This presentation discusses applications of artificial intelligence, machine learning, and deep learning in actuarial science. It provides an introduction to machine learning and deep learning, including different architectures like feedforward neural networks and embedding layers. It then discusses several potential applications of these techniques in actuarial problems, including non-life insurance pricing, IBNR reserving, analyzing telematics driving data, and mortality forecasting. The presentation concludes by noting that deep learning has the potential to enhance predictive modeling in actuarial science and that its application in the field seems to be an emerging area of research.
The slide has details on below points:
1. Introduction to Machine Learning
2. What are the challenges in acceptance of Machine Learning in Banks
3. How to overcome the challenges in adoption of Machine Learning in Banks
4. How to find new use cases of Machine Learning
5. Few current interesting use cases of Machine Learning
Please contact me (shekup@gmail.com) or connect with me on LinkedIn (https://www.linkedin.com/in/shekup/) for more explanation on ML and how it may help your business.
The slides are inspired by:
Survey & interviews done by me with Bankers & Technology Professionals
Presentation from Google NEXT 2017
Presentation by DATUM on Youtube
Royal Society Machine Learning
Big Data & Social Analytics Course from MIT & GetSmarter
Vertex has invested in companies across geographies addressing different industry applications leveraging AI to transform their service offerings. Read more on the trends and waves of AI developments observed.
The goal of this course is to offer data science and fintech enthusiasts a hand-on practical case study to understand the power of Data Science, ML and AI in Finance. We discuss two case studies; An NLP case study and a Credit Risk case study to reinforce concepts
Credit Risk Introduction and Pre-class preparation
Pre-class reading. We will be using the Lending club data set to build a credit risk model using machine learning techniques. This workshop was be delivered in Boston and Online by Sri Krishnamurthy.
Leveraging Computational Methods for Theorizing IS PhenomenaMalmi Amadoru
Â
The rapid development of computational methods expands the horizon of opportunities in research methods. Scholars have acknowledged the potential of computationally intensive research approaches for theorizing IS phenomena. However, computationally intensive theory building is still at a nascent stage. This presentation focuses on how to leverage computational methods in the theorizing process, associated challenges, and respective strategies.
The slide has details on below points:
1. Introduction to Machine Learning
2. What are the challenges in acceptance of Machine Learning in Banks
3. How to overcome the challenges in adoption of Machine Learning in Banks
4. How to find new use cases of Machine Learning
5. Few current interesting use cases of Machine Learning
Please contact me (shekup@gmail.com) or connect with me on LinkedIn (https://www.linkedin.com/in/shekup/) for more explanation on ML and how it may help your business.
The slides are inspired by:
Survey & interviews done by me with Bankers & Technology Professionals
Presentation from Google NEXT 2017
Presentation by DATUM on Youtube
Royal Society Machine Learning
Big Data & Social Analytics Course from MIT & GetSmarter
Vertex has invested in companies across geographies addressing different industry applications leveraging AI to transform their service offerings. Read more on the trends and waves of AI developments observed.
The goal of this course is to offer data science and fintech enthusiasts a hand-on practical case study to understand the power of Data Science, ML and AI in Finance. We discuss two case studies; An NLP case study and a Credit Risk case study to reinforce concepts
Credit Risk Introduction and Pre-class preparation
Pre-class reading. We will be using the Lending club data set to build a credit risk model using machine learning techniques. This workshop was be delivered in Boston and Online by Sri Krishnamurthy.
Leveraging Computational Methods for Theorizing IS PhenomenaMalmi Amadoru
Â
The rapid development of computational methods expands the horizon of opportunities in research methods. Scholars have acknowledged the potential of computationally intensive research approaches for theorizing IS phenomena. However, computationally intensive theory building is still at a nascent stage. This presentation focuses on how to leverage computational methods in the theorizing process, associated challenges, and respective strategies.
The state of the art in integrating machine learning into visual analyticsCagatay Turkay
Â
Slides for my talk on our paper at EuroVis 2017 on the STAR track:
Endert, A., Ribarsky, W., Turkay, C., Wong, B.L., Nabney, I., Blanco, I.D. and Rossi, F., 2017, March. The state of the art in integrating machine learning into visual analytics. In Computer Graphics Forum.
http://openaccess.city.ac.uk/16739/
Traditional Machine Learning and Deep Learning on OpenPOWER/POWER systemsGanesan Narayanasamy
Â
This presentation gave deep dive into various machine learning and deep learning algorithms followed by an overview of the hardware and software technologies for democratization of AI including OpenPOWER/POWER9 solutions.
