day Workshop
Proposers Day Workshop Monday, January 23, 2017 @srcjump, #JUMPpdw Cognitive Computing Vertical Research Center Mandy Pant Academic Research Director Intel Corporation Center Motivation Today s deep learning
Artificial Intelligence and Disaster Management
Artificial Intelligence and Disaster Management Dr. Jaziar Radianti Teknologidagene 2018 Trondheim, 31 October 2018 Agenda 1. Artificial Intelligence (AI) and Disaster Management 2. Research on disaster
Machines can learn, but what will we teach them? Geraldine Magarey
Machines can learn, but what will we teach them? Geraldine Magarey The technology AI is a field of computer science that includes o machine learning, o natural language processing, o speech processing,
AI Frontiers. Dr. Dario Gil Vice President IBM Research
AI Frontiers Dr. Dario Gil Vice President IBM Research 1 AI is the new IT MIT Intro to Machine Learning course: 2013 138 students 2016 302 students 2017 700 students 2 What is AI? Artificial Intelligence
COS 402 Machine Learning and Artificial Intelligence Fall Lecture 1: Intro
COS 402 Machine Learning and Artificial Intelligence Fall 2016 Lecture 1: Intro Sanjeev Arora Elad Hazan Today s Agenda Defining intelligence and AI state-of-the-art, goals Course outline AI by introspection
Representation Learning for Mobile Robots in Dynamic Environments
Representation Learning for Mobile Robots in Dynamic Environments Olivia Michael Supervised by A/Prof. Oliver Obst Western Sydney University Vacation Research Scholarships are funded jointly by the Department
Evidence Standards and AI
Evidence Standards and AI Dr Indra Joshi Digital Health and AI, Clinical Lead January 2019 We are creating a digital ecosystem that provides a consistent and trusted experience across digital tools and
The impact of rapid technological change on sustainable development
15-17 January 2019, Vienna The impact of rapid technological change on sustainable development Shamika N. Sirimanne Director, Division on Technology and Logistics UNCTAD 2018-2019 CSTD Intersessional Panel
Panel on Adaptive, Autonomous and Machine Learning: Applications, Challenges and Risks - Introduction
Panel on Adaptive, Autonomous and Machine Learning: Applications, Challenges and Risks - Introduction Prof. Dr. Andreas Rausch Februar 2018 Clausthal University of Technology Institute for Informatics
Introduction to Artificial Intelligence. Department of Electronic Engineering 2k10 Session - Artificial Intelligence
Introduction to Artificial Intelligence What is Intelligence??? Intelligence is the ability to learn about, to learn from, to understand about, and interact with one s environment. Intelligence is the
Norsk Regnesentral (NR) Norwegian Computing Center
Norsk Regnesentral (NR) Norwegian Computing Center Petter Abrahamsen Joining Forces 2018 www.nr.no NUSSE: - 512 9-digit numbers - 200 additions/second Our latest servers: - Four Titan X GPUs - 14 336 cores
Eur Ing Dr. Lei Zhang Faculty of Engineering and Applied Science University of Regina Canada
Eur Ing Dr. Lei Zhang Faculty of Engineering and Applied Science University of Regina Canada The Second International Conference on Neuroscience and Cognitive Brain Information BRAININFO 2017, July 22,
Powerful But Limited: A DARPA Perspective on AI. Arati Prabhakar Director, DARPA
Powerful But Limited: A DARPA Perspective on AI Arati Prabhakar Director, DARPA Artificial intelligence Three waves of AI technology (so far) Handcrafted knowledge Statistical learning Contextual adaptation
ARTIFICIAL INTELLIGENCE TRENDS AND POLICY ISSUES
International Institute of Communications AI Workshop Mexico City, October 9 2018 ARTIFICIAL INTELLIGENCE TRENDS AND POLICY ISSUES Roberto Martínez-Yllescas Head of the OECD Mexico Centre for Latin America
Building a more stable predictive logistic regression model. Anna Elizabeth Campain
Building a more stable predictive logistic regression model Anna Elizabeth Campain Common problems when working with clinical data Missing data Imbalanced class distribution Unstable logistic regression
Lesson 08. Convolutional Neural Network. Ing. Marek Hrúz, Ph.D. Katedra Kybernetiky Fakulta aplikovaných věd Západočeská univerzita v Plzni.
