Report on PhD Production in Computer Science highlights the Opportunity for PhDs

Leave a comment

On behalf of ACM India, I conducted the first survey on PhD production in Computer Science in India. The report has been published and can be found, along with the background, here.

As the report clearly shows, while the number is not as low as one thought, it is still about 125. And if you consider PhDs from only the top 20 institutions in the country, the number is in two digits. And the projections are that this number will only double in about 5 years.

This study actually highlights the tremendous opportunity for those who are doing PhD in CS in India. Academics is growing rapidly with so many new IITs, IIITs, and other Institutes coming up. Even if you consider each such Institute will need about 5 faculty members each year, 50 upcoming Institutes can easily consume 250 PhDs. Then there are at least 20 research labs in many software companies, including the large software companies which seem keen to expand their R&D capabilities rapidly, and various other companies that can consume PhDs. Overall, the private sector can also consume about 250 PhDs per year. There are other opportunities in Govt sector also. In a nutshell, the supply is significantly lesser than the demand. And this gap is likely to increase as demand is set to increase.

Due to this mismatch in demand and supply, and the growth of academics, the compensation for fresh PhDs is now very good. Companies will often pay a package starting from Rs 15 Lac to Rs 25 Lac or more for a fresh PhD. Academic packages are also quite good after the sixth pay commission – an Asst Prof can have a yearly compensation of Rs 8 to Rs 12 Lac. Compare this with the starting package for software jobs – except for a few multinationals, the starting package tends to be around Rs 3.5 Lac for the large and medium sized software houses (which is where 90% of the software jobs are), and Rs 4 to 6 for the niche players. Even if one counts for the 4 to 5 years that one has to spend in getting a PhD, compensation wise, a student who does a PhD will clearly come out ahead.

And then there are the really strong benefits of doing a PhD – the main reasons why people preferred this route even when the compensation was not good. And that is, the freedom to explore and chose your own work agenda, the non-repetitive and challenging nature of work, the culture of R&D, lack of hierarchy, being a member of the global community of researchers, etc.

Overall, while the PhD production report does not have too much good news for those who want to recruit PhDs, it is good news for those who are considering doing PhD.

Advertisements

Collaborative PhD Program

Leave a comment

A strong sentiment expressed in the survey I had conducted regarding doing PhD in India was the desire to have global exposure, and spending some time in a research group in overseas (there was also a strong sentiment in favor of spending some time in industry research lab). Though we provided for up to two international travels for presenting papers to PhD scholars so they can gain some international exposure, we felt that it will even be better if the student can spend a summer/semester/year in an overseas lab/institute.  Towards this end, we championed the “collaborative PhD program”, for which even in the first year we got about 10 collaborators.  First the key features of this program:

 

  • A PhD student is admitted into IIITD, from where he/she will get the degree. Interview for admission may be done jointly by IIITD and the partner group.
  • The main supervisor of the PhD scholar will be from IIITD; a co-supervisor will be from the partner group.
  • After doing the course work and starting the research, the scholar will spend 3 months to 1 year with the co-supervisor at the partner group/institution.
  • Most publications will be joint between the scholar, supervisor, and co-supervisor.

 

The financial model for this was that while the student is at IIIT-Delhi, we pay, and when the student is with the co-supervisor, the co-supervisor pays.

This program is a win-win for most involved – the student has the benefit of two supervisors, and also gets the exposure of working in another group/institute for some time. The supervisor and co-supervisor benefit by having a PhD student without having to pay for his scholarship for the full period, as well as build linkages and collaboration, which can prove to be very useful in future as international collaborations become more needed.

And this model avoids the complexities involved in joint degree programs, which require academic processes on both sides to converge on a common view, before proceeding – and this is an extremely complex and difficult exercise.

This model of collaborative PhD program is working well, as there is a lot of interest now across the world to establish collaborations in India. This year also we have many collaborations – most of them established by individual faculty members.

 

Completing PhD in Four Years in CS after BTech/BE

2 Comments

(Adaptation of a note I had written a few years ago)

PhD is a unique degree in that it is not focused on acquiring more knowledge, but is oriented around doing research. As research requires many abilities as well as creation of new knowledge, a PhD program is necessarily a long one.

