Due: Wednesday, April 9, 2025

---

Purpose: Consider alternative deep learning workflows and frameworks. This assignment is worth 50 points.

Reading:

• Chollet: Refocus on Chapter 3 Sections 3.5 and 3.6

Problems.

1. (34 points) Focus on the fundamental building blocks of a model in Tensor Flow by running the listings 3.13 through 3.21 in Section 3.5.4 (as well as the two code blocks that produce the plots) in a local Jupyter Notebook. You'll need to import tensorflow as tf first of course. It may be worthwhile to do run some of the examples prior to 3.5.4 in the previous parts of section 3.5 to get an idea of what the different pieces do.
    
   In Markdown cells, explain the purpose of these TensorFlow elements:
   • tf.ones
   • tf.matmul
   • tf.Variable
   • tf.GradientTape
   • What is the activation function?
   • What is the loss function?
   
   
   
2. (08 points) Run the pretrained ResNet101 model found in PyTorch. You can follow the steps shown in the linked PyTorch ResNet notebook along with the descriptions found here. However, don't run  the code blocks that access

   AlexNet

   or that contain just

   resnet

   as these produce pages of data we don't care that much about.
   
   As alluded to in class, you should run the model in the same way on three different images of your own: one with a single clear subject, one with at most three subjects of the same or different types (for example, a person in front of a car or motorcycle, a person with their dog, three friends in front of All Saint's, a dog and a monkey, you get the idea), and finally a picture with lots of subjects, which may or may not work well at all. Let's see what this model comes up with.
   
   
3. (08 points)  Find a paper or article online discussing the ethics or societal impact of the ChatGPT bot, or the GPT-3 model (or LLM models in general). It can be pro, con, or balanced.
   Summarize the arguments made by the author(s), and write a short response to those arguments, whether you agree, disagree, find the arguments unpersuasive, or just aren't sure. Include information about the paper such as title, authors, and a link to the paper or article.
   
   
   Here's an example of the sort of article you could use: ChatGPT is a bullshit generator. But it can still be amazingly useful.

Deliverables

• Send me a zip file containing all the ipynb files you've done. In your email, include a link to the HTML page you posted for first and second examples.
  
• Turn in your answers to written questions in class or in the box outside my door by 5:00 pm Friday
  
• Alternatively, email an electronic copy of written questions along with the rest.
  

---