- 80 Views
- Uploaded on
- Presentation posted in: General

Presenting Research

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.

- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -

Presenting Research

Dr. Anjum Naveed &

Dr. Peter Bloodsworth

- Before you start:
- Have your literature review handy
- Be very clear regarding what you are trying to prove
- Re-read your problem statement / research hypothesis
- List the contributions that you want to claim in your thesis
- For each point in your list you will need evidence to convince others that you have achieved it
- People won’t just take your word for it
- The most common reason for a rejected paper is getting this wrong

- Use your literature review
- How have others doing similar work to you evaluated their research?
- What metrics did they use and why?
- Are there any common data sets that are used widely?
- How much proof is normally expected / presented?
- Some fields require more proof than others

- Are there any systems that you can compare your work with?
- Caution 1 : It might sound clever to pick a legacy system to compare against
- Your system will look better – right?
- WRONG!! : Researchers will spot this straight away and it will harm your credibility in the future
- Caution 2 : Be sure that you make fair comparisons
- Don’t deliberately pick an inappropriate system to compare against
- Don’t choose a data set that favours your system
- Don’t make sweeping statements with limited proof!

- Carefully write down the assumptions that you have made
- Are they reasonable?
- Could others question them?
- How would you answer tough questions?
- Results need to be bullet-proof
- What initial conditions were set?
- Could they bias the results?
- What have you done to avoid this?
- Could other researchers repeat your tests to verify the results?

- Two main types of evaluation:
- Quantitative Evaluation
- Qualitative Evaluation

- The choice / mix of evaluation techniques depends on your thesis topic
- Generally quantitative results allow us to make stronger claims
- We have to be more careful when taking a qualitative approach can’t claim too much
- Ask for advice from your supervisor before starting on this
- Try to write an early paper to get some feedback on your evaluation technique

- Numerical comparisons
- System X performs 10% more accurately than System Y
- The algorithm is 90% effective in classifying brain tumors
- Makes use of statistical and other mathematical techniques
- Includes formal mathematical proof
- Logical proof and Model checking
- Regression to identify trends
- Is very powerful but care needs to be taken because errors can be very costly
- Examiners tend to be very numerate and will spot mistakes

- In some research areas you won’t have access to the required resources needed for testing
- In such cases simulating or modeling can help us to generate quantitative results
- Caution : The results will only be as good as our simulation / model!!
- Use established simulation / modeling techniques and packages wherever possible
- Carefully show that your simulation / model is accurate and that its configuration doesn’t introduce bias
- Run several experiments and gradually increase complexity

- Claiming too much, justifying too little
- Using an inappropriate mathematical technique which introduces bias to results
- Making a basic mathematical error
- Selecting data that isn’t representative of your problem domain
- Choosing data which is biased in some way
- Not doing enough testing
- Not having a large enough data sample
- Misinterpreting results – missing the point or drawing wrong conclusions

- Not using metrics that are expected in your domain
- Misunderstanding metrics and applying incorrectly
- Choosing the wrong simulation tool and trying to force it to fit your problem
- Badly configuring your simulation / model so that it doesn’t really describe your problem
- Doing things in a hurry at the last minute increases all of the above risks – take your time!!

- This is not numerical
- Is more descriptive
- May involve a criteria for success
- Create a list of necessary features that your system needs to show to be deemed a success
- Use a range of tests to show how the system behaves in response to stimuli
- Try to anticipate the possible inputs to the system
- Create a real use-case and make it the focus of your evaluation

- You claim too much – remember that qualitative results give you less real evidence
- You make sweeping statements that you don’t properly justify (avoid words like: generic, optimal , etc)
- You cover only a small range of possible inputs
- You create a basic prototype and try to claim that it shows much more than it does
- You don’t do enough testing
- You set biased tests in some way without noticing
- Your criteria is too limited to really test your work