Though many students who join the PhD program are not unduly worried about the time the program takes, there is also a section of students for whom the possibility of a long program is clearly a deterrent. So, even though the careers for PhDs in Computer Science have increased and continue to increase, in many cases, the perceived duration is discouraging these students from undertaking the PhD program.

In this note we will discuss how a PhD program can be comfortably finished in four years, after the B.Tech./BE, if the student is motivated and does not loosen up during the program. It is hoped that this note will help students who are inclined towards PhD take a more informed decision, and not just give up on PhD due to its large duration. (This is not just a hypothetical approach – I have successfully applied it on one candidate. Though one candidate is not enough data, as it was applied consciously and right from the start, it does provide a limited validation of its feasibility.)

There are essentially three phases in any PhD program. (1) Course work to gain breadth, (2) Identifying the area and problem to work on, and (3) doing the research work and writing results. All these three phases are roughly of equal duration – about 1.5 to 2 years each. They are often done linearly, leading to the program length of about 5 to 6 years.

The most difficult part of a PhD program is the phase two – identifying the area and defining the problem to be worked on in the thesis (which includes some a sense of how to approach the problem). It is often this phase in which a lot of time is spent – there is no tight course schedule to help, no deadlines, and no pressure of time as not too much time has passed. With some planning and initiative, however, it is possible to shorten the impact of this phase by doing parts of it in parallel with other phases.

To put this strategy in action, a rough schedule for a 4-year PhD is suggested, specifying what should be achieved in each semester.

Semester 1: Focus is on courses. The candidate should meet with different faculty; find out their areas of interest, and what type of problems they are working on, and try to Identify a faculty member with whom he/she may want to work with.

Semester 2: Do some advanced courses, and/or some “independent study” with the professor. Use these courses and extra study/independent study to engage with the professors to identify possible problem-areas for PhD. Here, it will help a lot if the potential advisor has some pre-defined problem which the student likes.

Summer, Semester 3: Finish the comprehensive (or qualifier or whatever it is called). Do further independent study to focus more on the problem-area, and generate (maybe small) ideas which can be developed further. Start active research.

Semester 4: Focus on developing the ideas and doing related reading. The developed ideas should be converted into research papers for conferences/journals, and where possible submit them for publication.

Semester 5, 6: Refine and/or expand the problem. Or work on some related ideas that got generated while working/reading. I.e. if the problem you are working on is not expandable, you can work on some related problems. Goal: Write at least two papers in this year. (It is better to start writing papers early, as they may not get accepted in the first attempt, and might need revision. As it is best to have some papers to your credit when you graduate, it is imperative that by the end of third year, a few papers have been written and submitted.)

Semester 7: Further development of ideas. Papers accepted in conferences which can be expanded, should be expanded as journal papers. Any pending or new ideas should be developed and papers written and submitted.

Semester 8: This is the closing semester. With papers written earlier, the thesis will be largely these papers plus a few chapters. The thesis writing effort reduces, but it is still a lot. The goal is to tie the loose threads, close pending issues to complete and submit the thesis.

The basic strategy in this approach is to start research work early, and to break the large problem of PhD into smaller problems of writing papers in closely related problems (or different aspects of some problem). In the process, it is ensuring that phase 2 of the PhD process is done in parallel with the other two phases of doing course work and doing research. Besides giving the satisfaction of visible progress in terms of publications, this approach has another advantage – with publications, the case becomes stronger and evident in the eyes of the thesis examiners.

In this schedule, the advisor has to tightly manage the program (assuming that completing the PhD soon is a shared objective of him and the student). This approach implicitly assumes that the advisor has a relatively well defined problem to offer to the candidate. With the problem defined, in the second year, the advisor has to really guide the student in doing the work – i.e. the work is to a large extent decided and directed by the advisor, but executed by the student. The paper writing exercise will also be the same – the advisor will guide the student in the art of paper writing, and through many rounds of the first paper, help the student develop the ability to properly write scientific papers. In the third year the role of the advisor matures to being a collaborator – actively engaged in developing and generating ideas together with the student and writing papers. In the final year, the role should actively become more passive so as to let the student take the lead. The advisor should become more of a bouncing board, giving suggestions and feedback.