Machine learning with an effective tools of data visualization for big dataKannanRamasamy25
Â
Arthur Samuel (1959) :
"Field of study that gives computers the ability to learn without being explicitly programmedâ
Tom Mitchell (1998) :
âA computer program is said to learn from experience E with respect to some task T and some performance measure P, if its performance on T, as measured by P, improves with experience Eâ.
There are several ways to implement machine learning algorithms
Automating automation
Getting computers to program themselves
Writing software is the bottleneck
Let the data do the work instead!
Innovations in technology has revolutionized financial services to an extent that large financial institutions like Goldman Sachs are claiming to be technology companies! It is no secret that technological innovations like Data science and AI are changing fundamentally how financial products are created, tested and delivered. While it is exciting to learn about technologies themselves, there is very little guidance available to companies and financial professionals should retool and gear themselves towards the upcoming revolution.
In this master class, we will discuss key innovations in Data Science and AI and connect applications of these novel fields in forecasting and optimization. Through case studies and examples, we will demonstrate why now is the time you should invest to learn about the topics that will reshape the financial services industry of the future!
AI in Finance
Deciphering AI - Unlocking the Black Box of AIML with State-of-the-Art Techno...Analytics India Magazine
Â
Most organizations understand the predictive power and the potential gains from AIML, but AI and ML are still now a black box technology for them. While deep learning and neural networks can provide excellent inputs to businesses, leaders are challenged to use them because of the complete blind faith required to âtrustâ AI. In this talk we will use the latest technological developments from researchers, the US defense department, and the industry to unbox the black box and provide businesses a clear understanding of the policy levers that they can pull, why, and by how much, to make effective decisions?
Machine Learning 2 deep Learning: An IntroSi Krishan
Â
Provides a brief introduction to machine learning, reasons for its popularity, a simple walk through example and then a need for deep learning and some of its characteristics. This is an updated version of an earlier presentation.
Lecture 2 - Introduction to Machine Learning, a lecture in subject module Sta...Maninda Edirisooriya
Â
Introduction to Statistical and Machine Learning. Explains basics of ML, fundamental concepts of ML, Statistical Learning and Deep Learning. Recommends the learning sources and techniques of Machine Learning. This was one of the lectures of a full course I taught in University of Moratuwa, Sri Lanka on 2023 second half of the year.
The state of the art in integrating machine learning into visual analyticsCagatay Turkay
Â
Slides for my talk on our paper at EuroVis 2017 on the STAR track:
Endert, A., Ribarsky, W., Turkay, C., Wong, B.L., Nabney, I., Blanco, I.D. and Rossi, F., 2017, March. The state of the art in integrating machine learning into visual analytics. In Computer Graphics Forum.
http://openaccess.city.ac.uk/16739/
Traditional Machine Learning and Deep Learning on OpenPOWER/POWER systemsGanesan Narayanasamy
Â
This presentation gave deep dive into various machine learning and deep learning algorithms followed by an overview of the hardware and software technologies for democratization of AI including OpenPOWER/POWER9 solutions.
Machine learning with an effective tools of data visualization for big dataKannanRamasamy25
Â
Arthur Samuel (1959) :
"Field of study that gives computers the ability to learn without being explicitly programmedâ
Tom Mitchell (1998) :
âA computer program is said to learn from experience E with respect to some task T and some performance measure P, if its performance on T, as measured by P, improves with experience Eâ.
There are several ways to implement machine learning algorithms
Automating automation
Getting computers to program themselves
Writing software is the bottleneck
Let the data do the work instead!
Innovations in technology has revolutionized financial services to an extent that large financial institutions like Goldman Sachs are claiming to be technology companies! It is no secret that technological innovations like Data science and AI are changing fundamentally how financial products are created, tested and delivered. While it is exciting to learn about technologies themselves, there is very little guidance available to companies and financial professionals should retool and gear themselves towards the upcoming revolution.
In this master class, we will discuss key innovations in Data Science and AI and connect applications of these novel fields in forecasting and optimization. Through case studies and examples, we will demonstrate why now is the time you should invest to learn about the topics that will reshape the financial services industry of the future!
