If you're wondering, a simple way to solve the limit is to take the log of the expression, compute the limit of that logarithm and then take the exponential of the limit. You can use the fact that log(n!) goes to infinity as nlogn The rest of the computation is trivial
To derive the formula at 26:07 (Poisson distribution) from 25:06 (Binomial distribution), you need to write combination as factorials, then use stirling approximation for factorials and the definition of e
Even Professor Tsitsiklis might stuck in the middle of the proof of Poisson distribution PMF...but the proof is perfectly written in his textbook, which is the BEST by far I've ever read in intro level probability and statistics. It took me a long time to fully understand Poisson process...thanks to MIT OCW and Professor Tsitsiklis.
sacredsoma The lector doesn't explain it, but in that part he is working with one of the several definitions of the Poisson process, in which the process is defined as a Lévy process. Deep knowledge of mathematical analysis, specifically measure theory, is required to understand these infitesimal properties of Lévy processes, that's why he doesn't say much about those second order terms.
I'm wondering, at 14:15, when Prof. is discussing arrival independence (second bullet point), how can they be truly "independent," if this is being modeled by a PMF that must sum to 1. In other words, don't we create a type of dependency exactly when an event has occurred ?
Although it is true intuitively that the T2, T3,.... all will be independent exponential random variable but the pdf for K=2,3,4 is not exponential...How can you argue about that??? anyone?
process has memorylessnessf property (provided the person entered the room doesn't see the future) continuous time variations definition of the poisson process and we use the concept of memorylessness in here time homogenity as time is continuous. Lambda*Tau will be the rate at very small interval diff time slots are independent of each other NUMBER of arrivals in the disjoint time interval will be independent of each other. delta is very small. pmf have the binomial pmf here we are taking a big interval and we are splitting that interval into many infinitesemely small interval. now prob of one of the intervals will be same as the bernoulli distriubution for poissson we take lambda tends to 0. merging of the poisson process
I am having a hard time making sense of the notation at 11:22. I believe the notation should be the conditional probability P(k|t) rather than P(k,t). I interpreted the latter to be the joint probability and if it is the case, the summation over all k of P(k,t) given a fixed t could not be equal to 1. Anyone, please help knock some sense to my head!
@@nguyentin6294 as mentioned in the video. k is a random variable, i.e. the summation over all k is 1. "t" just specifies omega, the set of all k's. it is only a parameter and not an event. Or otherwa around ask yourself if t would be an event what would P(t) be? it simply doesnt make sense to ask this, without specifing how much arrivals we are talking about.
I doubt that my comment will be useful to you at all since it's been 7 years, but I'll comment anyway for people who are going to view this lecture in the future. The second order terms exist because P(k,d) isn't 0 when k >=2. d here stands for delta. Regardless of how small an interval you pick, it is possible for more than one successes to occur in the interval. That probability, however, will converge to 0 much "faster" than d, so we can ignore those probabilities since they are trivial. This is an important assumption to use the binomial theorem, since the binomial theorem only works when there are two possible outcomes - success and failure. If we allow multiple successes, the theorem cannot be used. Of course, it follows that any distribution which doesn't satisfy the property that the probability of multiple successes in a given interval doesn't converge to 0 "faster" than the interval itself cannot be treated as a Poisson Distribution. In our course notes at University of Waterloo, this property is referred to as "individuality".
I really enjoyed this series of lectures just to refresh my memory. Thanks.But I am also surprised that A MIT professor of statistics can stump on such a simple task of deriving the Poisson PDF.
He wrote the textbook which contains the detailed derivation procedures. He would probably be able to figure out how to derive the final Poisson formula if he had spent perhaps some more minutes. But the professor decided to leave the students to read the textbook themselves. By the way, some people do have the ability to memorize all details once they take a look. But that is a painful disease.
If you're wondering, a simple way to solve the limit is to take the log of the expression, compute the limit of that logarithm and then take the exponential of the limit. You can use the fact that log(n!) goes to infinity as nlogn The rest of the computation is trivial
To derive the formula at 26:07 (Poisson distribution) from 25:06 (Binomial distribution), you need to write combination as factorials, then use stirling approximation for factorials and the definition of e
Well said! But, actually, stirling is not a necessity. Since when n -> infinity, we have (n k) = (n * (n-1) * ... * (n - k + 1)) / k! -> n^k / k!
you dont need stirling appoximation for that
Thanks 🙏🏻
This is a fantastic lecture on Poisson Process and way they come from, probably the best
he teaches excellently! make complicated things easy to understand, that's the magic!
Even Professor Tsitsiklis might stuck in the middle of the proof of Poisson distribution PMF...but the proof is perfectly written in his textbook, which is the BEST by far I've ever read in intro level probability and statistics. It took me a long time to fully understand Poisson process...thanks to MIT OCW and Professor Tsitsiklis.
Hi where I can find his textbook please?
@@elmrabti amazon
@@nadekang8198 Do you have it?