Lesson 08 Convolutional Neural Network Ing. Marek Hrúz, Ph.D. Katedra Kybernetiky Fakulta aplikovaných věd Západočeská univerzita v Plzni Lesson 08 Convolution we will consider 2D convolution the result
NHS Sutton CCG. Interim CCG code. This CCG has 29 practices¹, based on those with a registered population in April 2011.
CCG Profile version 0.32 PDF Created: 25/05/2012 NHS Sutton CCG Interim CCG code 08T Summary Statistics This CCG has 29 practices¹, based on those with a registered population in April 2011. Their total
Inclusive Marketing. jessica fish Leader
Inclusive Marketing jessica fish Leader Networks today How is your institution s commitment to diversity, equity, and inclusion represented in your marketing efforts? Do you feel like you could be more
Comparative Study of various Surveys on Sentiment Analysis
Comparative Study of various Surveys on Milanjit Kaur 1, Deepak Kumar 2. 1 Student (M.Tech Scholar), Computer Science and Engineering, Lovely Professional University, Punjab, India. 2 Assistant Professor,
NHS West London (K&C & QPP) CCG
CCG Profile version 0.32 PDF Created: 25/05/2012 NHS West London (K&C & QPP) CCG Interim CCG code 08Y Summary Statistics This CCG has 55 practices¹, based on those with a registered population in April
SPECIFICITY of MACHINE LEARNING PROJECTS. Borys Pratsiuk, Head of R&D, Ci
1 SPECIFICITY of MACHINE LEARNING PROJECTS Borys Pratsiuk, Head of R&D, Ci 2 Who am I? Senior Android Team Lead Android Architect Engineer, R&D Lab, Tescom, South Korea Android Developer Ph.D Solidstate
Data Anonymization Related Laws in the US and the EU. CS and Law Project Presentation Jaspal Singh
Data Anonymization Related Laws in the US and the EU CS and Law Project Presentation Jaspal Singh The Need for Anonymization To share a database packed with sensitive information with third parties or
Predicting the outcome of NFL games using machine learning Babak Hamadani bhamadan-at-stanford.edu cs229 - Stanford University
Predicting the outcome of NFL games using machine learning Babak Hamadani bhamadan-at-stanford.edu cs229 - Stanford University 1. Introduction: Professional football is a multi-billion industry. NFL is
TNO Whitepaper Marc van Lieshout Wessel Kraaij Hanneke Molema
TNO Whitepaper Marc van Lieshout Wessel Kraaij Hanneke Molema PRIVACY IN DIGITAL HEALTH, A POSITIVE DRIVER FOR INNOVATION 2 / 12 CONTENTS INTRODUCTION 3 PRIVACY: FROM DEFENSIVE TO POSITIVE CONCEPT 4 RESPECT4U:
BLUFF WITH AI. CS297 Report. Presented to. Dr. Chris Pollett. Department of Computer Science. San Jose State University. In Partial Fulfillment
BLUFF WITH AI CS297 Report Presented to Dr. Chris Pollett Department of Computer Science San Jose State University In Partial Fulfillment Of the Requirements for the Class CS 297 By Tina Philip May 2017
Elements of Artificial Intelligence and Expert Systems
Elements of Artificial Intelligence and Expert Systems Master in Data Science for Economics, Business & Finance Nicola Basilico Dipartimento di Informatica Via Comelico 39/41-20135 Milano (MI) Ufficio
Eleonora Escalante, MBA - MEng Strategic Corporate Advisory Services Creating Corporate Integral Value (CIV)
Eleonora Escalante, MBA - MEng Strategic Corporate Advisory Services Creating Corporate Integral Value (CIV) Leg 7. Trends in Competitive Advantage. 21 March 2018 Drawing Source: Edx, Delft University.
