Você está na página 1de 4

1.

The Principles of Scheduling


If we have a number of tasks to complete and know how long each will take, to minimize the total time
all jobs remain unprocessed, we should do the shortest jobs first.
If we have a number of tasks, each of which has a deadline by which it must be completed, then if it is
possible to schedule the jobs in such a way that they are all completed in time, then this can be
achieved by doing jobs with the nearest deadline first.
These results are very intuitive and we would expect them to be true in most cases. The critical part of
this is that we have proofs that they do in fact hold in all cases and can thus rely on these and can also
use these as heuristics (solutions that are good, but not necessarily the best) in more complicated cases.
This principle has applications in many different situations in the real world. In everyday life, if you
have to do a whole lot of tasks, doing the short ones first can make it seem much more manageable, as
you quickly only have to do one or two jobs. In industry, the applications are endless. For example, a
city runs a system of buses for public transportation. From time to time, these buses may need
maintenance, which varies from a quick cleaning to a transmission replacement. If these buses are
processed in the order they arrive, buses that need a regular cleaning will be stuck behind ones that
need a transmission replacement that takes a long time. Soon the city will not have enough buses to run
their transit system. However, if the jobs are processed in order of how much time they will take, this
problem disappears. Only a few buses will not be running at any given time.
2. In a convex feasible region a local minimum is a global minimum
A convex set can be thought of as a set where you can see any point from any other point.

The first set is convex, while the second one is not. This is an example of a two dimensional set, but
such sets can exist in any number of dimensions. Furthermore, you can define a continuous function
over such a set. In any function, a local maximum means that at a point, say x, f(x) is bigger than it is at
its neighboring points, say x-a and x+a, for small values of a. A global maximum means that at a point,
say x, f(x) is bigger than it is at any other point y. A local and global minimum is similarly defined. In a
convex set, since you can see any point from any other point, a local minimum is always a global
minimum. While this seems an underwhelming result, it is very useful. It is easy to show that a Linear
Programming Problem (discussed below) always produces a convex space and thus a local solution to
an LPP is also a global solution to the LPP.
3. The Simplex Method solves a Linear Programming Problem
The easiest way to explain a Linear Programming Problem is by an example. Suppose we have 5lbs of
wheat and 6 pounds of sugar and 2 gallons of milk. We can make biscuits or bread with these
ingredients. A loaf of bread required a pound of wheat, a pound of sugar and half a gallon of milk. A

batch of biscuits required half a pound of wheat, a pound and a half of sugar and half a gallon of milk.
A batch of biscuits sells for $15 and a loaf of bread sells for $10. What is the optimal number of loaves
of bread and batches of biscuits to make to maximize the money we make?
More generally, a LPP is any problem where we are trying to maximize or minimize some number (in
this case, our profit) subject to some constraints (the amount of wheat used has to be less than 5lbs etc.),
and all variables are of the first power.
The Simplex method relies on the fact that an optimal solution is always found on a corner of the
feasible region. In short, starting at any corner, check if the corners next to it have a better optimal
value. If they do, move to the one with the best value. If not, they you are at the optimal solution, since
the set is a convex set (discussed above) and a local maximum/minimum is a global
maximum/minimum.

(Alex Jones, Data Analytics for Business Leaders Explained)


4. Discrete Time Markov Chain
A Discrete Time Markov Chain is a way of modeling a process which has discrete or separate states,
and probabilities of transitioning between them. For example, if a person has the states {sick, well,
recovering}, he might have a small probability, say .01 of getting sick every day. For each day he is
sick, he will have some probability of starting to recover, and some probability of staying sick. If he is
starting to recover, he might get well, or he might get worse again (if for example, he does not eat

enough, or get enough rest). We call this a DTMC.


Well
Well
.99
Sick
0
Recovering
.30

Sick
.01
.70
.20`

Recovering
0
.30
.50

DTMCs are easy to study and we can figure out a large amount of information from them. For example
the long term average of how long a person will be sick or well, or how much time we would expect it
to take for a person who is sick to become well. DTMCs have a wide variety of applications in a
number of fields. Google PageRank, the algorithm behind the Google search engine is another such
example. The pages are the states and the probabilities are based on the hyperlinks between pages.

(https://sgmustadio.wordpress.com/2011/02/17/primer-on-hidden-markov-models-part-1/)

5. Lack of Memory of Exponential Distribution


The Exponential Distribution is a probability distribution, with some average value (1/).

(https://en.wikipedia.org/wiki/File:Exponential_cdf.svg)
A very important property of the distribution is that it is memoryless. This is best described by example.
How often you receive a text message on your phone is a memory distribution (with mean depending

on how popular you are). Say on average, you receive a text message every 30 minutes. Just because
you hadnt received a text message for 29 minutes, you would not expect to receive one soon. It is just
as likely that you will receive a text message 5 minutes after the previous one as it is 30 minutes after.
This is clearly true because none of your friends are keeping track of how often you receive these texts.
This lack of memory property makes using this distribution in more complex models very easy, without
losing sophistication in the model. We use this model to describe waiting times in applications ranging
from queues (for example, at the DMV or in a bank or in manufacturing), to radioactive decay.

Você também pode gostar