On Writing a PhD Thesis

1
On Writing a PhD Thesis

• Ian Miguel
• AI Group
• Department of Computer Science
• University of York

2
Overview

• Part I Before Writing Up
• The Scientific Method.
• Reading.
• Keeping Records.
• Organisation.
• Part II Writing Up
• Planning.
• Structure.
• Style.
• Worked Examples.

3
Part I Before Writing Up
4
The Scientific Method

• Identify a Problem.
• From observation/consideration of work of others.
• Form hypotheses.
• Test hypotheses.
• Experimentation/Further observation.
• Modify hypotheses according to results.
• Go to 3.

5
Dynamic Flexible Constraint Satisfaction and Its
Application to AI Planning

• The Constraint Satisfaction Problem (CSP)
• Given
• A finite set of variables.
• Each variable has an associated finite domain of
  potential values.
• A finite set of constraints over these variables.
• Find
• A complete assignment of values to variables that
  satisfies all constraints.

6
Example CSP Course Scheduling

• Decide the number of lecture, exercise and
  training sessions.
• So we need 3 variables.
• Constraints
• There must be a total of 8 sessions.
• Professor A will give 4 or 5 lectures.
• Dr B will give 3 or 4 exercise sessions.
• There must be 1 or 2 training sessions.

Lectures in 4, 5, Exercise in 3, 4, Training
in 1, 2Lectures Exercise Training 8
7
Course Scheduling A Solution

• Lectures 4
• Exercise 3
• Training 1

CSP is NP-completeObtained by some search
process.
8
Observation

• Classical CSP suffers from 2 weaknesses
• Static formulation Once variables, domains,
  constraints are specified, no change allowed.
• Hard constraints Imperative, inflexible.
• Individually, these weaknesses have been
  addressed.
• Dynamic CSP solution re-use.
• Flexible CSP allow soft constraints.
• Little work to support both dynamic and flexible
  CSP.

9
Central Hypothesis and Experimentation

• Developing techniques to support DFCSP will allow
  us to solve effectively new types of problem.
• Random Problems, generated by a set of
  parameters.
• Explore structure and properties of a certain
  class of DFCSPs.
• Compare against naïve approaches.
• Compare different DFCSP algorithms.
• AI Planning.
• More realistic application.

10
Reading

• The key to good writing is reading!
• Read all the way through the course of your PhD.
• Reading (good) prose is equally effective in
  improving your writing style.
• Can save many hours of work seeking to
  re-discover information already in the
  literature.
• Avoid disappointment!
• Learn by example.
• Relating your observations/ideas to earlier work
  will lead to a deeper understanding of your
  problem.
• You are not performing your research in a vacuum.

11
Reading
Paper D

• Sources (reverse order of maturity)
• Books.
• Journals.
• Conference papers.
• Workshop papers/technical reports.
• Maintain a reference graph of citations.
• Get in touch with researchers working in a
  similar direction to you.
• Workshops are good for this less formal.
• Also provide a preview of forthcoming conference
  papers.

Paper C
Paper B
Paper A
12
Reading

• You cant afford to read everything in full.
• Read the title/abstract.
• Read the introduction/conclusion.
• Search for the important results.
• If all goes well, read the whole thing.
• Read critically.
• How can I use this?
• Does this really do what the author claims?
• What if?

13
Reading

• Relevance.
• Devote most effort to the inner circles.
• Know the mostrelevant papers inthe middle
  circlewell.
• Know the majorresults in the outercircle.

SAT
Graph Colouring
Search

Wider CSPLiterature
Dynamic CSP
Flexible CSP
PlanningLiterature
14
Keep Records

• Writing helps you to remember.
• Think back to your lectures.
• Writing helps you to observe.
• Writing focuses the attention.
• A concise summary of a large set of computational
  experiments makes later interpretation much
  easier.
• Writing helps you to think.
• Writing is creative.
• Helps sort out ideas, preserve them for later
  consideration.

