User:Jadrian Miles/OKRs/Spring 2012

Today's Date: 12/10

• Meta-objective: make measurable progress toward the dissertation.
  • Key result: curve-clustering / macrostructure-initialization chapter is complete by 04/27.
  • Key result: chapter outline for the first-pass "real" problem is complete by 04/27.
    • Neither of these got done this semester.
• Objective: Finish the toy problem.
  • Key result: Success is defined by 01/10.
    • Demonstrate that the ground-truth configuration is a local optimum of the chi-squared goodness-of-fit measure.
    • Write a simple local search optimizer that improves goodness-of-fit—no need to go overboard tweaking it, though; a simple demonstration of improvement by geometrical adjustment, splitting, and joining is sufficient.
    • Wrote success definition 01/08--10.
  • Key result: Principled solver meets success criteria by 01/17.
    • Wrote an optimizer and wrapped up toy problem on 01/16.
• Objective: Submit a paper to MICCAI.
  • Key result: Potential paper subjects are defined by 01/17.
    • Subject: Streamline clustering based on chi-squared reconstruction error of coherent structures.
    • Done before 01/17.
  • Key result: Paper draft is ready by 01/27.
  • Key result: Paper is submitted by 03/01.
  • Key result: Additional OKRs for after the submission are defined by 03/02.
  • Can't write the code and the paper in time for the MICCAI deadline (decided 02/17). The rough outline (done before 01/27) is still a good guide though.
• Objective: Upgrade the toy problem to a simple 3-D case.
  • Key result: Clustering objective function is coded up by 01/27.
    • Objective function defined in terms of chi-squared before 01/27, but not implemented.
  • Key result: success defined for clustering by 02/24.
    • Clustering must perform better with respect to streamline reproduction, as measured by chi-squared goodness-of-fit, than at least two other competitor clustering algorithms. May want to compare to the Lawes cortical-ROI-based approach too.
    • Success defined late, 03/09.
  • Key result: success defined for real problem by 02/24.
    • There are four demonstrations of success I think are compelling. Demonstrations of success with synthetic data may be intermediate, but to finish the dissertation I think the following is required:
      1. Demonstrate (by pictures) specific challenging regions in which the multi-scale approach reconstructs structures better than two competitors.
      2. Demonstrate (numerically) that image reconstruction, as measured by chi-squared goodness-of-fit, is better than two competitors.
      3. Demonstrate (in prose, or maybe with a quick user study) that selection tasks for specific brain structures are easier and more precise than with ROI+tractography methods.
    • Success defined late, 03/09.
• Objective: Submit a paper to a journal in 2012.
  • Key result: Potential paper subjects are defined by 03/06.
    • The clustering paper described by the above definition of success would be a good submission.
    • Would it be possible to write up the toy problem as a case study for doing the Rician-to-Gaussian transformation for chi-squared analysis?
    • Paper subjects written up on time.