Thanks for the link! This looks really relevant and I look forward to reading it in detail!

It’s in the last post of the thread. 😊

These are the first student-led data to come from my Barnard lab, and I'm incredibly proud of Sage and Maria for leading this effort. Check it out at Journal of Vision: jov.arvojournals.org/article.aspx... 8/8

High semantic information actually facilitated detection, while mildly impairing categorization. It seems that high semantic information can give you a good sense that there is something meaningful in the stimulus. 7/

High visual information impaired both detection and categorization, suggesting that physical information may create a bottleneck to scene processing. Semantic information showed a more interesting pattern. 6/

We selected high- and low- information scenes for both semantic and visual complexity, and asked observers to do a detection task (i.e., scene versus noise) or a categorization task. 5/

Interestingly, these metrics are not strongly correlated! 4/

But images can have a lot of information in their pixels, but be visually simple (for example, patterns in carpets). So we assessed semantic complexity from human scene descriptions. Using NLP, we assessed the complexity of scene descriptions. 3/

How to operationalize information? Although computing entropy for scenes is computationally intractable, we used compressibility as a proxy. We took 67,000 photographs in RAW format and assessed relative file size when the images compressed. More compression=more redundancy=less information. 2/

I am a student at Brown in communication with a friend at RIBC.

There is an urgent need for blood donations to support RI. Donations in NY, NJ, MN, NB, DE or NJ can help supply RI.

Spread the word.

www.ribc.org/donate-blood/

Yeah, I think about that too. Hard to find a solution, given the incentive structures.