Computer Science
Work description
The Attention Atlas is a visual analysis tool designed for the multi-level exploration of attention patterns in Transformer models. The causal interpretation of attention is contested in the literature, and the scientific contribution of this research is to provide empirical evidence on an open question: Do scaffolding mechanisms at the interface measurably alter analysts’ reasoning (epistemic strength of conclusions, cross-checking, calibration), or do they merely affect usability?
Academic Qualifications
Undergraduate or graduate student in Computer Science, Computer Engineering, Data Science, Artificial Intelligence, or a related field
Minimum profile required
Knowledge of Python; strong oral and written communication skills in Portuguese and English; availability to work on-site (FCUP)
Preference factors
Experience with user studies or with the tool itself; familiarity with pre-registered protocols; experience with qualitative coding.
Application Period
Since 30 Jul 2026 to 12 Aug 2026
Centre
Advanced Computing Systems