Seminars and MTPs
Topics and Projects
Meeting schedule: Autumn 2016
- Current MTech2 students: Tue/Fri afternoons in CH.
Guidelines for Seminars
Weekly meetings
- We will meet once a week (for half hour or so) to discuss a
paper. You are expected to read one paper per week, though some
difficult papers may spill over to the next week. Overall, your
seminar is supposed to cover 8-9 good papers.
- When you read papers for the semester, focus not just on
the details of the paper, but on the overall "big
picture", and on how the paper fits in with other papers in
this topic. For example, after reading the paper, you should be
comfortable answering the following questions:
- Problem definition.What is the problem being solved by the paper? And
why is the problem important / relevant?
- Context. How does this paper improve upon existing work? Why
is this paper needed when other ideas exist? What does this paper claim
to do that other papers don't? (This point will become clearer as you
read more papers.)
- Key Idea.What is the main insight /
key idea of the paper? Often, it is okay to ignore the details in the
first reading - you should try to get a high level understanding of the
idea first before you dig into the details.
- Validation.How do the authors validate their idea? What
experiments / simulations do they do? Is it enough? Are you convinced the
idea works?
- Relationship with your project.How is this paper
related to your proposed research topic? Are you solving the same
problem? A related but complementary problem? A completely different
problem? It is good to understand how this paper fits in with your work.
- You
may find this tutorial on how to read a paper
helpful. When reading a paper, try to understand the big picture before
you dive into the details. For a first read, the details are less
important, and the overall context and high-level understanding is more
important. That is not to say that you shouldn’t read the details
(you must), but don’t get lost in the details before you understand
the big picture.
- I
recommend preparing a short presentation for every meeting, of not more
than 10-12 slides. You can organize your presentation along the 5 main
points above (or in any other way). Below are some tips for preparing your
weekly presentations.
- Please take your weekly presentations seriously, and prepare good
quality slides. I expect you to reuse these slides for your final
seminar presentation.
- You do not have to explain every single idea, every single graph,
and every single sentence in the paper. Suffices to explain the key
idea(s) of the paper in your own words, and at a high-level, to
convince me (and your examiner, eventually) that you have actually
understood the paper. Your understanding will be seen in how well you
can identify the main parts of the paper, and summarize them in your
own words, in a concise fashion. Stick to the limit of 10-12
slides. Do not repeat/reproduce what's in the paper without
understanding it.
- Come up with running examples that illustrate the ideas in
the paper. You can even use the same example across multiple papers,
to illustrate how a particular problem is solved by various
papers. Coming up with your own examples will convey a certain level
of understanding.
- Draw figures, use animations etc to explain concepts. Avoid
long blurbs of text.
- Do NOT copy complicated equations, pseudocode and other such
things straight from the paper. It will be very difficult to explain
and understand. Instead, explain such content in your own words.
- If there are any flaws in the paper, you should be able to point
them out. Think critically about the paper, and do not take everything
that's in the paper as gospel. After the first few papers, you should
develop a taste for what is a good idea and what is not.
- Bring your laptops or email me the presentation beforehand so that
we can view it on my computer.
Seminar report and presentation
- You will be required to submit a final report and make a
presentation to an examiner to get graded in your seminar.
- Have a first draft of your seminar report at least two
weeks before your presentation date, so that I can review and give
feedback. Show me a draft of your presentation at least one
week before your presentation day. You must get feedback on your talk
from me, and from at least one other classmate of yours (preferably
from someone in the same area of systems/networking, but NOT working
on the exact same topic as you.) Two mock presentations, one to me,
and one to your peer, are a must.
- Your seminar report should be no more than 20 pages. It should
roughly have the following structure.
- Introduction (1-2 pages): describe the broad problem that
your set of papers are trying to solve. Why is this an important
problem? Use an example if needed to motivate the problem area and the
main solution techniques. Do not just list all the papers you have
read; instead, you must provide a taxonomy of the papers /
solution techniques you have studied. That is, you must be able to
classify the papers you have read into some broad bins based on some
criteria, and describe this broad classification in the
introduction. This high level understanding of the problem and
solution space is the most important contribution you can show from
your seminar. At the end of the introduction, the reader should have a
decent idea of the problem statement, and the high level classes of
solutions to the problem.
- You may have an optional background section that provides
more details and terminology to understand the rest of the report.
- Survey of papers read (5-6 pages). Briefly describe (in
half page or so each) the key idea behind each of the papers you have
read, along with a citation to the paper as appropriate. Describe the
key idea of the paper in your own words / figures. Comment critically
on the paper (merits / flaws etc). You must organize these papers
according to the broad classes you have come up with in the
introduction.
- Compare and contrast the solutions (1-2 pages): how do the
various approaches described above compare with each other? Which is
applicable in what scenarios? What are the pros and cons of each?
Ideally you should have a table or some such concise summary of all
papers. Or you can use one example and see how various systems handle
it. How you compare is left to you - this is another place where your
contribution / understanding will show.
- Conclude with describing how this seminar will lead into possible
topics for your MTP. Most of you should have a fairly good idea of
your MTP topic by the time your seminar is done.
- Your presentation should also be structured along similar
lines. You should prepare 18-20 slides for a 20 minute presentation,
leaving 5-10 minutes for Q&A. Please keep the following in mind for
the presentation.
- Keep your audience in mind when you give the talk. For the
seminar, you should assume that your examiner (and other students who
come to attend) knows the basics of the field, but not any
details. Make your seminar self contained, so that anyone who has done
a CS641 type course should be able to follow. You cannot speak the way
you speak to me at our weekly meetings, assuming all background
knowledge.
- Start your seminar with an overview of the problem, the solution
techniques, the context, and why it is important to solve the
problem. Define all terms you will use. Spend about 4-5 mins easing
the audience into your presentation. Do not start with a paper on
slide 1 and expect people to follow along.
