\(~\)
Objectives:
- Familiarize ourselves with influential literature in the field,
including the terminology, notation, and style commonly used in academic
writing.
- Gain a historical perspective on how important machine learning
ideas and best practices have developed over time.
- Connect concepts from academic papers to machine learning workflows
and approaches we’ve covered in class.
\(~\)
Directions (before class):
- Aim to spend approximately 1-hour reading the paper and focus on
gaining a high-level understanding of the authors’ contribution. That
is, direct your attention to the problem the authors are trying to solve
and their main conclusions, not the technical details.
- It is okay to skim some of the jargon and mathematical definitions
so long as you can extract the main ideas. Most professionals will do
this when reading a paper for the first time.
- It is okay to not understand parts of the paper, even after
re-reading. Even experienced researchers struggle to understand some
parts of technical papers in their fields.
- While reading the paper, aim to identify at least three interesting
discussion points. These could be key assertions made by the authors,
connections with course material, points where you were confused, or
examples you found particularly interesting. Prepare a formal discussion
question for each discussion point, but note that you likely won’t have
time to discuss all three points you identified.
\(~\)
Directions (during class):
- Start by having each person briefly explain their understanding of
the paper’s purpose, main findings, and intended contribution. Ask
questions when someone brings up a point you hadn’t thought of while
reading, and try not to repeat things that have already been
shared.
- Rotate through the discussion questions prepared by group members,
aiming to have everyone contribute to each discussion topic. Don’t feel
obligated to share one thought then never speak again, instead strive to
have an earnest conversation about each item for as long as the topic
remains interesting. It’s okay if you don’t have enough time to get
through all of the group’s interest items.
- When discussing an item, try to stay grounded within the context of
the paper and our course. This means references to specific evidence,
experiments, figures, or claims from the paper whenever possible.
\(~\)
Directions (after class):
- Submit a 1-2 paragraph reflection (at most one page) describing the
most important idea your discussion group took away from the paper, how
it connects to topics from class, and how the discussion changed or
deepened your understanding of the paper. Simply summarizing the paper
is not enough and will not receive full credit. Focus on what you
learned, what surprised you, or how your thinking changed from specific
moments in your group’s discussion. You are expected to make at least
one specific reference to something your group discussed.
- Fill out the peer feedback
form. Note that your responses will remain confidential.
\(~\)
What makes a good discussion item?
Good discussion items:
- identify something others would agree is important or
surprising
- connect an idea from the paper to one from our lectures, labs, or
homework assignments
- address how the paper’s arguments influence the way we develop or
evaluate machine learning models
- raise questions on which reasonable people could disagree
- point out a limitation or unanswered question within the paper
- evaluate whether the evidence presented by the authors is
convincing
Less effective discussion items:
- focus too narrowly on facts that can easily be looked up
- don’t leave room for differences in interpretation or
application