AI in Finance
Deciphering AI - Unlocking the Black Box of AIML with State-of-the-Art Techno...Analytics India Magazine
Â
Most organizations understand the predictive power and the potential gains from AIML, but AI and ML are still now a black box technology for them. While deep learning and neural networks can provide excellent inputs to businesses, leaders are challenged to use them because of the complete blind faith required to âtrustâ AI. In this talk we will use the latest technological developments from researchers, the US defense department, and the industry to unbox the black box and provide businesses a clear understanding of the policy levers that they can pull, why, and by how much, to make effective decisions?
Machine Learning 2 deep Learning: An IntroSi Krishan
Â
Provides a brief introduction to machine learning, reasons for its popularity, a simple walk through example and then a need for deep learning and some of its characteristics. This is an updated version of an earlier presentation.
Lecture 2 - Introduction to Machine Learning, a lecture in subject module Sta...Maninda Edirisooriya
Â
Introduction to Statistical and Machine Learning. Explains basics of ML, fundamental concepts of ML, Statistical Learning and Deep Learning. Recommends the learning sources and techniques of Machine Learning. This was one of the lectures of a full course I taught in University of Moratuwa, Sri Lanka on 2023 second half of the year.
Synthetic Fiber Construction in lab .pptxPavel ( NSTU)
Â
Synthetic fiber production is a fascinating and complex field that blends chemistry, engineering, and environmental science. By understanding these aspects, students can gain a comprehensive view of synthetic fiber production, its impact on society and the environment, and the potential for future innovations. Synthetic fibers play a crucial role in modern society, impacting various aspects of daily life, industry, and the environment. ynthetic fibers are integral to modern life, offering a range of benefits from cost-effectiveness and versatility to innovative applications and performance characteristics. While they pose environmental challenges, ongoing research and development aim to create more sustainable and eco-friendly alternatives. Understanding the importance of synthetic fibers helps in appreciating their role in the economy, industry, and daily life, while also emphasizing the need for sustainable practices and innovation.
The Roman Empire A Historical Colossus.pdfkaushalkr1407
Â
The Roman Empire, a vast and enduring power, stands as one of history's most remarkable civilizations, leaving an indelible imprint on the world. It emerged from the Roman Republic, transitioning into an imperial powerhouse under the leadership of Augustus Caesar in 27 BCE. This transformation marked the beginning of an era defined by unprecedented territorial expansion, architectural marvels, and profound cultural influence.
The empire's roots lie in the city of Rome, founded, according to legend, by Romulus in 753 BCE. Over centuries, Rome evolved from a small settlement to a formidable republic, characterized by a complex political system with elected officials and checks on power. However, internal strife, class conflicts, and military ambitions paved the way for the end of the Republic. Julius Caesarâs dictatorship and subsequent assassination in 44 BCE created a power vacuum, leading to a civil war. Octavian, later Augustus, emerged victorious, heralding the Roman Empireâs birth.
Under Augustus, the empire experienced the Pax Romana, a 200-year period of relative peace and stability. Augustus reformed the military, established efficient administrative systems, and initiated grand construction projects. The empire's borders expanded, encompassing territories from Britain to Egypt and from Spain to the Euphrates. Roman legions, renowned for their discipline and engineering prowess, secured and maintained these vast territories, building roads, fortifications, and cities that facilitated control and integration.
The Roman Empireâs society was hierarchical, with a rigid class system. At the top were the patricians, wealthy elites who held significant political power. Below them were the plebeians, free citizens with limited political influence, and the vast numbers of slaves who formed the backbone of the economy. The family unit was central, governed by the paterfamilias, the male head who held absolute authority.
Culturally, the Romans were eclectic, absorbing and adapting elements from the civilizations they encountered, particularly the Greeks. Roman art, literature, and philosophy reflected this synthesis, creating a rich cultural tapestry. Latin, the Roman language, became the lingua franca of the Western world, influencing numerous modern languages.
Roman architecture and engineering achievements were monumental. They perfected the arch, vault, and dome, constructing enduring structures like the Colosseum, Pantheon, and aqueducts. These engineering marvels not only showcased Roman ingenuity but also served practical purposes, from public entertainment to water supply.
Palestine last event orientationfvgnh .pptxRaedMohamed3
Â
An EFL lesson about the current events in Palestine. It is intended to be for intermediate students who wish to increase their listening skills through a short lesson in power point.
