Your 15% journal club mark consists of two components:
- Skimming each paper and participating in weekly discussions (5%)
- Reading a single paper more thoroughly and being the "expert" on that paper (10%)
The week you have signed up to be the expert, read the paper and answer the following questions before Wednesday's class. Journal club discussions will take place Wednesday or Friday so that we've covered the topic a bit before discussing the paper.
The following questions should be used to guide the discussion during class. Since there are multiple experts each week, submitting a written response allows me to see your thoughts in case someone else has the same ideas and gets a chance to speak first.
You may use whatever file format you wish, and feel free to copy and paste relevant diagrams or quotes from the paper.
These questions should be answered concisely, but I am looking for more than a simple yes/no.
For example, for the question "Are the methods described in the paper still relevant today?" you might answer "yes, this paper outlined the Adam optimizer that is still the most popular choice for gradient descent in Tensorflow and Pytorch".
- What are the main contributions of this work?
- What is the context (timeframe, prior knowledge, etc) of this paper?
- How did the authors justify their conclusions?
With the power of hindsight, we can look back and see what impact each paper had on the field of machine learning. Most of them are widely cited. Consider the following:
- If applicable, try to find implementations of the algorithms described by the paper. These may be shared by the authors (for more recent papers), or implemented by others (for pre-2000s papers)
- Are the methods described in the paper still relevant today?
- What is your impression of the impact that this paper had on the field of machine learning?
Note that the first ("Deep Learning") and last ("Stochastic Parrots") papers chosen for discussion don't quite fit these questions, as they discuss general concepts rather than a specific technical method.
- What did you like about the paper?
- What was the most confusing or challenging to understand?
- Any additional thoughts?
Each of the three sections will be assessed on a 4-point scale, with an additional score for sharing your thoughts and responding to classmates' questions during the discussion period.
| Component | Score out of 4 |
|---|---|
| Paper summary | |
| Paper's influence | |
| Your opinions | |
| Discussion | |
| Total out of 16 |
| Score | Description |
|---|---|
| 4 | Excellent - thoughtful and creative without any errors or omissions |
| 3 | Pretty good, but with minor errors or omissions |
| 2 | Mostly complete, but with major errors or omissions, lacking in detail |
| 1 | A minimal effort was made, incomplete or incorrect |
| 0 | No effort was made, or the submission is plagiarized |
Note: I am aware of LLMs that can read PDFs and produce answers to these questions. If it looks like your answers were generated by an LLM, this will result in a score of 0.