15
Keep Records

• Keep records of
• Reading, and your opinions.
• Results, and their trends.
• Ideas.
• No-one will read these except you. Add
• Speculation.
• Current problems in your work.
• Writing down the problem clearly may lead you to
  the answer.
• Possible solutions.

16
Keep Records

• Read over your notes periodically.
• Perhaps make a summary for easy reference.
• Identify gaps in your knowledge.
• Gain a deeper understanding of past work in light
  of what you know now.
• Your records can form the backbone of a paper.
• Your papers form the backbone of your thesis.
• Reviewers comments provide expert feedback.
• Invaluable when writing up.

17
Organisation

• Keep a plan of things to do.
• Tick off things achieved.
• Satisfying, and helps to structure writing.
• A regular working schedule
• Not necessarily 9-5!
• But have a definite start time.
• Leave something simple to do first thing.
• Similarly, do simple tasks when tired.
• E.g. diagram drawing.

18
Part II Writing Up
19
Writing Up

• Always takes (much) longer than you think.
• There are many distractions in the final year
• Finding a job.
• Moving house.
• Attending conferences/revising journal papers.
• Writing a grant proposal
• So start early!
• Surprising how progress can be made when you sort
  your notes/papers into chapters.

20
Planning

• Do not underestimate the need for planning.
• A thesis is a large undertaking.
• Failing to plan leads to a lot of wasted time.
• Develop a plan with your supervisor.
• Structure your thesis into logical sections.
• Leave plenty of time for revision, taking into
  account the comments of your supervisor and
  others.

Planning
Writing
Revision
21
Structure

• Your thesis should tell a coherent story.
• Progress through thesis reflects progress from
  problem to solution.
• Structure is important to help reader.
• Partitioning of the thesis into chapters.
• Partitioning of the chapters into sections.
• Smooth linkage of chapters/sections/paragraphs.
• Make a good job of this and your thesis will be a
  pleasure to read.

22
Structure

• Introduction
• The Constraint Satisfaction Problem.
• Dynamic Flexible Constraint Satisfaction.
• An Empirical Study of Fuzzy rrDFCSPs.
• Dynamic CSP in Domain-independent AI Planning.
• GP-rrDCSP Experimental Results.
• Flexible Planning Problems and Flexible
  Graphplan.
• FGP Experimental Results.
• Conclusion.

Pt. 1
Pt. 2
Theory/Results Interleaved
23
Structure

• Introduction.
• Non-technical keep symbols to a minimum.
• Set the scene.
• Overview of main strands of the thesis.
• Define the challenges.
• Example
• Importance of constraints for knowledge
  representation and inference.
• Limitations of classical CSP
• Static formulation, hard constraints.
• Objective Dynamic Flexible CSP.

24
Structure

• Literature review.
• The foundation of your thesis.
• If you cant convince the reader that you
  understand the existing literature, your original
  work is on shaky ground.
• Formal definitions.
• Describe the key ideas you are building on.
• Worked examples.
• Example
• Classical CSP. Definitions, solution procedures,
  etc.
• Dynamic CSP. Types, solution procedures.
• Flexible CSP. Types, solution procedures.
• All with examples.

25
Structure

• Middle Chapters.
• Your original contribution.
• Each chapter should express a key idea.
• It should be clear how each chapter fits into the
  story you are telling.
• Example Dynamic Flexible Constraint
  Satisfaction.
• Outlines need for combined DFCSP techniques.
• Describes matrix of instances of DFCSP.
• Defines new DFCSP algorithms.
• Lead into next chapter experimentation.

26
Structure

• Conclusion.
• Relatively short, but important - write
  carefully.
• Outline main achievements.
• Gather various contributions, draw general
  lessons.
• Criticism of thesis.
• Take a step back from your work.
• Further work.
• In-depth analysis of other types of DFCSP.
• Further development of planning systems.
• Further applications, e.g. flexible compositional
  modelling.