Biological screening of herbal drugs: Introduction and Need for
Phyto-Pharmacological Screening, New Strategies for evaluating
Natural Products, In vitro evaluation techniques for Antioxidants, Antimicrobial and Anticancer drugs. In vivo evaluation techniques
for Anti-inflammatory, Antiulcer, Anticancer, Wound healing, Antidiabetic, Hepatoprotective, Cardio protective, Diuretics and
Antifertility, Toxicity studies as per OECD guidelines
Operation âBlue Starâ is the only event in the history of Independent India where the state went into war with its own people. Even after about 40 years it is not clear if it was culmination of states anger over people of the region, a political game of power or start of dictatorial chapter in the democratic setup.
The people of Punjab felt alienated from main stream due to denial of their just demands during a long democratic struggle since independence. As it happen all over the word, it led to militant struggle with great loss of lives of military, police and civilian personnel. Killing of Indira Gandhi and massacre of innocent Sikhs in Delhi and other India cities was also associated with this movement.
How to Make a Field invisible in Odoo 17Celine George
Â
It is possible to hide or invisible some fields in odoo. Commonly using âinvisibleâ attribute in the field definition to invisible the fields. This slide will show how to make a field invisible in odoo 17.
Model Attribute Check Company Auto PropertyCeline George
Â
In Odoo, the multi-company feature allows you to manage multiple companies within a single Odoo database instance. Each company can have its own configurations while still sharing common resources such as products, customers, and suppliers.
Introduction to AI for Nonprofits with Tapp NetworkTechSoup
Â
Dive into the world of AI! Experts Jon Hill and Tareq Monaur will guide you through AI's role in enhancing nonprofit websites and basic marketing strategies, making it easy to understand and apply.
The French Revolution, which began in 1789, was a period of radical social and political upheaval in France. It marked the decline of absolute monarchies, the rise of secular and democratic republics, and the eventual rise of Napoleon Bonaparte. This revolutionary period is crucial in understanding the transition from feudalism to modernity in Europe.
For more information, visit-www.vavaclasses.com
2. This presentation is given solely in my personal capacity and the views
expressed in these slides do not necessarily represent those of the AIG
Group and its subsidiaries, nor the professional organizations to which I
belong (the IFoA, ASSA or IRMSA).
Any software or code referred to in the presentation is provided as is for
demonstration purposes only, without any implied warranty, and is licensed
under the MIT License which can be viewed on the associated GitHub
repository.
Disclaimer
3. ⢠Introduction
⢠Machine Learning
⢠Deep Learning
⢠Applications in Actuarial Science
⢠Discussion and Conclusion
Agenda
4. Introduction
⢠This talk is about 3 things:
⢠Provide context to understand deep learning
⢠Discuss applications of deep learning in actuarial science
⢠Provide code to experiment (see last slide)
⢠Inspiration of paper and talk:
"The future of insurance will be staffed by bots rather than brokers and AI
in favor of actuaries"- Daniel Schreiber, CEO, Lemonade Inc.
5. Deep Learning in the Wild
⢠We all use Deep Learning today:
⢠Google/Apple/Facebook/Instagram
/PinterestâŚ
⢠⌠and might use it more in the
medium term (self-driving
cars/medical applications)
⢠We help to train DL - Recaptcha
⢠DL is good enough to trick us
⢠But, are actuaries benefiting from
Deep Learning?
Man from www.thispersondoesnotexist.com/
YOLO from https://github.com/pjreddie/darknet/wiki/YOLO:-Real-Time-Object-
Detection
6. ⢠Computer vision starting with AlexNet architecture of Krizhevsky,
Sutskever and Hinton (2012)
⢠Speech recognition (Hannun, Case, Casper et al. 2014).
⢠Natural language processing, e.g. Googleâs neural translation machine
(Wu, Schuster, Chen et al. 2016)
⢠Winning method in 2018 M4 time series forecasting competition
(Makridakis, Spiliotis and Assimakopoulos 2018a).
⢠Analysis of GPS data (BrÊbisson, Simon, Auvolat et al. 2015)
⢠Analysis of tabular data (Guo and Berkhahn 2016) (plus other Kaggle
competitions)
Practical Successes of Deep Learning
7. ⢠Introduction
⢠Machine Learning
⢠Deep Learning
⢠Applications in Actuarial Science
⢠Discussion and Conclusion
Agenda
8. Machine Learning
⢠Machine Learning is concerned with âthe study of algorithms that allow
computer programs to automatically improve through experienceâ
(Mitchell 1997)
⢠Machine learning approach to AI - systems trained to recognize patterns within
data to acquire knowledge (Goodfellow, Bengio and Courville 2016).