This approach is primarily geared towards completing the PhD quickly and developing research doing capability, on which a research career can be built. There are, however, some disadvantages of this approach. Due to its focus on time, it discourages looking at hard problems, which are risky and may require much more time. Also, the approach of having papers in related problems can provide breadth, but it can discourage working at a larger problem for a longer periods where results may come only towards the end. And it can also reduce the development of the problem-finding ability of the PhD candidate, as it relies a lot on the advisor proposing the problem. Overall, this approach  is geared towards someone who treats PhD as a starting stage of a research career where he is learning the skills of doing research, which he will further develop in the rest of his career.

 

Reducing the Duration of the PhD program

2 Comments

The basic objective of a PhD program is to develop researchers. PhD programs across the world vary a lot in the method they follow to achieve this objective. There are programs which have no course work (e.g. in UK and Australia), to programs which have a huge amount of course work requirement (e.g. in some Universities in US).  Some programs have elaborate, and often time consuming, PhD thesis evaluation process (e.g. in almost all IITs and most Universities in India), while some have far shorter with greater reliance on the evaluation of the PhD supervisor (e.g. most Universities in US).

How long it takes a student to finish a PhD in Computer Science also varies a lot – from 3 to 4 years in UK and Australia to 5 to 7 years in many universities in US, as well as in IITs. I have strongly believed for a long time that it is possible to have a high quality PhD program where PhD students can expect to finish their PhD in 4 years. To achieve this objective of completing in 4 years but without diluting quality, we took some key decisions.

  • Course Requirement. A PhD should have some course requirement – that was very clear as I had seen that  often graduates of PhD programs with no course work requirement often do not have the breadth and rigor in their approach and are insufficiently prepared for a career in research, particularly the academic career (this is the primary reason why, for example, PhDs from Australia, where course work requirement is minimal, are often not preferred for academic positions in India, particularly when compared to PhDs from US, where most universities have decent course requirements.) After a lot of discussion and thinking, we settled on a requirement of doing 8 courses after BTech (and 50% of this after doing MTech), while keeping the option open to increase/decrease it, depending on the background of the student. This is equivalent to about 2 semester’s worth of effort (though a student may spread it over 3 or 4 semester, as he/she may spend some time doing research even in early semesters). This requirement will ensure that our PhD students will have decent breadth.
  • Identifying the area of interest early. It is well known that in PhD programs that take 5-6 years, the first 2 to 3 years are spent in identifying the area the student wants to work in and the problem to work on. We decided that we will require students admitted to the PhD program to have some clear ideas of the areas they want to work in, and we will take them only if we have strength in those areas. Towards this end, even in our fliers and advertisements, we listed the areas in which we are looking for PhD students. By “forcing” the student to identify the areas early, we hoped to reduce the front end of the PhD cycle – the student can then easily identify the group and the advisor almost in the start (our rules require the student to identify the advisor within 2 semesters), and start identifying the problem to work on. This approach has the disadvantage that it disallows exploration, and it does not allow “uncommitted” students,  but we felt that it is better to stick to students who have a clearer idea of what they want when they are coming to do PhD. This seems to have at least one desired effect – students are getting involved in research and publishing early – within the first two years, many of the PhD students of the first batch have published some papers.
  • Tight progress monitoring. Given the unstructured nature of PhD, it is very easy for a student to lose a year or two if the student relaxes a bit – it is far too common to see students who spent a few years just reading and exploring, but not pursuing any focused agenda. We decided that we will monitor each PhD student regularly for progress. For this we instituted a rule that each PhD student has to give a seminar once a year, which will be attended by most of the faculty, who will carefully evaluate the progress. (Of course, this also helped improve the communication abilities of the PhD student.) Later we added that in the winter semester, a short presentation will be given by the advisor of each student on progress of the student to the faculty. In other words, the progress is monitored each semester.
  • Thesis evaluation. Thesis evaluation can easily take a year or more in many Indian Institutes. This is clearly one area where at least six months can be saved easily from the PhD duration, without any effect on thesis quality. We took the basic structure of evaluation in IITs, which requires the thesis to be evaluated by a few external examiners, and combined it with the use of technology and the evaluation process used in CS conferences, in a hope to reduce evaluation time. Essentially, the evaluators will be like PC committee members who are assigned a paper (the thesis) for review – they will submit their review in 8 weeks, after which there will be a PC meeting (the thesis defense) in which the final decision will be taken by the committee.

With these measures, we feel confident that a good student can comfortably complete his/her PhD in 4 years, without diluting the rigor and quality of the PhD.

