Stanford Seminar - Pushing the Boundaries of "Doing" Research Papers in Computing

()

Writing Style in HCI and Computer Vision Papers

Conventional style emphasizes clarity, elegance, and theoretical claims.
Typical qualitative paper structure includes abstract, introduction, related work, methods, findings, discussion, and design implications.

Introduces Stewart's book and its affect theory approach to capturing emotions and feelings.
Shares excerpts from the book to illustrate the writing style and focus on intensity, texture, and reader engagement.
Mentions reactions and critiques of scholars, highlighting the experimental nature and invitation to new forms of writing and reading.

Author experimented with Stewart's style for a chapter on the AIA outdoors, drawing from personal experiences and reflections on "rule."
Students had mixed reactions, some finding it enjoyable but tiring, others appreciating the standalone nature of the vignettes.
Author added a motivation section to the chapter, discussing the purpose of ethnography and HCI and encouraging readers' interpretations.
Author and colleagues used Stewart's style again in a study on external critique's impact on tech workers.
Converted interview data into scenes, foregrounding workers' stories without excessive authorial intrusion.
Interview excerpt described a tech worker's encounter with a stranger's criticism, highlighting hypervisibility and emotional numbness.

Computer vision papers are designed social and cultural interfaces created by disciplines and creating scientific truth.
Teaser images and tables of results visually represent state-of-the-art and reflect the attention economy.
Transition from applied math to quantitative empirical discipline, adopting tables of results from runtime benchmarking.
PDF paper format viewed as screen-based, with images often illegible when printed, requiring zooming.
Overall argument: computer vision papers are about attention and reflect the attention economy, with contributions reduced to their exchange value in terms of the attention they generate.

Collected 56 stories from the computer vision community during the pandemic to understand the emotional impact of the field's explosive growth.
Participants wrote non-fiction stories involving themselves, computer vision or machine learning, recent changes, and an emotional impact.
Stories were creative, powerful, and varied in length.
Took a more conventional HCI approach to analyze the findings, which was still relatively radical for the CVPR conference.
Wrote a rebuttal to address reviewers' comments about sample size and need for quantitative results, explaining qualitative vs. quantitative research approaches.
Paper accepted and received positive attention on Twitter, with CVPR Twitter account highlighting its importance in discussing community changes and challenges.

Suggests using qualitative methods to convey emotions and experiences, giving readers more agency in interpreting the paper.
Discusses challenges of getting experimental papers published and the need for a method paper discussing the meta-approach to writing papers.
Acknowledges current PDF format's potential for further exploration and innovation within its constraints.
Compares writing style to William Faulkner, acknowledging difficulty but arguing for its value and worthiness of effort.
Suggests style's suitability for conveying affect, the mundane, and ambiguity, citing its effective use in a paper on personal informatics.

Discusses challenges of writing academic papers that give readers more agency, acknowledging potential increased reading cost and loss of some readers.
Expresses uncertainty about making current ML benchmarks more accessible and usable for experimentation, despite their usefulness for competition and progress.