titanlab.org

Don’t Teach!

I was reading some PhD guidelines and happened to come across a very meaningful remark. I felt exciting and decided to copy the inspiring paragraph here.

Don’t teach!

. . . more than you have to. For many, teaching is attached to a stipend or is otherwise economically unavoidable. In this case, do what you must! Moreover, there are some real intellectual and practical advantages from doing a couple of terms of TA work. Explaining the concepts to others is very useful in consolidating them in yourself. But beyond this, the returns become strongly negative. Your job is research – and anything that distracts you from this is a heavy cost. The first cost, which may seem remote at the time that you are deciding on the teaching, is that it could delay completion of the thesis by a year or more. An even larger cost is if it crowds out time to write a really great thesis. As a PhD student, your time is very valuable; treat it that way.

Ph.D. Thesis Research: Where do I Start? (PDF) 

Demystify Convolution

There are may ways to explain convolution. But I prefer this example.

# 1

Imagine you are throwing a stone into a lake.

As the stone hit to the lake, you will see some ripples on the surface of the lake.

That is the response of the lake towards your projection of stone.

# 2

Now, let say we want to model this.

Your stone is an unit impulse ( ) and the lake  is a transfer function .

We don’t what exactly is yet.

However, when your stone hits into the lake , we see a ripple or response . That ripple is the response of the lake towards your stone.

Or, mathematically,  convolute with and we see an output something like the figure below.

(Of cause, the response of the lake is not a pure rectangular wave form. This is just for our ease of illustration.)

Read more…

The Random Variables and Random Process

Bucket of Apples

Someone ask me to explain the different between random variables and random process. Here are my explanation.

Say we are a wholesaler of apples in a city. One day a worker moves down a bucket of apples from a truck. By looking at the apples in this bucket, we can measure the expected weight and variation of apples in this bucket. Let denote the expected weight as , in which it is a random variable that possesses finite mean and variation.

Then, the worker moves down another bucket of apples. We not sure if this bucket of apples possess the same statistical properties as the previous bucket. By intuition, we know these apples in the second bucket possesses a finite expected weight as well (or you will see a HUGE apple!!). Next, the worker continue to move down a lot of buckets of apples from the truck (big truck!). For convenience, we denote the expected weight as where denotes the sequence where the worker moves down the buckets.

The manager of the wholesaler is not interested at the weight distribution of apples in any particular bucket but among certain groups of buckets. For example, he wants to know if the first bucket has worm, what is the chance that the nth bucket has worm as well. This question the inter-relation among the buckets. So, we have no choice but to extend the notion “a single bucket” into “groups of buckets”. In mathematics, we write this as and call this the random process (or stochastic process) of X. In such a way, we can describe the correlation (or other statistical properties) between the first bucket and the second bucket or any bucket in the later sequence to the manager.

As for the summary, in random process (stochastic process) we are dealing with a group of “homogeneous” random variables in the sense that they are of the same function (e.g. apples) but they may possess different distribution function.

[Math] My Math Note

Taken somewhere from Internet.

Variance, covariance and autocovariance:

• Variance is one of the parameter to describe the distribution. It tells how the distribution extends from the mean value.
• Moments are always used to describe the distribution due to its computational simplicity (i.e. the power series). 1st moment is mean value, 2nd moment is variance, 3rd moment is skewness.
• Let and denote two random variables. Covariance is a measure of how much and change together.
• In this case, variance is a special case of the covariance when two variables are identical.
• Given a stochastic process , the autocovariance is the covariance of againsts itself in a time-shifting. That is .

Correlation and Autocorrelation:

• Correlation is the measure of the strength of association between two variables. For example, there is a strong correlation between black eyes and and black hair as you may seem them moving together all the time. But, there is a weak correlation between blue wallpaper and puppies. Bear in mind that correlation is not causation. Though the death of babies increase with the cases of vehicle accidents. They have a strong correlation. But we cannot say the death of babies causes the happen of vehicle accidents. It does not make sense. [Taken from Yahoo! Answer]
• Autocorrelation is the cross-correlation of the variable itself, in which separated by certain amount of time delay.

Random Variable:

• Say we run an experiment in which it gives a finite number of possible outcome, named as sample space .
• For the convenience of mathematical calculation, each outcome (denoted as ) of is translated into a real number (if possible) and that translation process is given by a function where . We name as random variable and most of the time we just denote it as if the context is well understood.

Classification of data.

Stationary Random Process:

• Let denote a random process, in which and denote its mean value and autocorrelation function.
• is non-stationary if and vary as varies.
• Otherwise, is said stationary or weakly stationary in wide-sense.
• For weakly stationary process, and .
• For strongly stationary process, all possible moments and join moments are time-invariant.