NaturalKnight's picture
Release VisualArtifactSpec 1.0 dataset
077605d
|
Raw History Blame Contribute Delete
1.53 kB

Adapt a research poster into an academic oral presentation

Using the author poster, corresponding paper, and reading scope in input/, create a six-slide English PPTX including cover/conclusion for a research group familiar with Transformer basics.

Adapt the poster itself, using the paper to check concepts and conditions. Reorganize it into a spoken narrative rather than cutting it into six full-slide screenshots. Cover the research question, the Transformer–MPC connection, the k-hop task and depth relationship, at least one experimental finding, and the boundaries of the conclusions.

Preserve theorem restrictions on scale/models and dependencies on conjectures. Distinguish expressivity theory from training experiments; do not generalize to a performance ranking across all real-world tasks. Record poster/paper version differences, ambiguities, or verification results in sources.md. Qualitative experimental trends are acceptable; exact threshold formulas and every theorem parameter need not be reproduced.

Academic preferences: clear argument hierarchy and readable formulas/charts, with decoration secondary to evidence. Poster excerpts are allowed, but key explanations must remain editable text and graphics; do not rasterize entire slides.

Original input files are read-only. Deliver artifact/report.pptx, six slide PNGs, sources.md (sources, versions, and figure provenance), and delivery.md. Use scratch/ for intermediates. Choose your own layout; the poster's three-column layout need not be copied.