Search This Blog

Tuesday, March 22, 2016

GATE 2016: Problem with CSE key

I have never been fond of GATE exam. In fact, this is a terrible exam and I have written about it several times on this blog. So often one finds ambiguous questions. This blog is to report one such ambiguous question in GATE 2016 Computer Science and Engineering paper.

Q. Which one of the following protocols is NOT used to resolve one form of address to another one?

(A) DNS
(B) ARP
(C) DHCP
(D) RARP

The official answer is (C).

Obviously, the question was set by someone who does not understand protocols. They would have checked some book which would have said that Dynamic Host Configuration Protocol is used to dynamically assign IP addresses to hosts. They would have assumed that DHCP can only do this, and nothing else. In particular, they would have assumed that DHCP is NOT used to resolve one form of address to another one. But, of course, DHCP can very much be used to provide static addresses by using MAC address as one possible identity of the node seeking an IP address.

So far, so good. I guess that the committee that sets up a question paper can not be a very large committee in the interest of secrecy. And hence one cannot possibly have an expert from each sub-discipline of Computer Science. And therefore, such misunderstandings can happen. And this is really the cause of ambiguous questions in GATE, for which there is no easy solution.

I was hoping that at least after such an ambiguity has crept in, and they have been informed about it, they will take appropriate measures to reduce the impact of such an error on students. But, unfortunately, at least in this case, they decided to ignore the appeal, which is very strange to me.

7 comments:

Nagendra said...

I fully agree that mistakes in exam question papers, especially a public exam like this taken by a large number of students is very bad. The number of question paper setters and checkers is large enough, and there is even a post-exam check, to cover all areas and filter out such mistakes or make corrections to scoring if it is detected post-facto. Not sure what happened here.

But I am curious why you characterize GATE as a terrible exam. Do you feel that there shouldn't be such a common exam for PG entrance and other assorted purposes? or that GATE isn't correctly designed for such a purpose? My own feeling is that the exam is well enough designed (within the limitations of any such exams). The reduced number of questions and addition of some common aptitude questions have improved it in the last few years. Also, we have found that PG performance (GPA) of students is well correlated with the GATE scores with which they entered. (None of this is to claim that someone with GATE score of 900 is better than 895, or that there are no outliers at all; But given that a line has to be drawn as cutoff for admitting PG students, it is a good enough one. Also, my experience is with EC, not CS, but can't imagine that the large branches are all that different from each other.).

Nagendra

Dheeraj Sanghi said...

@Nagendra, GATE is terrible for the purpose it is designed. Can we say with reasonable statistical validity that students with 99.5 percentile are more likely to perform better in our MTech program than students with 99.0 percentile. Most people that I have worked with would say that GATE is a random process of ranking students and the only advantage of GATE is that we get a method to shortlist which can not be challenged in a court. (I am only commenting on CS paper, of course.) If you look at reliability of any assessment, it is severely impacted if the questions are ambiguous.

The question paper in CS in the past have been so bad (I haven't looked at in the last couple of years) that on an average a 12th class student should be able to score higher than an average 4th year student in CS. The 12th class student will only need to be trained to not attempt CS questions and avoid negative marks. This partly reflects the quality of CS education in colleges, and partly reflects the quality of GATE exam.

One can get into top 1 percentile without solving even one question in vast areas of computer science. In fact, one needs to be good in only 3-4 courses out of 20-25 courses one would have done in a computer science program. This would be alright if the purpose of the exam was to test basic CS knowledge only and encourage migration of non-CS students to MTech (CS), but it is not OK if the purpose was to check overall CS preparedness of the student. (And if the purpose was to encourage migration, they should allow students to give GATE in two subjects.)

But the bottomline is this. If you tell me that your GATE score is 80 percentile in CS, I know you have a poor background in CS. But if you tell me that your GATE score is 99 percentile in CS, I am not sure if you have a good CS background or not. What is the point of an exam which declares you to be in the top 1 percent of the country, and yet it gives one no guarantee in terms of CS preparedness.

Nagendra said...