⢠Earlier attempts to build AI systems = hard code knowledge into
knowledge bases
⢠But doesnât work for highly complex tasks e.g. image recognition, scene
understanding and inferring semantic concepts (Bengio 2009)
⢠ML Paradigm â feed data to the machine and let it figure it out!
9. Map of Machine Learning
Reinforcement Learning
Regression
Deep Learning
Machine Learning
Unsupervised Learning
Supervised Learning
Classification
10. Supervised Learning
⢠Supervised learning = application of machine learning to datasets that
contain features and outputs with the goal of predicting the outputs
from the features (Friedman, Hastie and Tibshirani 2009).
X (features)
y (outputs)
11. So, ML is just regression, right?
⢠Not exactly. ML relies on a different approach to building,
parameterizing and testing statistical models, based on statistical
learning theory. For other ideas â see Richman (2018)
⢠Distinction between tasks of predicting and explaining, see Shmueli
(2010). Focus on predictive performance leads to:
⢠Building algorithms to predict responses instead of specifying a stochastic data
generating model (Breiman 2001)âŚ
⢠⌠favouring models with good predictive performance at expense of
intepretablity.
⢠Accepting bias in models if this is expected to reduce the overall prediction error.
⢠Quantifying predictive error (i.e. out-of-sample error)
12. Unsupervised learning
⢠Unsupervised learning = application of machine learning to datasets
containing only features to find structure within these datasets (Sutton
and Barto 2018).
⢠Task of unsupervised learning is to find meaningful patterns using only
the features.
⢠Recent examples:
⢠modelling yield curves using Principal Components Analysis (PCA) for the
Interest Rate SCR in SII
⢠mortality modelling â Lee-Carter model uses PCA to reconstruct mortality curves
13. The ML Actuary
⢠Actuarial problems are often supervised regressions =>
⢠If an actuarial problem can be expressed as a regression, then machine
and deep learning techniques can be applied:
⢠P&C pricing
⢠IBNR reserving
⢠Experience analysis
⢠Mortality modelling
⢠Lite valuation models
⢠But donât forget about unsupervised learning either!
14. ⢠Introduction
⢠Machine Learning
⢠Deep Learning
⢠Applications in Actuarial Science
⢠Discussion and Conclusion
Agenda
15. Feature Engineering (Model Specification)
⢠Suppose we realize that Claims depends on Age^2 => enlarge feature
space by adding Age^2 to data. Other options â add interactions/basis
functions e.g. splines
X (features)
y (outputs)
0.06
0.09
0.12
20 40 60 80
DrivAge
rate
16. Representation learning
⢠In many domains, including actuarial science, traditional approach to
designing machine learning systems relies on humans for feature
engineering. But:
⢠designing features is time consuming/tedious
⢠relies on expert knowledge that may not be transferable to a new domain
⢠becomes difficult with very high dimensional data
⢠Representation Learning = ML technique where algorithms
automatically design features that are optimal for a particular task.
Traditional examples are PCA (unsupervised) and PLS (supervised)
⢠Simple/naive RL approaches often fail when applied to high dimensional
data
17. Deep Learning
⢠Deep Learning = representation learning technique that automatically
constructs hierarchies of complex features
⢠Modern example of deep learning is feed-forward neural networks,
which are multi-layered machine learning models, where each layer
learns a new representation of the features.
⢠The principle: Provide data to the network and let it figure out what and
how to learn.
⢠Desiderata for AI by Bengio (2009):
⢠âAbility to learn with little human input the low-level, intermediate, and high-
level abstractions that would be useful to represent the kind of complex
functions needed for AI tasks.â
18. ⢠Single layer neural network
⢠Circles = variables
⢠Lines = connections between inputs
and outputs
⢠Input layer holds the variables that are
input to the networkâŚ
⢠⌠multiplied by weights (coefficients)
to get to result
⢠Single layer neural network is a linear
regression!