Ένα μεγάλο ευχαριστώ, από Έλληνες φοιτητές!
You are a star, dear lecturer.
sacredsoma The lector doesn't explain it, but in that part he is working with one of the several definitions of the Poisson process, in which the process is defined as a Lévy process. Deep knowledge of mathematical analysis, specifically measure theory, is required to understand these infitesimal properties of Lévy processes, that's why he doesn't say much about those second order terms.
Thanks for that clarification. I wrote on my notes that this part needed some more explanation.
Begin's poisson at 9:20
I'm wondering, at 14:15, when Prof. is discussing arrival independence (second bullet point), how can they be truly "independent," if this is being modeled by a PMF that must sum to 1. In other words, don't we create a type of dependency exactly when an event has occurred ?
my probability professor skipped poisson distribution saying it is not so important.
haha
Although it is true intuitively that the T2, T3,.... all will be independent exponential random variable but the pdf for K=2,3,4 is not exponential...How can you argue about that??? anyone?
Excellent explanation !
51:46 he probably meant "this plus that" instead of "this plus that".
too much information to grasp in one go.. woah nice lecture!
process has memorylessnessf property (provided the person entered the room doesn't see the future)
continuous time variations
definition of the poisson process and we use the concept of memorylessness in here
time homogenity as time is continuous.
Lambda*Tau will be the rate at very small interval
diff time slots are independent of each other
NUMBER of arrivals in the disjoint time interval will be independent of each other.
delta is very small.
pmf have the binomial pmf
here we are taking a big interval and we are splitting that interval into many infinitesemely small interval.
now prob of one of the intervals will be same as the bernoulli distriubution
for poissson we take lambda tends to 0.
merging of the poisson process
thank you so much sir,you made learning very easy....
Why probability of arrival for k>1 in time delta is zero.?
wonderful lecture
master piece of video
I enjoyed watching this maths lecture as watching a movie
Not even a chuckle at the "B movie" joke?
I am having a hard time making sense of the notation at 11:22. I believe the notation should be the conditional probability P(k|t) rather than P(k,t). I interpreted the latter to be the joint probability and if it is the case, the summation over all k of P(k,t) given a fixed t could not be equal to 1. Anyone, please help knock some sense to my head!
t is a different nature and not a random variable.
@@schobihh2703 could you clarify it a bit more?
@@nguyentin6294 as mentioned in the video. k is a random variable, i.e. the summation over all k is 1. "t" just specifies omega, the set of all k's. it is only a parameter and not an event. Or otherwa around ask yourself if t would be an event what would P(t) be? it simply doesnt make sense to ask this, without specifing how much arrivals we are talking about.
@@schobihh2703 thank you for the insight :)
Prof John's lecture on Poissan PMF is available at: th-cam.com/video/eTLa9GDID8A/w-d-xo.html
what is the reason behind considering second order terms of delta? since it is just an interval.
I dont get the bit @ 15:50 about small interval probabilities being a limiting case. Why are there any second order terms?
I doubt that my comment will be useful to you at all since it's been 7 years, but I'll comment anyway for people who are going to view this lecture in the future.
The second order terms exist because P(k,d) isn't 0 when k >=2. d here stands for delta.
Regardless of how small an interval you pick, it is possible for more than one successes to occur in the interval. That probability, however, will converge to 0 much "faster" than d, so we can ignore those probabilities since they are trivial.
This is an important assumption to use the binomial theorem, since the binomial theorem only works when there are two possible outcomes - success and failure. If we allow multiple successes, the theorem cannot be used.
Of course, it follows that any distribution which doesn't satisfy the property that the probability of multiple successes in a given interval doesn't converge to 0 "faster" than the interval itself cannot be treated as a Poisson Distribution. In our course notes at University of Waterloo, this property is referred to as "individuality".
@@aryanparikh9085 If possible, can you share the notes of this particular characteristic for qualifying a distribution as Poisson?
good explaination though a little more rigorous derivation wud have helped lot... intended for overview
the course is applied probability, the point is not rigorous derivation
I would look into a textbook for complete rigorous derivations.
This is amazing
I really enjoyed this series of lectures just to refresh my memory. Thanks.But I am also surprised that A MIT professor of statistics can stump on such a simple task of deriving the Poisson PDF.
Anyone can stump on something once in a while.
He wrote the textbook which contains the detailed derivation procedures. He would probably be able to figure out how to derive the final Poisson formula if he had spent perhaps some more minutes. But the professor decided to leave the students to read the textbook themselves. By the way, some people do have the ability to memorize all details once they take a look. But that is a painful disease.
Περίφανος που βλέπω κι άλλον Έλληνα να διδάσκει εδώ
Πώς ξέρετε ότι είναι Έλληνας; Αγνοήστε τα λάθη, χρησιμοποιώ λογισμικό μετάφρασης.
@@lekokotonteso Tsitsiklis is a greek surname. Furthermore focus on his accent
hahaha, this guy does not get ready for lectures hahah. so funny..