Can we say with reasonable statistical validity that students with 99.5 percentile are more likely to perform better in our MTech; ... if you tell me that your GATE score is 99 percentile in CS, I am not sure if you have a good CS background.

Certainly not. I am not sure if this type of general exam can have that kind of resolution. For that, an exam specifically designed for the top few % is required. That would probably leave majority (even larger than currently) of GATE exam takers unqualified. The only thing I can say from statistics at IITM is that the top 1% or 0.5% are not likely to perform very poorly in our MTech.

One can get into top 1 percentile without solving even one question in vast areas of computer science.

This is also true. I'd say the purpose of GATE is to test the knowledge of core courses-maybe 6 to 8. Of these, omitting a couple certainly won't damage your prospects. This a problem in EC as well. The kind of comprehensive evaluation over 20-25 courses can only be obtained from a grade card of a UG program with some rigor in evaluation.

on an average a 12th class student should be able to score higher than an average 4th year student in CS

This I believe is not true for EC. I have to see question-wise data for this(do you have this for CS?), but couldn't find it in the public domain. GATE EC rank of 1000 corresponds to about 55-60 marks out of 100, but these are also after some normalization across sessions-not sure how much these move things. I believe one an get maybe 20 marks only knowing the general aptitude part which carries 15 marks and maybe a bit more of general algebra-far from 55-60 raw marks. Also, the move to computerized tests has reduced the number of multiple choice questions which can be gamed more easily and made the test better.

... which declares you to be in the top 1 percent of the country ...

Here, I disagree with the implied exclusivity of the top 1%. If a large number of unqualified people join in, percentile of the top improves without any change in their performance. 80-85% of GATE registrants don't even get the qualifying score. What is your estimate of the number of graduating students "who know what they are doing"? [I don't know, but have the following wild guesswork] If there are 200 classes in the country and 20 in each class know what they are doing, and half of them chose not to write GATE, we'll have only 2000 students. It is my speculation that GATE correctly identifies those. Like I mentioned earlier, in my experience, they don't perform poorly in our MTech program. The failures here(of small numbers) are entirely in the UG educational system.

Ajit R. Jadhav said...

This topic interests me (though not as much as it used to).

Dheeraj, you seem to be making a CS-specific case. ...

In CS, what are the 3--4 courses in which the student needs to be good so that even if he doesn't prepare for the others, he can still land into the top 1 percentile?

I mean to ask: are these 3--4 courses very specific ones (say the more basic/fundamental ones like data structures and algorithms, discrete maths), or is it the case that selective preparation in *any* 3--4 courses (out of some 10--15 CS-specific courses covered in the second- and third-year) would be good enough (provided the student had good test-beating skills)? The issue would seem to be worrisome only if the latter is the case.

Best,

--Ajit
[E&OE]

Ungrateful Alive said...

Actually, solution bug or not, the question is quite silly and useless.
Regarding 80 vs 99 percentile, sure, even an IIT UG education can no longer guarantee CS preparedness comparable to Harvey Mudd or Mills or Pace. It's all inflation. Just like to watch youtube videos without choking you need about a 41597G mobile network, going by how well 3G and 4G works today.

Dheeraj Sanghi said...

@Nagendra, I think you got me wrong. I am not saying that the exam should be such that 99.5 percentile student should be statistically better than 99.4 percentile student. I am only saying that the exam should be such that the top 1 percent should be good students. I am assuming that about 10 percent of the CS graduates in the country are getting decent quality of education. Of this only half are sitting for GATE. But then 100% of the graduates are not sitting for GATE. So among those who are sitting for GATE, I would assume that upwards of 5 percent students would be good. I think a properly designed exam should ensure that most people in the top 1 percent are from these 5 percent good students. And I am expressing my feeling that GATE is unable to achieve this.

Also, when I say that one can get into top 1 percentile without solving even one question in vast areas of computer science, I am only including those subjects that are part of GATE syllabus.

Dheeraj Sanghi said...

@Ajit, The 3-4 basic CSE courses like Data Structures, Organization, Algorithms and Programming appear to be enough. However, if there were an exam where the questions in those extra courses were not confusing and not very difficult, then we will be able to have a better sense of how much core CS does a student knows.