Single Layer NN = Linear Regression
19. ⢠Deep = multiple layers
⢠Feedforward = data travels from left to
right
⢠Fully connected network = all neurons
in layer connected to all neurons in
previous layer
⢠More complicated representations of
input data learned in hidden layers
⢠Subsequent layers represent
regressions on the variables in hidden
layers
Deep Feedforward Net
20. ⢠Several specialized types of neural networks depending on purpose
⢠Embedding layer learns dense vector transformation of sparse input
vectors and clusters similar categories together; see Section 3.3 in
Richman (2018)
⢠Embeddings often capture actuarially meaningful relationships in
categorical data â can be interpreted as relativities
Embedding layers
Actuary Accountant Quant Statistician Economist Underwriter
Actuary 1 0 0 0 0 0
Accountant 0 1 0 0 0 0
Quant 0 0 1 0 0 0
Statistician 0 0 0 1 0 0
Economist 0 0 0 0 1 0
Underwriter 0 0 0 0 0 1
Finance Math Stastistics Liabilities
Actuary 0.5 0.25 0.5 0.5
Accountant 0.5 0 0 0
Quant 0.75 0.25 0.25 0
Statistician 0 0.5 0.85 0
Economist 0.5 0.25 0.5 0
Underwriter 0 0.1 0.05 0.75
21. Summary of architectures
⢠Key principle - Use architecture that expresses useful priors about the
data => major performance gains:
⢠Deep feedforward network â structured (tabular) data
⢠Embedding layers â categorical data (or real values restructured as
categorical data)
⢠Deep autoencoder (non-linear PCA) â unsupervised learning
⢠Convolutional neural network â data with spatial/temporal dimension
e.g. images and time series
⢠Recurrent neural network â data with temporal structure
22. ⢠Introduction
⢠Machine Learning
⢠Deep Learning
⢠Applications in Actuarial Science
⢠Discussion and Conclusion
Agenda
23. Summary of architectures
⢠Searches within actuarial literature confined to articles written after
2006, when current resurgence of interest in neural networks began
(Goodfellow, Bengio and Courville 2016).
⢠Pricing of non-life insurance (Noll, Salzmann and Wßthrich 2018; Wßthrich and
Buser 2018) X
⢠IBNR Reserving (Kuo 2018b; Wßthrich 2018b; Zarkadoulas 2017) X
⢠Analysis of telematics data (Gao, Meng and Wßthrich 2018; Gao and Wßthrich
2017; WĂźthrich and Buser 2018; WĂźthrich 2017) X
⢠Mortality forecasting (Hainaut 2018; Richman and Wßthrich 2018)
⢠Approximating nested stochastic simulations (Hejazi and Jackson 2016, 2017)
⢠Forecasting financial markets (Smith, Beyers and De Villiers 2016)
24. Non-life pricing (1)
⢠Non-life Pricing (tabular data fit with GLMs) seems like obvious
application of ML/DL
⢠Noll, Salzmann and Wßthrich (2018) is tutorial paper (with code) in
which apply GLMs, regression trees, boosting and (shallow) neural
networks to French TPL dataset to model frequency
⢠ML approaches outperform GLM
⢠Boosted tree performs about as well as neural networkâŚ
⢠âŚ.mainly because ML approaches capture some interactions automatically
⢠In own analysis, found that surprisingly, off the shelf approaches do not perform
particularly well on frequency models.
⢠These include XGBoost and âvanillaâ deep networks
25. Non-life pricing (2)
⢠Deep neural network applied to
raw data (i.e. no feature
engineering) did not perform
well
⢠Embedding layers provide
significant gain in performance
over GLM and other NN
architectures
⢠Layers learn a (multi-
dimensional) schedule of
relativities at each age (shown
after applying t-SNE)
⢠Transfer learning â can boost
performance of GLM
-50
-25
0
25
20 40 60 80
Drivage
value
variable dim1 dim2
Model OutOfSample
GLM 0.3217
GLM_Keras 0.3217
NN_shallow 0.3150
NN_no_FE 0.3258
NN_embed 0.3068
GLM_embed 0.3194
NN_learned_embed 0.2925
26. IBNR Reserving
⢠IBNR Reserving boils down to regression of future reported claim
amounts on past => good potential for ML/DL approaches
⢠Granular reserving for claim type/property damaged/region/age etc difficult
with normal chain-ladder approach as too much data to derive LDFs
judgementally
⢠Wßthrich (2018b) (who provides code + data) extends chain-ladder as a
regression model to incorporate features into derivation of LDF
⢠DeepTriangle of Kuo (2018b) is less traditional approach. Joint prediction of Paid
+ Outstanding claims using Recurrent Neural Networks and Embedding Layers
⢠Better performance than CL/GLM/Bayesian techniques on Schedule P data from
USA
j
i
j
i C
X
f
C ,
, ).
