Did you buy a gold jewelry a few months
back and now cursing why you ever did?! Do you
think there should have been some sort of a crystal
which should have predicted this fall in the
price of gold? The tech world is working on such a
machine which is called surprise modeling.
So what does surprise modeling do? It
works on the principle that by analyzing the things
in the past which have brought a surprise, it’s possible
to model a system that would predict any
surprises in the future. A system to function surprise
modeling must be able to do two things,
firstly it must have a knowledge of what the user
considers surprising and secondly must be able to
predict it.
So what is a surprise? A surprise occurs in
the case of uncertainty, i.e. when some information
is missing or is incomplete. And surprise is a
relative matter, what may surprise one person
may not surprise another. In probability and decision
theory it can be shown that, under a small set
of axioms, the only consistent way for modeling
and reasoning about uncertainty is provided by
the Bayesian theory of probability.
For a given set of events in a model M, we
have the prior probabilities. So, from the new set
of events, we calculate the new probabilities
(posterior probabilities). They both are compared
(prior and posterior) and if found equal, there is
no element of surprise.Else the surprise is measured
using Kullback-Leibler (KL) divergence.
In sum, unsurprising data yields little
difference between posterior and prior distributions
of beliefs over models, while surprising
data yields a large shift: in mathematical terms,
an event is surprising when the distance between
posterior and prior distributions of beliefs
over all models is large.
Let us consider, the example of trying to
analyze whether there would be a surge in traffic
in a particular area. This is being done at Seattle.
They analyze traffic scenarios for every 15
minutes and came up with the probability distribution
of traffic for each segment. They also
have a database of how traffic was influenced by
various factors such as rain, storm, heat wave, a
holiday and even high profile visits.
With such information on hand, they
find out any deviations if they exist using The
Bayesian theory and thus inform users about the
sudden and unexpected fluctuations in traffic.
Thus, who knows the day is not far when we
may log on to our systems to know when it is
profitable to buy gold.
back and now cursing why you ever did?! Do you
think there should have been some sort of a crystal
which should have predicted this fall in the
price of gold? The tech world is working on such a
machine which is called surprise modeling.
So what does surprise modeling do? It
works on the principle that by analyzing the things
in the past which have brought a surprise, it’s possible
to model a system that would predict any
surprises in the future. A system to function surprise
modeling must be able to do two things,
firstly it must have a knowledge of what the user
considers surprising and secondly must be able to
predict it.
So what is a surprise? A surprise occurs in
the case of uncertainty, i.e. when some information
is missing or is incomplete. And surprise is a
relative matter, what may surprise one person
may not surprise another. In probability and decision
theory it can be shown that, under a small set
of axioms, the only consistent way for modeling
and reasoning about uncertainty is provided by
the Bayesian theory of probability.
For a given set of events in a model M, we
have the prior probabilities. So, from the new set
of events, we calculate the new probabilities
(posterior probabilities). They both are compared
(prior and posterior) and if found equal, there is
no element of surprise.Else the surprise is measured
using Kullback-Leibler (KL) divergence.
In sum, unsurprising data yields little
difference between posterior and prior distributions
of beliefs over models, while surprising
data yields a large shift: in mathematical terms,
an event is surprising when the distance between
posterior and prior distributions of beliefs
over all models is large.
Let us consider, the example of trying to
analyze whether there would be a surge in traffic
in a particular area. This is being done at Seattle.
They analyze traffic scenarios for every 15
minutes and came up with the probability distribution
of traffic for each segment. They also
have a database of how traffic was influenced by
various factors such as rain, storm, heat wave, a
holiday and even high profile visits.
With such information on hand, they
find out any deviations if they exist using The
Bayesian theory and thus inform users about the
sudden and unexpected fluctuations in traffic.
Thus, who knows the day is not far when we
may log on to our systems to know when it is
profitable to buy gold.
0 Responses to "Surprise Modeling"
Post a Comment