Weekly update for September 18

Sample List

• This
• Is
• Sparta!!
• (And Sparta in Italics)
• (And Sparta in bold)
• (And Sparta in bold Italics)
• (And Sparta with a link)

Sample Image

Sample Table with Equations

For me (Guillermo), it’s been a nightmare to type equations and tables in RMarkdowns. But you can try…

Judge education and outreach

• Phone call with Alicia and some other awesome people:
  • Judge Barbara Hervey
  • Judge Stephanie Domitrovich
  • Felice Schur, Assoc. Dir., Judicial Div., ABA
• Devise a course to teach statistics to judges
  • Use cases where statistical testimony has been / could be used
  • Focus on concepts, not equations
  • Ballistics, fraud, paternity, discrimination, mixture DNA
  • e.g. U.S v. Glynn (2008) (ballistics)
• Work with Hal Stern at UCI

Developing more human factors studies

• Using ULTR language in the evidence conditions
• Describing the ULTR language as approved by the FBI or not
• Telling participants the known error rates or not (where applicable)
• More evidence conditions: mixture DNA, ballistics, fingerprints
• Including the statement below as a condition:

Even though these two items are indistinguishable, we do not know how many other items would have also been indistinguishable from the crime scene evidence."

• Work with Bill Thompson at UCI.

Soyoung’s slides

• Working on many comparisons between known matches (replicates)
• There are 10 possible pairwise comparisons for 5 replicated images
• Example. Comparison between rep2 and rep3

Soyoung’s slides

Soyoung’s slides

• Example. Comparison between rep2 and rep3

Example

Soyoung’s slides

• Get the results of 20 comparisons (2 pairs)
• The goal is to find similarity attributes as many as possible
• Euclidean distance of centers from 3 circles,
• Average of Overlap features,
• Standard deviation of rotation angle estimations.

Soyoung’s slides

Image Alignment

Forgotten Image from Last Week

London With Nodes

London As Graph

Jason’s Slide

• Re-implemented, then re-re-implemented the above model
• Attempted to code poorly-specified steganalysis models
• Learned that being a Bayesian in neural network-land is pretty darn simple
• Introduction to Neural Networks for Statisticians

Heike’s slide

• visit from Srinivasan Rathinam from LA PD
• processing scans from Berettas
• upload Phoenix data to local DB

FlashID Exported Measurements

• 3 writers, 3 documents each (REU data: London and Cedar letters)

FlashID Exported Measurements

FlashID Exported Measurements

The Plan

• Each file will have a number of columns that depend on the graph being measured.
• Moving forward
  • Extract grapheme counts
  • Re-create the FlashID scoring
  • Grapheme importance & feature importance (number of features varies by grapheme type)
• This week: Science Bound Forensics Session (Saturday, Sept. 23)
  • Overview, handwriting, and fingerprints