- After the introduction, you may start talking about each
paper. Keep the discussion simple and accessible. Use examples and
figures. Concrete examples that illustrate your problem and solution
are a thousand times better than an abstract explanation. Keep
yourself in the shoes of the poor audience who is trying to grasp
several months of your work in 30 minutes.
- Spend the last few minutes summarizing all papers, pros and cons,
comparisons, take-away points etc. If you feel you are running out of
time, skip some intermediate papers but do not skip this part.
- Prepare for no more than 20 minutes of talking time. It will end
up being 30 minutes with questions. A rough guideline is about 20
slides for a 20 minute talk, but no more.
- Make your slides interesting with figures, colors etc. where
appropriate.
- Finally, please take your presentations seriously. The goal of the
seminar is not just doing a literature survey for your thesis. A
primary goal is to teach you how to read papers, identify the crux of
the paper, and present it in simple terms to a general audience. It is
not enough if you are technically sound (though it helps a long way),
you should also be able to take people along with you and convince
them of your technical capabilities. I cannot stress enough the
importance of good communication: prepare seriously for your talks
(this one, your MTP presentations, and so on).
- And attend a few other seminars, not just your own, so you know
what works and what doesn't.
Optional R&D project
Some of you may opt to do an R&D project with me, along with your
seminar, so that you can get familiar with the tools and techniques
needed for your MTP. An R&D project can also be used to get a hands-on
feel for some of the techniques you are reading about in the seminar,
and can reinforce your understanding of the papers. In projects where
you will be working with the code developed by your seniors, an R&D
project also lets you have an overlap with them and learn from
them. If none of the above apply to you, then not doing an R&D project
is also fine.
If you end up doing an R&D project, we will setup
weekly meetings, roughly aligned with your seminar slot. You can also
drop by the lab and meet me with the senior students. Finally,
you will have to submit a report at the end of the semester, and do a
small demo of what you have learned.
Guidelines for MTP
During the semester
- We will aim to meet twice a week during the semester for regular
updates. However, if you are stuck on something, please don't wait for
the next meeting slot, and approach me earlier.
- You should have a clear idea of the deliverable expected by the
end of the semester fairly early (within the first month of starting
your stage I or stage II).
- You are expected to put in 30-40 hours a week on an
average. Maintain a consistent pace throughout the semester, so that
work never stalls.
- Work in the lab as far as possible. Coming to the lab
regularly everyday will help you focus better. Further, the projects
of several students are closely related, and working together can
boost your productivity.
- Flag any requirements for equipment, or any other needs / concerns
you have to me, so that I can act on them in time.
MTP Stage I/II report and presentation
- Keep your reports to around 20 pages for stage I and 30 pages for
stage II. Plan a 30 minute presentation (20 min + Q&A) for your stage
I, and 45 minutes (30 min + Q&A) for your stage II.
- Start working on an outline of your report one month before
your presentation. I expect to receive a draft about 3 weeks before,
and we can iterate over it in parallel with your work. And you should
have a final report one week before your
presentation. Treat this as a hard deadline.
- Similarly, I expect to see the first draft of your presentation
two weeks before your MTP presentation. In the week before your
presentation, you must do at least two mock presentations , one to me,
and one to your peer (someone in this general broad area, but
not working on the same topic).
- Have clarity on what your problem statement is, and pick a
suitable suggestive title for your thesis.
- Both your report and presentation should start out with a short
introduction / overview section. You should give a high-level
background and motivation, an overview of the problem you have solved,
the work you have done (in seminar, R&D, stage I, stage II etc.), and
your main result and contributions. You should get through this
content within the first 1-2 pages of your report, and within the
first 5-6 slides of your presentation (in under 5 minutes). So, anyone
reading your report or attending your talk should get a good idea of
the nature/quantity/quality of your work, and your key idea /
contribution (obviously, not all the details) in under 5 minutes.
- Next, you can have a section of more related work / background to
clarify the context for your problem. You may devote more space to it
in the report, but during the presentation, talk about only what is
absolutely required to understand the rest of your talk.
- Next, you will describe your design and implementation in detail,
followed by your experiments and evaluation. Use figures, tables, flow
charts etc. liberally to explain your ideas, instead of long blurbs of
text.
- You should have a small number of well-explained graphs in your
evaluation, instead of a large number of results. For each graph, you
must carefully explain what the experimental scenario is, what the X
and Y axes are in the graph, what the graph/result means, and what
conclusions to draw. Graphs should be easily readable on paper, and
visible when projected on screen.
- You must conclude by pointing out your major results and conclusions.
Grading
In general, your grade will depend upon several components:
- Your actual work, both quantity and quality. Ideally, you should
have completed enough work to meet the goal set out at the start of
the semester.
- Your initiative / independence in doing things on your own
(without me pushing you or having to tell you every single thing), and
your enthusiasm and sincerity during your project.
- Your ability to communicate well in your report and presentation.
Here are general guidelines on how your work translates to a grade (these are only rough guidelines: grading is always done on a case-by-case basis).
- AA: You have done exceedingly well on all three aspects above.
- AB: You have done very well in most aspects, but missed on a few
things (e.g., work was good but presentation wasn't great, or didn't
show enough intiative on your own and needed lot of hand-holding).
- BB: Decent effort, average output, needs improvement in most aspects.
- BC: Below average work and effort.
- CC: Bad quality of work, did not learn much or accomplish much in the project.
Of course, the most important aspect of a project (for me, and hopefully for you as well) is that we all learn something out of it, and have a good time working together. If you like your work and enjoy it, the grade is usually a secondary concern. I hope you all work on things more out of passion and interest, and less for the grade you get.