 

Starting the PhD Program in Computer Science

2 Comments

As our goal was to develop IIIT-Delhi as a leading research-led Institute in the country, after starting the BTech program in 2008, we decided to start the PhD program in 2009. We felt that this will give a clear message that we would like to be a research-led Institute. Thankfully, by the end of the academic year 08-09, we had reasonable number of faculty acceptances, and a few research groups had evolved. So, we had decent capability in these areas to guide PhD thesis.

For designing the PhD program, the key input I took was from a survey I had conducted earlier of students from IITs at Kanpur, Delhi, and Mumbai on why they don’t do PhD in India and what will they want in their PhD program in India. From the survey a few things we picked some key inputs for designing our program:

  1. A strong reason for not doing PhD was its duration and the uncertainty in the duration – good students did not want to spend another 5 to 6 years for PhD.
  2. They felt that the stipend was too low, making the opportunity cost too high.
  3. They wanted global exposure during their PhD program itself, possibly also of spending sometime in another lab/country.
  4. GATE as a requirement was also considered an irritant by some

Reason 2 is relatively easier to handle. We decided to increase the stipend for our PhD scholars from the currently prevailing norms of around Rs 8000 to 15000 to Rs 20000 in the start (which has been further increased to 22000), which can increase to Rs 25,000 in later years – this ensured that their compensation during the PhD days was not to far off from the starting salaries in much of the software industry.

For reason 4, we decided not to insist on GATE as a necessary requirement. We also realized that by requiring GATE, we will be restricted to a very limited pool of students, which will be same as the pool that IITs are targeting. We believed that there were many good students who did not take GATE, but who might for some reason or the other decide to do PhD – good examples of these would be those who wanted to do a job, but after finding out about the jobs are not excited about the prospects; those who wanted to go abroad for higher studies, but finally were not able to do so due to visa, family reasons, lack of scholarship, etc; and those who have been in industry for a few years and now want to pursue higher studies. And for us, it will be best to target this group, besides, of course, also considering students with GATE.

This decision of not requiring GATE is one of the smartest decisions we took, as by this we opened new input sources for our PhD program, which were generally not tapped by other Institutes. And in our first batch of 8 PhD students, only two had GATE. We also later observed that most of the students who did not have GATE are doing very well in the PhD program.

Another very smart decision we took was to focus on BTech for intake into the PhD program. It is clear that the BTech pool provides the best candidates in the largest numbers – this is a fact that has been understood and used effectively by US universities for last many decades as they took BTechs from India directly into the PhD program, but not well appreciated by the more rigid Indian system which expects primary intake as MTechs, with BTechs to be taken only if they are “exceptional”. Taking BTechs also gives the ability to train them better as they will generally do many more courses in the Institute. The key to attract BTechs directly into PhD is to address their main concern about the “risk” involved in making such a long-term commitment. We directly addressed this by allowing an “exit clause” in our regulations – a student can leave the PhD program for any reason, and may graduate and leave with an MTech. This simple method, something that I had argued earlier in IIT Kanpur but was not successful in convincing the concerned bodies about it, provides a clear risk-mitigation to bright BTech students, who are then more likely to consider PhD option.

Then we took some steps for containing the duration of the PhD program such that a good student can finish his/her PhD in 4 years after BTech – this topic will be discussed in another post.  For global exposure we took two steps – one we decided that PhD students will be provided travel support for attending upto two international conferences of repute for presenting papers. Second, we designed a collaborative PhD program which allowed the student a possibility of having an external co-advisor from a top lab/institute anywhere in the world, and spending time with the co-advisor. The collaborative program is also described later in another post.

So, we designed a program that addressed many of the key concerns of the youth about doing a PhD, without diluting the rigor of the program. For selecting the PhD scholars, we identified three basic capabilities that we felt are desired: (1) Strong conceptual/theory foundations, (2) Strong programming skills, and (3) Strong motivation/drive. We decided to evaluate all three aspects – for the first two we conducted tests – a programming test and a theory test. For the third, we relied on the interview. Thankfully we got over 200 applications in the first year – larger than what many IITs get. Finally, after a full day of exams and interview, we selected some candidates, eight of which finally joined us (this number is also very good as compared to many IITs), which gave our PhD program a very good start!