Nist Scans: Observed Type I error rate
Nist Scans: Type II Error rate and Failed Tests
Updates
- CSAFE scans: Testing setup ready
- Results are not much better at the moment than NIST scans, in terms of FP and FN error rates.
- Investigating, possible problems of missing values in signature, smoothing and excessive failing of different shift comparisons
Kiegan
Updates
- Chapman and Hall book: ‘Officially’ happening! Positive reviews.
- Finishing chapter on analysis of evidence by end of March
- Trace evidence case studies
- DNA case studies (non-mixture)
- NIJ Graduate Fellowship: Submitted over break. :)
- Groove Identification: Looking into median absolute devation, iterative robust linear model fits
Nick
Updates
- Removed dependency on “imager”
- Cleaned up image to fill holes that shouldn’t be there
- Redid plotting to make it a little faster and changed little red circles.
Cleaning Effects
Cleaning Effects
Cleaning Effects
New Plots
Guillermo
Renaming files
Soyoung
Shoe matching score
- Mated score and Non-mated score
Reasonable model
- Kolmogorov-Smirnov test, Cremer-von Mises test, Anderson-Darling test, and so on..
- Start with Beta distribution, Beta(s1, s2).
- Get MLE of s1 and s2 and make some variations in each parameter.
- Test K-S test and find the reasonable shape parameters that pass the test.
- Get density of KM score and KNM score and take the ratio (SLR) at desired score.
- KM score : 559 reasonable sets of parameters passed K-S test out of 1005.
- KNM score : 2162 reasonable sets of parameters passed K-S test out of 4221.
Beta distribution with K-S test
- example model
Distribution of SLR
Amy
Updates
- NIJ Graduate Fellowship (submitted last week!!)
- STEMversity
- ~ mid-June (13 - 18)
- Middle and high school students
- Evidence collection/lab work - a few ideas!
- Handwriting
- IRB application for data collection
- Nov. 2017 Vastrick et al. handwritten numerals (Gary was involved)
- Jan. 2017 Johnson et al. handwriting and hand printing (NIJ funded - unavailable)
Martin
My C++ version of Matlab’s “imregister”
- Objective is to align images got with the EverOS 2D scanner
- Function works well for small images (around 1/8 of the actual dimensions)
- However, for larger images it’s not clear, since it takes so long
- This is not surprising, since each iteration computes pixel difference for almost 7 million pixels
- Working on techniques to accelerate this process, specifically, parallelization in C++
Jason
Bullet Matching Intervals
- BART intervals are still crazy, even after basing the intervals on predictions from 100 models rather than posterior predictive intervals from a single fit.
- High variability in the training algorithm, which is scary.
- This is probably due to insufficient burn-in time.
- Extended hyperparameter optimization work planned for this week.
Sam
Projects
- Judge training: draft waiting for feedback from Hal & Alicia, then will be sent to judges for feedback
- REU: phone interviews nearly complete! Project planning - data science focus, glass data & mock jury study data. Lecture/event/activity ideas welcome!
- ROpenSci Fellowship: CRAN task view draft complete. Feedback welcome! Going to be “rotating” on projects for 4-6 weeks, starting with glass.
Ganesh
Nist Scans: Observed Type I error rate
Nist Scans: Type II Error rate and Failed Tests
Updates
- CSAFE scans: Testing setup ready
- Results are not much better at the moment than NIST scans, in terms of FP and FN error rates.
- Investigating, possible problems of missing values in signature, smoothing and excessive failing of different shift comparisons
Kiegan
Updates
- Chapman and Hall book: ‘Officially’ happening! Positive reviews.
- Finishing chapter on analysis of evidence by end of March
- Trace evidence case studies
- DNA case studies (non-mixture)
- NIJ Graduate Fellowship: Submitted over break. :)
- Groove Identification: Looking into median absolute devation, iterative robust linear model fits
Nick
Updates
- Removed dependency on “imager”
- Cleaned up image to fill holes that shouldn’t be there
- Redid plotting to make it a little faster and changed little red circles.
Cleaning Effects
Cleaning Effects
Cleaning Effects
New Plots
Soyoung
Shoe matching score
- Mated score and Non-mated score
Reasonable model
- Kolmogorov-Smirnov test, Cremer-von Mises test, Anderson-Darling test, and so on..
- Start with Beta distribution, Beta(s1, s2).
- Get MLE of s1 and s2 and make some variations in each parameter.
- Test K-S test and find the reasonable shape parameters that pass the test.
- Get density of KM score and KNM score and take the ratio (SLR) at desired score.
- KM score : 559 reasonable sets of parameters passed K-S test out of 1005.
- KNM score : 2162 reasonable sets of parameters passed K-S test out of 4221.
Beta distribution with K-S test
- example model
Distribution of SLR
Martin
My C++ version of Matlab’s “imregister”
- Objective is to align images got with the EverOS 2D scanner
- Function works well for small images (around 1/8 of the actual dimensions)
- However, for larger images it’s not clear, since it takes so long
- This is not surprising, since each iteration computes pixel difference for almost 7 million pixels
- Working on techniques to accelerate this process, specifically, parallelization in C++