Machine Learning. Classification, Discriminative learning. Marc Toussaint University of Stuttgart Summer 2014
Machine Learning Classification, Discriminative learning Structured output, structured input, discriminative function, joint input-output features, Likelihood Maximization, Logistic regression, binary
Big Data, privacy and ethics: current trends and future challenges
Sébastien Gambs Big Data, privacy and ethics 1 Big Data, privacy and ethics: current trends and future challenges Sébastien Gambs Université du Québec à Montréal (UQAM) gambs.sebastien@uqam.ca 24 April
Privacy preserving data mining multiplicative perturbation techniques
Privacy preserving data mining multiplicative perturbation techniques Li Xiong CS573 Data Privacy and Anonymity Outline Review and critique of randomization approaches (additive noise) Multiplicative data
How AI and wearables will take health to the next level - AI Med
How AI and wearables will take health to the next level By AIMed 22 By Nick Van Terheyden, MD Wearables are everywhere and like many technology terms the early entrants have become synonymous and part
Evolutionary Computation for Creativity and Intelligence. By Darwin Johnson, Alice Quintanilla, and Isabel Tweraser
Evolutionary Computation for Creativity and Intelligence By Darwin Johnson, Alice Quintanilla, and Isabel Tweraser Introduction to NEAT Stands for NeuroEvolution of Augmenting Topologies (NEAT) Evolves
AI for Autonomous Ships Challenges in Design and Validation
VTT TECHNICAL RESEARCH CENTRE OF FINLAND LTD AI for Autonomous Ships Challenges in Design and Validation ISSAV 2018 Eetu Heikkilä Autonomous ships - activities in VTT Autonomous ship systems Unmanned engine
Kay Firth-Butterfield, Executive Director Barrister-at-Law, LL.M, M.A., F.R.S.A
Kay Firth-Butterfield, Executive Director Barrister-at-Law, LL.M, M.A., F.R.S.A Copyright 2017. All Rights Reserved. Do not circulate or use without express permission Marc Benioff Davos 2017 AI technology
AI: will the machines save the world (and make me redundant)? Gregor Russell R&D Director Consultant Old Age Psychiatrist Honorary Senior Lecturer
AI: will the machines save the world (and make me redundant)? Gregor Russell R&D Director Consultant Old Age Psychiatrist Honorary Senior Lecturer Your therapist will see you now. The rise of the machines?
AI & Law. What is AI?
AI & Law Gary E. Marchant, J.D., Ph.D. gary.marchant@asu.edu What is AI? A machine that displays intelligent behavior, such as reasoning, learning and sensory processing. AI involves tasks that have historically
Dota2 is a very popular video game currently.
Dota2 Outcome Prediction Zhengyao Li 1, Dingyue Cui 2 and Chen Li 3 1 ID: A53210709, Email: zhl380@eng.ucsd.edu 2 ID: A53211051, Email: dicui@eng.ucsd.edu 3 ID: A53218665, Email: lic055@eng.ucsd.edu March
Beyond Reliability: Advanced Analytics for Predicting Quality
Beyond Reliability: Advanced Analytics for Predicting Quality William J. Goodrum, Jr., PhD Elder Research, Inc. william.goodrum@elderresearch.com Headquarters 300 W. Main Street, Suite 301 Charlottesville,
PHARMACEUTICALS: WHEN AI ADOPTION HAS GATHERED MOST MOMENTUM.