(
Ë ď˝
ďŤ1
27. Telematics data (1)
⢠Telematics produces high dimensional data (position, velocity,
acceleration, road type, time of day) at high frequencies â not
immediately obvious how to incorporate into pricing
⢠Sophisticated approaches to analysing telematics data from outside actuarial
literature using recurrent neural networks plus embedding layers such as Dong,
Li, Yao et al. (2016), Dong, Yuan, Yang et al. (2017) and Wijnands, Thompson,
Aschwanden et al. (2018)
⢠Within actuarial literature, series of papers by Wßthrich (2017), Gao and
WĂźthrich (2017) and Gao, Meng and WĂźthrich (2018) discuss analysis of velocity
and acceleration information from telematics data feed
⢠Focus on v-a heatmaps which capture velocity and acceleration profile of driver
but these are also high dimensional
28. Telematics data (2)
⢠Heatmap generated using code in
WĂźthrich (2018c)
⢠Shows density i.e. probability that driver is
found at location in heatmap
⢠Wßthrich (2017) and Gao and Wßthrich
(2017) apply unsupervised learning
methods to summarize v-a heat-maps:
⢠K-means, PCA and shallow auto-encoders
⢠Stunning result = continuous features are
highly predictive
⢠Why? Goodfellow, Bengio and Courville
(2016) : âbasic idea is features useful for
the unsupervised task also be useful for
the supervised learning taskâ
0.000
0.005
0.010
0.015
6 8 10 12 14 16 18 20
-2
-1
0
1
2
v-a heatmap of driver 20
speed in km/h
acceleration
in
m/s^2
29. ⢠Introduction
⢠Machine Learning
⢠Deep Learning
⢠Applications in Actuarial Science
⢠Discussion and Conclusion
Agenda
30. Discussion
⢠Emphasis on predictive performance and potential gains of moving from
traditional actuarial and statistical methods to machine and deep
learning approaches.
⢠Measurement framework utilized within machine learning â focus on
testing predictive performance => focus on measurable improvements
in predictive performance led to refinements and enhancements of
deep learning architectures
⢠Learned representations from deep neural networks often have readily
interpretable meaning
⢠Very useful for high-frequency and high-dimensional data
31. Conclusion
⢠Deep learning can enhance the predictive power of models built by
actuaries
⢠Application of deep learning techniques to actuarial problems seems to
be rapidly emerging field within actuarial science => appears reasonable
to predict more advances in the near-term.
⢠Deep learning is not a panacea for all modelling issues - applied to the
wrong domain, deep learning will not produce better or more useful
results than other techniques.
⢠Winter might be coming â if actuaries do not take the lead in applying
deep learning, someone else will.
32. Bengio, Y. 2009. "Learning deep architectures for AI", Foundations and trendsÂŽ in Machine Learning 2(1):1-127.
De BrĂŠbisson, A., Ă. Simon, A. Auvolat, P. Vincent et al. 2015. "Artificial neural networks applied to taxi destination prediction", arXiv arXiv:1508.00021
Dong, W., J. Li, R. Yao, C. Li et al. 2016. "Characterizing driving styles with deep learning", arXiv arXiv:1607.03611
Dong, W., T. Yuan, K. Yang, C. Li et al. 2017. "Autoencoder regularized network for driving style representation learning", arXiv arXiv:1701.01272
Friedman, J., T. Hastie and R. Tibshirani. 2009. The Elements of Statistical Learning : Data Mining, Inference, and Prediction. New York: Springer-Verlag.
http://dx.doi.org/10.1007/978-0-387-84858-7.
Gao, G., S. Meng and M. WĂźthrich. 2018. Claims Frequency Modeling Using Telematics Car Driving Data. SSRN. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3102371.
Accessed: 29 June 2018.
Gao, G. and M. WĂźthrich. 2017. Feature Extraction from Telematics Car Driving Heatmaps. SSRN. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3070069. Accessed:
June 29 2018.
Goodfellow, I., Y. Bengio and A. Courville. 2016. Deep Learning. MIT Press.
