week 11 - visual digital turn

Question 1

Quantification & The
Humanities
* Mainly limited by

Answer 1

Compute
▪ Digitizatio

Question 2

Gold and Klein1 define DH as the following

Answer 2

digital archives, quantitative analysis, and tool-building
projects

Question 3

AI

Answer 3

Consist of systems that can handle given tasks on their
own

• search algorithms, rule based system

Question 4

ML

Answer 4

omputer Algorithms that automate actions without
explicit programming. Can learn and improve

Clustering algorithms, K-Means, Gradient Boosting, PC

Question 5

DEEP LEARNING

Answer 5

Deep Learning is part of ML and makes use of ‘Deep’
Neural Networks

Convolutional Neural Nets, Recurrent Neural Networks,
LLM’s (ChatGPT)

Question 6

Supervised Machine Learning

Answer 6

Model learns by example
* Human labelled datasets
* Model learns based on examples
* Measures how ‘wrong’ it is and parameters get adjusted.
* Parameters start of randomly.

Question 7

Unsupervised Machine
Learning

Answer 7

Does not rely on labelled data
* Aims to identify structures or patterns in data.

Question 8

Training a Network - STEPS

Answer 8

1. Data goes through the model.
2. Calculate how ‘wrong’ the model is from the labels.
3. Calculate what parts in a network need to be adjusted to
   perform slightly better.
4. Change the parameters
5. Repeat until model no longer improves.

Question 9

Gradient Descent

Answer 9

: ‘Direction’ to reduce ‘wrongness’.

It helps AI models learn better by slowly improving.

It’s used in training almost all machine learning models (like ChatGPT, image recognition, etc.).

Question 10

CNNs (Convolutional Neural Networks):

Answer 10

Specialized for image processing

Identify features like edges, shapes, textures

Use layers of convolutions to detect increasingly complex patterns

Question 11

Wevers and Smits

Answer 11

focus on the potential of Computer Vision in the
Humanities.
* Shows three different ways to use CNN

Question 12

Wevers and Smits highlight three simple uses of CNNs in the humanities

Answer 12

Detecting
Medium (1)

Clustering
with CNNs (2)

Your own
classifier (3)

Question 13

detecting medium

Answer 13

They used CNNs to tell the type of media in an image — like whether it’s a photo, drawing, or printed ad. This helps researchers sort and study different kinds of visual content in archives.

Question 14

2. Clustering with CNNs

Answer 14

They grouped images that look visually similar (like sorting ads with the same layout or style) by using the features from a pre-trained CNN. This helps spot visual trends and design patterns in things like advertisements

*used pre trained CNN

Question 15

3. Your Own Classifier

Answer 15

They trained a custom CNN to recognize recurring image types in newspapers (like weather icons or political sketches). With only a few examples, the model got good at picking out these common image categories.

Question 16

What has traditionally been the main focus of Digital Humanities?
A) Visual content analysis
B) Text-based analysis using OCR and computational tools
C) Audio and video processing
D) Network security

Answer 16

b

Question 17

What is meant by the “visual turn” in Digital Humanities?
A) Moving from text analysis to also include visual data
B) Focusing only on photographs in archives
C) Using VR technology for humanities research
D) Ignoring textual data completely

Answer 17

a

Question 18

Which type of machine learning uses labeled data to train models?
A) Supervised Learning
B) Unsupervised Learning
C) Reinforcement Learning
D) Deep Reinforcement Learning

Answer 18

a

Question 19

What is a Convolutional Neural Network (CNN) primarily used for?
A) Text translation
B) Image processing and pattern recognition
C) Audio synthesis
D) Database management

Answer 19

b

Question 20

In Wevers & Smits’ study, what shift did the CNN detect in newspaper images after 1900?
A) More illustrations than photos
B) A surge in photo usage, overtaking illustrations by 1925
C) No change in image types
D) A decline in image usage altogether

Answer 20

b

Question 21

What does clustering with CNNs involve?
A) Grouping visually similar images based on extracted features
B) Labeling images with text metadata
C) Translating images into text descriptions
D) Removing noise from images

Answer 21

a

Question 22

Which dataset did Wevers & Smits use to analyze advertisements?
A) CHRONIC
B) SIAMESET
C) ImageNet
D) MNIST

Answer 22

b

Question 23

. What is a limitation of clustering CNNs mentioned in the study?
A) It can’t process color images
B) It struggles with abstract or text-heavy ads
C) It requires millions of labeled images
D) It only works on handwritten documents

Answer 23

b

Question 24

How many labeled images did Wevers & Smits use to retrain their custom classifier?
A) About 100
B) About 500
C) About 10,000
D) Over 100,00

Answer 24

b

Question 25

What is one major challenge in using CNNs for humanities research? A) CNNs cannot process black and white images B) Training data bias can affect results C) CNNs only work on video data D) Lack of computing power

Answer 25

b

Question 26

What is the main focus of your research on colonial Korean print shops? ) Analyzing the sound quality of radio broadcasts B) Studying how the visual style of printed texts changed over time C) Translating Korean texts into English D) Mapping geographic locations of print shop

Answer 26

b

Question 27

Which question is central to understanding print shops’ design choices? A) How did they adapt to historical events? B) How fast were their printing machines? C) What languages did they print in? D) How many workers did each shop have?

Answer 27

a

Question 28

What types of changes are you trying to detect in the printed texts? A) Changes in paper quality B) Changes in fonts, layouts, or printing style C) Changes in ink composition D) Changes in shipping routes

Answer 28

b

Question 29

. How large is the dataset you are working with from Korean newspapers? A) About 1,000 pages B) Over 150,000 pages C) 10,000 pages D) 500 pages

Answer 29

b

Question 30

Which tool do you plan to use for parsing the layout of newspaper pages? A) OCR Only B) SAM (Segment Anything Model) C) Manual inspection only D) Excel spreadsheets

Answer 30

b