PHARMACEUTICALS: WHEN AI ADOPTION HAS GATHERED MOST MOMENTUM www.infosys.com/aimaturity It is critical to make sure that society can take full advantage of the capabilities of AI systems while minimizing
Privacy-Preserving Collaborative Recommendation Systems Based on the Scalar Product
Privacy-Preserving Collaborative Recommendation Systems Based on the Scalar Product Justin Zhan I-Cheng Wang Abstract In the e-commerce era, recommendation systems were introduced to share customer experience
Applications of Machine Learning Techniques in Human Activity Recognition
Applications of Machine Learning Techniques in Human Activity Recognition Jitenkumar B Rana Tanya Jha Rashmi Shetty Abstract Human activity detection has seen a tremendous growth in the last decade playing
The IEEE Global Initiative for Ethical Considerations in Artificial Intelligence and Autonomous Systems. Overview April, 2017
The IEEE Global Initiative for Ethical Considerations in Artificial Intelligence and Autonomous Systems Overview April, 2017 @johnchavens 3 IEEE Standards Association IEEE s Technology Ethics Landscape
Your Name and ID. (a) ( 3 points) Breadth First Search is complete even if zero step-costs are allowed.
1 UC Davis: Winter 2003 ECS 170 Introduction to Artificial Intelligence Final Examination, Open Text Book and Open Class Notes. Answer All questions on the question paper in the spaces provided Show all
Automating NSF HERD Reporting Using Machine Learning and Administrative Data
Automating NSF HERD Reporting Using Machine Learning and Administrative Data Rodolfo H. Torres CIMA Session: The Use of Advance Analytics to Drive Decisions 2018 APLU Annual Meeting New Orleans Marriott,
The Information Commissioner s response to the Draft AI Ethics Guidelines of the High-Level Expert Group on Artificial Intelligence
Wycliffe House, Water Lane, Wilmslow, Cheshire, SK9 5AF T. 0303 123 1113 F. 01625 524510 www.ico.org.uk The Information Commissioner s response to the Draft AI Ethics Guidelines of the High-Level Expert
Computing Science (CMPUT) 496
Computing Science (CMPUT) 496 Search, Knowledge, and Simulations Martin Müller Department of Computing Science University of Alberta mmueller@ualberta.ca Winter 2017 Part IV Knowledge 496 Today - Mar 9
AIMed Artificial Intelligence in Medicine
Medical Intelligence and Innovation Institute (MI3) Presents The First International Multidisciplinary Symposium on Artificial Intelligence in Medicine: Analytics and Algorithms, Big Data, Cloud and Cognitive
Policies for the Commissioning of Health and Healthcare
Policies for the Commissioning of Health and Healthcare Statement of Principles REFERENCE NUMBER Commissioning policies statement of principles VERSION V1.0 APPROVING COMMITTEE & DATE Governing Body 26.5.15
Workshop on anonymization Berlin, March 19, Basic Knowledge Terms, Definitions and general techniques. Murat Sariyar TMF
Workshop on anonymization Berlin, March 19, 2015 Basic Knowledge Terms, Definitions and general techniques Murat Sariyar TMF Workshop Anonymisation, March 19, 2015 Outline Background Aims of Anonymization
1 2-step and other basic conditional probability problems
Name M362K Exam 2 Instructions: Show all of your work. You do not have to simplify your answers. No calculators allowed. 1 2-step and other basic conditional probability problems 1. Suppose A, B, C are
Rise of the Intelligent Internet of Things
Rise of the Intelligent Internet of Things Opportunities and Challenges of Connected Patients, Environments and Machines Mark Wolff, M.S., Ph.D. Advisory Industry Consultant Chief Health Analytics Strategist
The Future of Patient Data The Global View Key Insights Berlin 18 April The world s leading open foresight program
The Future of Patient Data The Global View Key Insights Berlin 18 April 2018 The world s leading open foresight program Context Over a 6 month period, 12 expert discussions have taken place around the
100 Year Study on AI: 1st Study Panel Report
Artificial Intelligence and Life in 2030 100 Year Study on AI: 1st Study Panel Report Prof. Peter Stone* Study Panel Chair Department of Computer Science The University of Texas at Austin *Also Cogitai,
Internet of Things and Healthcare Analytics for Better Healthcare Solution: Applications and Challenges
Internet of Things and Healthcare Analytics for Better Healthcare Solution: Applications and Challenges Zuraida Abal Abas 1, Zaheera Zainal Abidin 2, Ahmad Fadzli Nizam Abdul Rahman 3, Hidayah Rahmalan
Utah Advance Directive Form & Instructions
Utah Advance Directive Form & Instructions 2009 Edition published by Utah Medical Association 310 E. 4500 South, Suite 500 Salt Lake City, UT 84107 Instructions for Completing the Advance Health Care Directive
Three Minute Thesis & Research Presentations.