Guo, C. and F. Berkhahn. 2016. "Entity embeddings of categorical variables", arXiv arXiv:1604.06737
Hainaut, D. 2018. "A neural-network analyzer for mortality forecast", Astin Bulletin 48(2):481-508.
Hannun, A., C. Case, J. Casper, B. Catanzaro et al. 2014. "Deep speech: Scaling up end-to-end speech recognition", arXiv arXiv:1412.5567
Hejazi, S. and K. Jackson. 2016. "A neural network approach to efficient valuation of large portfolios of variable annuities", Insurance: Mathematics and Economics 70:169-181.
Hejazi, S. and K. Jackson. 2017. "Efficient valuation of SCR via a neural network approach", Journal of Computational and Applied Mathematics 313:427-439.
Krizhevsky, A., I. Sutskever and G. Hinton. 2012. "Imagenet classification with deep convolutional neural networks," Paper presented at Advances in Neural Information
Processing Systems. 1097-1105.
Kuo, K. 2018. "DeepTriangle: A Deep Learning Approach to Loss Reserving", arXiv arXiv:1804.09253
References (1)
33. Makridakis, S., E. Spiliotis and V. Assimakopoulos. 2018. "The M4 Competition: Results, findings, conclusion and way forward", International Journal of Forecasting
Mitchell, T. 1997. Machine learning. McGraw-Hill Boston, MA.
Noll, A., R. Salzmann and M. WĂźthrich. 2018. Case Study: French Motor Third-Party Liability Claims. SSRN. https://ssrn.com/abstract=3164764 Accessed: 17 June 2018.
Richman, R. 2018. Thoughts on the M4 Conference. http://ronaldrichman.co.za/2018/12/17/thoughts-on-the-m4-conference/. Accessed: 20 March 2019.
Richman, R. and M. WĂźthrich. 2018. "A Neural Network Extension of the Lee-Carter Model to Multiple Populations", Available at SSRN: https://ssrn.com/abstract=3270877
Schreiber, D. 2017. The Future of Insurance. DIA Munich 2017: https://www.youtube.com/watch?time_continue=1&v=LDOhFHJqKqI. Accessed: 17 June 2018.
Shmueli, G. 2010. "To explain or to predict?", Statistical Science:289-310.
Smith, M., F. Beyers and J. De Villiers. 2016. "A method of parameterising a feed forward multi-layered perceptron artificial neural network, with reference to South African
financial markets", South African Actuarial Journal 16(1):35-67.
Sutton, R. and A. Barto. 2018. Reinforcement learning: An introduction, Second Edition. MIT Press.
Wijnands, J., J. Thompson, G. Aschwanden and M. Stevenson. 2018. "Identifying behavioural change among drivers using Long Short-Term Memory recurrent neural networks",
Transportation Research Part F: Traffic Psychology and Behaviour 53:34-49.
Wu, Y., M. Schuster, Z. Chen, Q. Le et al. 2016. "Google's neural machine translation system: Bridging the gap between human and machine translation", arXiv arXiv:1609.08144
WĂźthrich, M. 2018a. "Neural networks applied to chainâladder reserving", European Actuarial Journal 8(2):407-436.
WĂźthrich, M. 2018b. v-a Heatmap Simulation Machine. https://people.math.ethz.ch/~wueth/simulation.html. Accessed: 1 July 2018.
WĂźthrich, M. and C. Buser. 2018. Data analytics for non-life insurance pricing. Swiss Finance Institute Research Paper. https://ssrn.com/abstract=2870308. Accessed: 17 June
2018.
WĂźthrich, M.V. 2017. "Covariate selection from telematics car driving data", European Actuarial Journal 7(1):89-108.
Zarkadoulas, A. 2017. "Neural network algorithms for the development of individual losses." Unpublished thesis, Lausanne: University of Lausanne.
References (2)
34. Get involved
⢠Insurance Data Science conference - 14 June 2019
⢠ETH Zurich
⢠https://insurancedatascience.org/
⢠Amazing line-up of papers, presentations and speakers!
⢠Kasa.ai â launching soon, led by Kevin Kuo of Rstudio
⢠An open research group encouraging innovation in insurance analytics
⢠Some interesting projects planned
35. Thanks for listening - Any questions?
Paper: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3218082
Contact: ron@ronaldrichman.co.za