Skip to main content

Research Project

In this section, we discuss qualities of a good research project. Goal of a research project is to answer a question or solve a problem better than what has been done before. So, novelty is a key quality and it can come from a new problem, a new task within a known problem, or a new method to solve a known problem.

Sources of novelty:​
  • New problem: a problem that has not been solved before.
  • New task within a known problem: a new task that has not been solved before within a known problem. For example, a new task within image classification.
  • New method to solve a known problem: a new approach to solving a problem that has already been solved.
  • Insights and analysis: a new insight or analysis that has not been done before
  • Dataset/Software: a new dataset or software that has not been released before.
Questions to ask yourself when evaluating a research project:​

What are our contributions?

  • New Problem?
  • New Task within a Known Problem?
  • Insights or Analysis?
  • Technique or Method?
  • Empirical Results?
  • Dataset or Software?

Researchers at universities usually take the stand to open source everything they do. This enables other researchers to work on their problem, makes it easy for other researchers to get started and to measure their progress on this problem.

To successfully do this, provide:​
  • Problem statement: what is the problem you are trying to solve?
  • Baseline: what is the current state of the art on this problem?
  • Dataset: what data will you use to train and evaluate your model?
    • Provide a link to the dataset
    • Test dataset: to evaluate final systems
    • Development/Validation dataset: to evaluate progress during development
  • Evaluation metric: formula for measuring how well a systems does on the dev or test data.
Known Problems that lead to failure of a research project:​
  • Dataset: dataset can be hard to find, or it can be too small, or it can be too large, or it can be too noisy, or it can be too biased.
  • Not enough idea or insight contributions: Improvement and why that improvement happened are both important.
  • Overly ambitious: trying to solve a problem that is too hard, or trying to solve a problem that is too broad, or trying to solve a problem that is too vague. Also access to resources and time are important. If you don't have enough time or resources, you may not be able to finish the project.
  • Not well thought out: not having a clear plan, not having a clear goal.
When evaluating a project, write down:​
  • Has this problem been solved before? If yes, how is your approach different?
  • Can you formulate this problem in Math?
  • Is my techical approach new?
  • Do I have a dataset?
  • Do I have a baseline?
  • Write down the evaluation metric and how you will evaluate your model.