27
Structure

• Within a Chapter.
• Start with a brief opening paragraph.
• Motivation, connect to previous chapter.
• Summarise what will come without giving away the
  conclusions.
• Body.
• Try to be as self-contained as possible.
• End with a summary.
• Review and emphasise the chapters achievements.
• Set the chapter in context of the thesis as a
  whole.
• Connect to next chapter.

28
Writing Style

• Your writing style reflects on you and your work.
• Poor/sloppy writing irritates the reader and
  distracts him/her from the points you are making.
• It is essential to polish your writing style
  practice.
• Be prepared to revise your writing to get it
  right. The first draft is rarely good enough.
• The secret
• Have something to say, and say it as simply as
  you can.

• Verbs has to agree with their subjects.

29
Writing Style

• Explanation.
• Consider the needs of the reader.
• What does he/she know?
• Good introduction/background chapters assume that
  the reader knows very little.
• What further explanation is required?
• Think about these questions
• What is it? How does it work? Why is it used?
• What have you done? Why and how was it done?
• What have you found?

• And dont start a sentence with a conjunction.

30
Writing Style

• Clarity.
• Good work requires clear thinking.
• But good work can be obscured by bad writing.
• Take the trouble to find the words which express
  your meaning precisely avoid ambiguity.
• Prefer a short word to a long one (unless the
  longer one is better).
• do vs. accomplish, so vs. consequently, use vs.
  application, end vs. terminate

• Avoid clichés like the plague.

31
Writing Style

• Clarity (2).
• Avoid unnecessary qualification
• The actual number. The smallest possible
  minimum.
• Very true. Wholly new. Quite impossible.
• Technical terms.
• Undefined use assumes both that the reader is
  familiar with the concept and recognises it by
  its technical name.
• Would you have known what a CSP was?
• Give clear definitions.
• Do not use, undefined, to impress the reader.

• Do not use no double negatives.

32
Writing Style

• Clarity (3).
• Avoid circumlocution use of many words where few
  would do better
• In view of the fact that ? because. If it is
  assumed that ? if.
• During the time that ? while. Have been shown to
  be ? are.
• On two separate occasions ? twice.
• In view of the foregoing circumstances ?
  therefore.
• Sources of verbosity
• Padding.
• Writing is not speech no need to repeat for
  emphasis nor give reader time to think.
• Lazy sentence construction/revision.

• Be more or less specific.

33
Writing Style

• Completeness.
• Be comprehensive.
• Every line of argument should be followed through
  to a logical conclusion.
• Dont leave half-developed ideas hanging.
• Avoid errors of omission.
• Dont leave out evidence that is against your
  hypothesis.
• Dont undervalue the findings of other
  scientists.
• But show an awareness of the limits of your
  knowledge.
• Dont claim to know more than you do.

• Proofread carefully to see if you any words out.

34
Writing Style

• Impartiality.
• Show an awareness of all sides of a question.
• Try not to be biased by preconceived ideas.
• My complex new algorithm must be better than the
  simple old one
• Make clear any assumptions underlying your
  arguments.
• If these are incorrect, your conclusions may be
  incorrect.
• Assumptions should be based on sufficient
  evidence.
• Words that introduce assumptions obviously,
  surely, of course.

• Also, always absolutely avoid annoying
  alliteration.

35
Writing Style

• Accuracy.
• Good science depends upon
• Care in observation.
• Precision in measurement.
• Care in recording the above.
• Care in analysis (again, choose words carefully).
• Every experiment should be repeatable.
• Do not hide important details.
• Every conclusion should be verifiable.

• Prepositions are not words to end sentences with.

36
Writing Style

• Objectivity.
• Every statement should be based on evidence not
  unsupported opinion.
• Never state an opinion (even of the majority) as
  fact.
• Warning words/phrases
• Possible, probable, it is likely to.
• Have you considered the evidence sufficiently?
• Is there enough evidence to omit the
  qualification?
• Is further work necessary? Dont be afraid to
  say so.