Three Minute Thesis & Research Presentations Ludovica Luisa Vissat Modelling and analysis of spatial stochastic systems Case-study: disease spread Average infected population Probability of an epidemic
Early HTA to inform value driven market access and reimbursement planning
Early HTA to inform value driven market access and reimbursement planning Lotte MG Steuten, PhD Associate Prof. Health Technology & Services Research Program Director Health Sciences University of Twente,
GPU ACCELERATED DEEP LEARNING WITH CUDNN
GPU ACCELERATED DEEP LEARNING WITH CUDNN Larry Brown Ph.D. March 2015 AGENDA 1 Introducing cudnn and GPUs 2 Deep Learning Context 3 cudnn V2 4 Using cudnn 2 Introducing cudnn and GPUs 3 HOW GPU ACCELERATION
Community Health Services Medical Patient Experience Survey Results Organization
1. Patient Information What is your age? 0-12 45 9.9% 13-19 33 7.3% 20-29 50 11.0% 30-39 60 13.2% 40-49 68 15.0% 50-64 129 28.4% 65+ 69 15.2% What is your gender? Male 86 29.8% Female 203 70.2% Transgender
NHS Ipswich and East Suffolk CCG
CCG Profile version 0.32 PDF Created: 25/05/2012 NHS Ipswich and East Suffolk CCG Interim CCG code 06L Summary Statistics This CCG has 42 practices¹, based on those with a registered population in April
Community Health Services Medical Patient Experience Survey Results Organization - November 2018
1. Patient Information What is your age? 0-12 24 7.2% 13-19 13 3.9% 20-29 41 12.3% 30-39 50 15.1% 40-49 55 16.6% 50-64 95 28.6% 65+ 54 16.3% What is your gender? Male 64 30.0% Female 148 69.5% Transgender
Surveillance and Privacy in the Information Age. Image courtesy of Josh Bancroft on flickr. License CC-BY-NC.
Surveillance and Privacy in the Information Age Image courtesy of Josh Bancroft on flickr. License CC-BY-NC. 1 Basic attributes (Kitchin, 2014) High-volume High-velocity High-variety Exhaustivity (n=all)
TRUSTING THE MIND OF A MACHINE
TRUSTING THE MIND OF A MACHINE AUTHORS Chris DeBrusk, Partner Ege Gürdeniz, Principal Shriram Santhanam, Partner Til Schuermann, Partner INTRODUCTION If you can t explain it simply, you don t understand
Probability (Devore Chapter Two)
Probability (Devore Chapter Two) 1016-351-01 Probability Winter 2011-2012 Contents 1 Axiomatic Probability 2 1.1 Outcomes and Events............................... 2 1.2 Rules of Probability................................
Inference of Opponent s Uncertain States in Ghosts Game using Machine Learning
Inference of Opponent s Uncertain States in Ghosts Game using Machine Learning Sehar Shahzad Farooq, HyunSoo Park, and Kyung-Joong Kim* sehar146@gmail.com, hspark8312@gmail.com,kimkj@sejong.ac.kr* Department
Session 2: 10 Year Vision session (11:00-12:20) - Tuesday. Session 3: Poster Highlights A (14:00-15:00) - Tuesday 20 posters (3minutes per poster)
Lessons from Collecting a Million Biometric Samples 109 Expression Robust 3D Face Recognition by Matching Multi-component Local Shape Descriptors on the Nasal and Adjoining Cheek Regions 177 Shared Representation