Weekly Update for July 9

class: center, middle, inverse, title-slide

# Weekly Update for July 9
### 2018/07/09

---

class: inverse
# About Slides

---
class: primary 
# New CSAFE slide template

We're now using [`xaringan`](https://github.com/yihui/xaringan)

What's changed:

- New person slide:

````
---
class: inverse
# Your Name
````

---
class: secondary

- New content slide with title:

````
---
class: primary
# Title of slide

Slide content
````

- New content slide without title:

````
---
class: secondary

Slide content with no title on slide
````

---
class: inverse
# Sample User

---
class: primary
# Sample Slide

- Sample Table:

| Team     | Probability | 
| : ------ | :---------- | 
| France   | 29%         |  
| England  | 27%         | 
| Belgium  | 26%         | 
| Croatia  | 18%         |

---
class: secondary

- If you are going to add an image, **create a directory** with your name within "images" folder. For example, "images/**guillermo**/sample_image.JPG"

---
class:inverse
# Miranda

---
class:primary
# Neural Networks

- Most recent model (classifying circles vs. triangles) using VGG16 has **94.8% accuracy** on test set!
- Next questions
    - Is VGG19 (more layers) better than VGG16?
    - Effect of different classes on accuracy? (e.g., quadrilateral, text)
    - How to handle multi-class images? 
        - Binary encoding/classification for each shape
        - Multi-label neural network
    
---
class: primary
# Architecture

- **Convolutional layers** learn local patterns by analyzing the image through a sliding window
- **Max-Pooling layers** combine patterns learned in single windows, downsampling huge feature maps and creating feature hierarchies of importance

---
class:primary
# Visualizing Filters

<img src = "images/miranda/aerosoles-pin-down-black-leather_product_9010211_color_72_crop2.png" width="70%"/>
 
---
class:secondary

1st Convolutional Layer
 <img src = "images/miranda/activations/1_conv2d_17.png" width="80%"/>
 
2nd Convolutional Layer
 <img src = "images/miranda/activations/3_conv2d_18.png" width="80%"/>

*A blank square means the filter's pattern is not found in the image

---
class:secondary

3rd Convolutional Layer
 <img src = "images/miranda/activations/5_conv2d_19.png" width="80%"/>
 
---
class:secondary

4th Convolutional Layer
 <img src = "images/miranda/activations/7_conv2d_20.png" width="80%"/>

---
class: inverse  
# Jenny and Ben

---
class: primary
# Shoe tread classifications

- 1645 classifications total

---
class: primary
# Realistic Print Experimental Setup

![3d](images/susan/IMG_20180703_083851.jpg)

---
class: primary
# Realistic Print Experimental Setup

![2d](images/susan/IMG_20180703_112534.jpg)

---
class: inverse  
# Susan

---
class: primary
# Realistic Print Experimental Setup

!["Stolen" Dirt](images/susan/IMG_20180704_084458.jpg)
---
class: primary
# Realistic Print Experimental Setup

!["Stolen" Dirt](images/susan/IMG_20180704_093237.jpg)

---
class:primary
# Truthiness

- 6 facts down, 6 more to go
- 10 pictures/plots per fact
- Formalized the experimental design

It's surprisingly hard to find specific bits of data, such as the length of a border with a body of water. :-/

---
class: inverse  
# Kiegan

---
class: primary
# Sorry I can't be there!

<img src = "images/kiegan/this-is-fine.0.jpg" width="50%"/>  
 
 
---
class: inverse  
# Martin

---
class: primary
# Brief update

- Working on alignment of two different pattern shoeprints
- Still struggling with the speed of processing the 4243x1645 pixels full size image
- Currently working on doing the processing using RCpp
- I also filtered and preprocessed some EverOS images from the database that Soyoung needs
 
---
class: inverse
# Sam

---
class: primary
# Summer activities thus far

- REU https://csafe-isu.github.io/reu18/
- Judge/lawyer training for ABA annual meeting in Chicago in August
- Working on submitting paper to JCGS before JSM

---
class: inverse
# Jimmy

---
class: primary
# Longitudinal Data Collection
* Document Development
   + Detailed outline of all changes made during collection
   + Updates on all procedure manuals
* Automated File development
   + Mat Scanner Files
   + These should be all remaining errors

---
class: inverse
# Nate

---
class: primary
# Bullet Changepoint
 - Changepoint analysis with Gibbs is currently running
    - Currently the samples is a two step, random walk Metropolis within Gibbs.
 - Preliminary results on next few slides
---
class: secondary
- Vertical lines show posterior means of the changepoints
- So far, so good.

<img src = "images/Nate/hamby1_cp.PNG" width="75%" height="75%"/> 
---
class: secondary
- This is what happens when only one shoulder(?) is present.

<img src = "images/Nate/hamby3_cp.PNG" width="75%" height="75%"/>
---
class: secondary
- The case when there is little data after a shoulder might be challenging.

<img src = "images/Nate/hamby5_cp.PNG" width="75%" height="75%"/>
---
class: secondary
- This looks fine so far. Starting values are dashed lines.

<img src = "images/Nate/hamby6_cp.PNG" width="75%" height="75%"/>
---
class: secondary
- When bad starting values are used.

<img src = "images/Nate/hamby6_cp_bad_svals.PNG" width="75%" height="75%"/> 
---
class: secondary
- When bad starting values are used but a high proposal variance is used.

---
class: inverse
# Guillermo

---
class: secondary

- Automating extraction of data from MatScan files with AutoIt
  + Make different windows active 
  + Need to come up with a way to move already processed files
- Set the right name for every file with R
- Fixing renamed vinyl pictures

---
class: inverse
# REU Students

---	
class: primary	
# Malisha Jones	
	
![My Colorful Plot](images/MalishaPlot.png)	
- Goal: To Create The Best Possible Graph! :)

---
class: primary
# Yolonda

- 10/54 documents completed
- Problems with telling letters apart

![](images/0026_4.png)
![](images/0026_4 edit.png)
Author 26, document 4

---
class: primary
# Da'Monie
Handwriting
- Assigned a total of 54 documents from 9 different writers to classify handwritten letters.
- In a week, I have completed 8/54 documents.

Errors: 
- Can only select a "square" region.
- Some letters overlap with the next letter in the word.

---
class: primary
# Alese
Making models and graphs using mock jury survey data.

---
class: primary
# Badiah 
- Jury project
  + Initial Verdict
- Fitting different models
- Changing ordinal data to numerical data

---
class: primary
# Carley
- Taking images of letters to train computer to classify letters
- Completed 12 out of 54 of my documents
- Issues
  - Letters that look the same
  
  <img src = "images/Carley_letter n.png" width="25%" height="25%"/>
  n
  <img src = "images/Carley_letter a.png" width="25%" height="25%"/>
  a
  <img src = "images/Carley_letter u.png" width="25%" height="25%"/>
  u

---
class: primary
# Mark Lancaster

Mind map regarding the evaluation of evidence

![mindmap](images/markjlancaster/mindmap.png)

https://mm.tt/1119875312?t=Z5g2sHX4AO