WEBVTT
00:00:00.820 --> 00:00:04.030
We now start our discussion
of stochastic processes
00:00:04.030 --> 00:00:07.580
by starting with the simplest
stochastic process there is.
00:00:07.580 --> 00:00:10.160
This is the so-called
Bernoulli process,
00:00:10.160 --> 00:00:14.020
which is nothing but a sequence
of independent Bernoulli
00:00:14.020 --> 00:00:15.310
trials.
00:00:15.310 --> 00:00:19.570
We let Xi stand for the
random variable that
00:00:19.570 --> 00:00:22.700
describes the result
in the i trial.
00:00:22.700 --> 00:00:26.460
The assumptions that we
make are that at each trial
00:00:26.460 --> 00:00:31.360
there is a certain probability,
p, that the trial results in 1.
00:00:31.360 --> 00:00:33.840
And in that case, we
usually say that there
00:00:33.840 --> 00:00:36.090
is a success at the ith trial.
00:00:36.090 --> 00:00:38.310
And the remaining
probability, 1 minus p,
00:00:38.310 --> 00:00:42.070
is assigned to the possibility
that the random variable Xi
00:00:42.070 --> 00:00:44.310
takes a value of
0, in which case
00:00:44.310 --> 00:00:46.950
sometimes we say that
there was a failure.
00:00:46.950 --> 00:00:49.330
Now, to keep things
nontrivial, we
00:00:49.330 --> 00:00:53.390
will always assume that p is
a number strictly between 0
00:00:53.390 --> 00:00:57.620
and 1, because otherwise, in the
extreme cases of p equal to 0
00:00:57.620 --> 00:01:00.165
or p equal to 1, there
isn't really any randomness.
00:01:02.700 --> 00:01:05.930
This process is something
that we have already seen.
00:01:05.930 --> 00:01:09.130
We have worked plenty
of examples involving
00:01:09.130 --> 00:01:13.210
repeated Bernoulli trials
or repeated coin flips.
00:01:13.210 --> 00:01:17.730
We have solved several problems,
we have seen several formulas.
00:01:17.730 --> 00:01:21.030
Here we will recapitulate
some of them.
00:01:21.030 --> 00:01:23.700
But we will also start
looking at the process
00:01:23.700 --> 00:01:26.910
from a new point of view.
00:01:26.910 --> 00:01:30.680
Before continuing, let me
emphasize the assumptions that
00:01:30.680 --> 00:01:32.950
come into the Bernoulli process.
00:01:32.950 --> 00:01:36.810
One key assumption is
that we have independence.
00:01:36.810 --> 00:01:40.920
The different trials
are independent.
00:01:40.920 --> 00:01:43.150
The second assumption
that we make
00:01:43.150 --> 00:01:46.420
is that the model
is time homogeneous.
00:01:46.420 --> 00:01:49.009
What I mean by this is
that this probability
00:01:49.009 --> 00:01:54.170
p of success at a given trial
is the same for all trials.
00:01:54.170 --> 00:01:57.229
It does not depend on i.
00:01:57.229 --> 00:02:01.920
So this process is as simple as
a stochastic process could be.
00:02:01.920 --> 00:02:05.360
But nevertheless, it
can be used as a model
00:02:05.360 --> 00:02:07.080
in various situations.
00:02:07.080 --> 00:02:09.460
Sometimes it's clear that
we're dealing with a Bernoulli
00:02:09.460 --> 00:02:13.160
process, but sometimes it also
shows up in unexpected ways.
00:02:13.160 --> 00:02:16.400
In any case, the first simple
example could be the following.
00:02:16.400 --> 00:02:19.920
Every week you play the
lottery, and either you win
00:02:19.920 --> 00:02:21.250
or you do not to win.
00:02:21.250 --> 00:02:23.920
And assuming that it is
the same kind of lottery
00:02:23.920 --> 00:02:27.040
that you play each
week, the constant p
00:02:27.040 --> 00:02:30.100
would be the same, the
probability of success.
00:02:30.100 --> 00:02:33.690
And assuming a lottery that
is not rigged in any way,
00:02:33.690 --> 00:02:35.730
whether you win on
one week, should
00:02:35.730 --> 00:02:39.570
be independent from what
happens in other weeks.
00:02:39.570 --> 00:02:41.350
Quite often, the
Bernoulli process
00:02:41.350 --> 00:02:45.720
is used as a model of
arrivals, in which case,
00:02:45.720 --> 00:02:47.740
instead of saying
probability of success,
00:02:47.740 --> 00:02:50.280
we would say probably
of an arrival.
00:02:50.280 --> 00:02:53.170
The idea is that
time is slotted,
00:02:53.170 --> 00:02:55.970
let's say, in
seconds, for example.
00:02:55.970 --> 00:02:59.420
And each second you're sitting
at the entrance of a bank
00:02:59.420 --> 00:03:02.430
and you make a note
whether somebody
00:03:02.430 --> 00:03:04.620
came into the
bank, in which case
00:03:04.620 --> 00:03:08.420
we have an arrival or success,
or whether no one came
00:03:08.420 --> 00:03:11.760
during that time interval.
00:03:11.760 --> 00:03:15.560
If one here believes
that different slots,
00:03:15.560 --> 00:03:18.320
different seconds, are
independent of each other,
00:03:18.320 --> 00:03:21.140
then you do have a
Bernoulli process.
00:03:21.140 --> 00:03:23.890
It might not be an
exactly accurate model,
00:03:23.890 --> 00:03:25.780
but it is a good
first approximation
00:03:25.780 --> 00:03:29.260
to start working with a
model of this situation.
00:03:29.260 --> 00:03:32.540
Similarly, if you have a
server, a computer, that
00:03:32.540 --> 00:03:36.579
takes jobs to process and
jobs are coming randomly,
00:03:36.579 --> 00:03:38.940
you divide time into
slots, and during each slot
00:03:38.940 --> 00:03:42.090
a job might arrive
or might not arrive.
00:03:42.090 --> 00:03:45.480
And as a first approach
to a model of this kind,
00:03:45.480 --> 00:03:49.950
you might as well employ
the Bernoulli process.
00:03:49.950 --> 00:03:53.100
A final note, why is this
process called the Bernoulli
00:03:53.100 --> 00:03:54.380
process?
00:03:54.380 --> 00:03:58.340
Well, the name comes
from a famous family
00:03:58.340 --> 00:04:01.310
of Swiss mathematicians,
the Bernoulli family.
00:04:01.310 --> 00:04:04.420
And one of them,
Jacob Bernoulli,
00:04:04.420 --> 00:04:07.260
did many contributions to
many branches of mathematics.
00:04:07.260 --> 00:04:09.180
But an important
one was in the field
00:04:09.180 --> 00:04:11.190
of probability,
where he actually
00:04:11.190 --> 00:04:17.920
derived quite deep results
on a sequence of what we now
00:04:17.920 --> 00:04:20.370
call Bernoulli trials.