• Use words correctly, irregardless of how others
  use them.

37
Writing Style

• Nothing should be implied or left to the readers
  imagination.
• You are not writing a novel!
• Phrases to avoid
• As is well known (I think).
• It is perhaps true to say (I do not know what to
  think).
• For obvious reasons (I have no evidence).
• It is likely that (I have not got enough
  evidence).

• No sentence fragments.

38
Writing Style

• Simplicity.
• William of Ockham (14th Century Theologian).
• Occams razor
• One should not increase, beyond what is
  necessary, the number of entities required to
  explain anything.
• I.e. If you have two theories which both explain
  the observed facts then you should use the
  simpler at least until more evidence is
  obtained.
• Simplicity in writing or in constructing figures
  is the outward sign of clarity of thought.
• Makes a good impression on the reader.

• It is wrong to ever split an infinitive.

39
Worked Examples

• Very important to promote understanding.
• Make good use of figures.
• A good figure is far easier to understand than
  text alone.
• Try to use the same example as far as possible.
• Construct an example to illustrate several
  points.
• Much easier for the reader to follow than several
  individual examples.

40
Worked Examples
c3
r2
r3
guard1
pkg1
pkg2
r1
c1
c2

• Plan course of action to achieve pre-specified
  goals.
• Components of a planning problem
• Plan objects.
• Initial state.
• Goal state.
• Operators.

41
Worked Examples
c3
r2
r3
guard1
pkg1
pkg2
r1
c1
c2

• Goals
• Operators
• Load/Unload-truck (guard present if valuable).
• Drive-truck (non-mountainous roads only).
• Guard-boards-truck.

42
Worked Examples
c3
r2
r3
guard1
pkg1
pkg2
r1
c1
c2

• Drive-truck truck c1 to c2 via r1.

43
Worked Examples
c3
r2
guard1
r3
pkg2
pkg1
r1
c1
c2

• Load-truck pkg2.Guard-boards-truck.

44
Worked Examples
c3
r2
guard1
r3
pkg2
pkg1
r1
c1
c2

• Drive-truck truck c2 to c1 via r1.

45
Worked Examples
c3
r2
r3
guard1
Pkgs1, 2
r1
c1
c2

• Load-truck pkg1.

46
Worked Examples
Pkgs1, 2
guard1
c3
r2
r3
r1
c1
c2

• Drive-truck truck c1 to c2 via r1.
• Drive-truck truck c2 to c3 via r2.

47
Worked Examples
guard1
pkg1
pkg2
c3
r2
r3
r1
c1
c2

• Unload-truck pkg1.Unload-truck pkg2.
• Goals achieved.

48
Worked Examples
c3
r2
r3
guard1
pkg1
pkg2
r1
c1
c2

• Hopefully, this example is easier to follow than
  a dry description of plan objects, operators,
  goals, and solutions.
• It is also used to describe
• Planning algorithms.
• Flexible planning problems (in an extended form).
• Flexible planning algorithms.

49
Summary

• Before writing up
• The Scientific Method.
• Problem, Hypothesis, Experimentation,
  Modification.
• Organisation.
• Reading, keeping records.
• Writing up
• Structure
• Style
• Worked examples.

50
Conclusion

• You should enjoy reading what you write.
• Seek advice from people who have recently written
  up.
• Swap drafts with others in the same position.
• Remember, it always takes longer than you think.
• But dont give up.
• Good luck!

51
Resources

• Scientists must write, R. Barrass,Chapman and
  Hall 1978.
• How to Documents
• www.dai.ed.ac.uk/homes/bundy/how-tos/how-tos.html
• www.cs.indiana.edu/mit.research.how.to/mit.researc
  h.how.to.html
• www.ifs.tuwien.ac.at/silvia/research-tips/
• Rools for Riting Good Inglish
• www.york.ac.uk/depts/arch/ugrad/bos/rools.htm