Hello everyone and welcome to the voiceover of lecture 4. The topic for today is supervised
machine learning. A machine, an algorithm is learning when it becomes better and better at
doing a task when it receives additional data.
FSA Lecture 4 1
, DATA-DRIVEN AND AI: LEARNING PERSONALIZATION
“KBC has translated the customer “Customers can ask Kate questions
journey (from data analysis to regarding their basic financial
trigger and then solution) into an transactions (money transfer, insurance
AI-powered personal digital claim, etc.). But customers will also
assistant called Kate who gradually receive regularly personalised proposals
becomes more accurate or what from KBC in their mobile app to ensure
you might call smarter over time maximum convenience. They are
and is therefore continuously completely free to choose whether or not
improving the user's overall to accept a proposal. If they do, the
customer experience.” solution will be offered and processed
completely digitally.”
https://www.kbc.com/en/newsroom/innovation/innovation-2020.html 2
An example of supervised machine learning from a banking app is the digital assistant Kate at
KBC. Kate is a digital assistant that tries to predict what you as client like to purchase as a
service from KBC, through your relation with KBC. KBC collects data about you and from that
data can KBC predict what you would like to buy from them. As they get to know you better
and better, that is as they collect more and more data about you, the algorithm will perform
better and better. That's what they also write on their website that Kate gradually becomes
more accurate or what you might call smarter overtime. And thus by knowing you better they
can make better personalized proposals which is beneficial to KBC because they have more
success from doing proposal and also a beneficial to you because your user experience will
improve.
FSA Lecture 4 2
,More formally you have here a definition of machine learning. A computer program is learning
from data with respect to a prediction task and some performance measure if its performance
on that task as measured by this performance measure improves with increasing availability
of data. So machine learning can be quantified, so we have to be very precise in terms of what
is the task that the algorithm needs to perform? How do we evaluate, this is a performance
measure. And as we give more data to the algorithm, the algorithm should get better and
better in terms of improving the performance measure. So in machine learning we are focused
mainly on prediction of a certain outcome, could be serious price of a house, could be the
stock return, could be the next service your client would like to purchase, … So that would be
the prediction task. The input data can be whatever feature that is useful to do this prediction
and the performance measure is the fact that what you predict and the actual outcome should
be as close as possible. So in this lecture we will use for this the mean squared error as
measure of prediction accuracy, it is the average squared prediction error.
FSA Lecture 4 3
, So the focus is on prediction and prediction happens through into data, the features. Features
are mapped to the prediction using a mapping function f. This function f must be computable
by a machine and therefore it's also called an algorithm which is a function that can be
computed by a machine. The function typically has parameters that need to be estimate and
in machine learning we call the task of estimation: training of the algorithm.
FSA Lecture 4 4
machine learning. A machine, an algorithm is learning when it becomes better and better at
doing a task when it receives additional data.
FSA Lecture 4 1
, DATA-DRIVEN AND AI: LEARNING PERSONALIZATION
“KBC has translated the customer “Customers can ask Kate questions
journey (from data analysis to regarding their basic financial
trigger and then solution) into an transactions (money transfer, insurance
AI-powered personal digital claim, etc.). But customers will also
assistant called Kate who gradually receive regularly personalised proposals
becomes more accurate or what from KBC in their mobile app to ensure
you might call smarter over time maximum convenience. They are
and is therefore continuously completely free to choose whether or not
improving the user's overall to accept a proposal. If they do, the
customer experience.” solution will be offered and processed
completely digitally.”
https://www.kbc.com/en/newsroom/innovation/innovation-2020.html 2
An example of supervised machine learning from a banking app is the digital assistant Kate at
KBC. Kate is a digital assistant that tries to predict what you as client like to purchase as a
service from KBC, through your relation with KBC. KBC collects data about you and from that
data can KBC predict what you would like to buy from them. As they get to know you better
and better, that is as they collect more and more data about you, the algorithm will perform
better and better. That's what they also write on their website that Kate gradually becomes
more accurate or what you might call smarter overtime. And thus by knowing you better they
can make better personalized proposals which is beneficial to KBC because they have more
success from doing proposal and also a beneficial to you because your user experience will
improve.
FSA Lecture 4 2
,More formally you have here a definition of machine learning. A computer program is learning
from data with respect to a prediction task and some performance measure if its performance
on that task as measured by this performance measure improves with increasing availability
of data. So machine learning can be quantified, so we have to be very precise in terms of what
is the task that the algorithm needs to perform? How do we evaluate, this is a performance
measure. And as we give more data to the algorithm, the algorithm should get better and
better in terms of improving the performance measure. So in machine learning we are focused
mainly on prediction of a certain outcome, could be serious price of a house, could be the
stock return, could be the next service your client would like to purchase, … So that would be
the prediction task. The input data can be whatever feature that is useful to do this prediction
and the performance measure is the fact that what you predict and the actual outcome should
be as close as possible. So in this lecture we will use for this the mean squared error as
measure of prediction accuracy, it is the average squared prediction error.
FSA Lecture 4 3
, So the focus is on prediction and prediction happens through into data, the features. Features
are mapped to the prediction using a mapping function f. This function f must be computable
by a machine and therefore it's also called an algorithm which is a function that can be
computed by a machine. The function typically has parameters that need to be estimate and
in machine learning we call the task of estimation: training of the algorithm.
FSA Lecture 4 4
