CAOS 2025 - 1 | Rovereto, May 7-9 | Wilma Bainbridge

Value of the Information & Strength of the Argument

The talk provides substantial value by presenting a robust, replicable methodology for quantifying memorability and demonstrating its predictive power in naturalistic settings. The argumentation is solid, grounded in large-scale experiments and statistical models. Bainbridge carefully addresses potential confounds (e.g., context, size) and acknowledges limitations, such as the unexplained role of interestingness. The progression from online experiments to real-world validation strengthens the claims.

Scientific Rigor, Source Quality, Title Accuracy

The scientific rigor is high, with peer-reviewed publications underlying the presented work. The speaker cites specific databases (THINGS) and tools (ResMem) and provides a link to the workshop for further information. The title accurately reflects the content, focusing on the predictability of memory. No public comments were provided for analysis.

The title accurately reflects the core theme: the predictability of memory across individuals and contexts.

8/10

The talk presents peer-reviewed research from a recognized expert, with large-scale datasets and transparent methodology. The claims are supported by empirical evidence, though some results (e.g., the role of interestingness) are still under investigation.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 00:00Introduction by the session chair, welcoming attendees to the 16th CAOS workshop.
• 02:00Wilma Bainbridge begins her talk, thanking the committee and sharing her history with CAOS.
• 05:00Interactive demonstration: audience claps when recognizing repeated faces, illustrating memorability.
• 10:00Definition of memorability and evidence for its consistency across people and stimuli types.
• 20:00Introduction of ResMem, a deep learning model for predicting image memorability.
• 30:00Study at the Art Institute of Chicago: predicting memory for artworks in a real-world setting.
• 40:00Results: neural network predictions significantly correlate with human memory, while beauty and emotion do not.
• 50:00Study using the THINGS database: over 1 million memory ratings for 26,000 object images.
• 1:00:00Semantic dimensions influence memorability more than low-level visual features.
• 1:10:00Discussion of whether distinctive or prototypical items are more memorable, and future directions.

• CIMeC CAOS Workshop — Workshop page with information about the event and speakers.

• Bainbridge Lab Publications — List of peer-reviewed articles supporting the presented findings.

• No discordant sources identified — No conflicting sources were mentioned in the talk.