Thursday, January 29, 2009


Excessive business this week leads to lack of internet downtime and concomitant paucity of blog posts! Head about to explode! (But with very clean teeth, thanks to dentist visit this afternoon.)

Closing a few browser tabs:

Stephen Elliott interviews Margaret Cho at the Rumpus. Also courtesy of that excellent new internet time-waster: build your own virtual squid!

Hillary Clinton never got to meet America's Angriest Consul.

Courtesy of Bookforum, Elizabeth Quill at Science News on the science of human attractiveness and a great science-fictional twist:
Caring about specific features is one thing, articulating those preferences is another. Even people who consistently rate symmetrical faces as attractive, for example, have trouble identifying symmetrical faces. People just know an attractive face when they see it.

So does at least one computer. Eytan Ruppin of Tel Aviv University in Israel and colleagues have trained a computer to recognize what humans would rate as an attractive female face. The machine, described in January 2008 in Vision Research, automatically extracted measurements of facial features from raw images rated by study participants for attractiveness. It considered thousands of features and then condensed them. Then it went to work on a fresh set of faces. The computer predicted attractiveness in these new faces in line with human preferences.

Even more intriguing, the computer replicated at least one human bias. Symmetry studies often involve taking the right side of a face and mirror imaging it to create a full face or taking the left side and doing the same. Humans show a surprising bias; in two-thirds of cases, they prefer left-left images (from the point of view of the onlooker). Somehow, this bias must have been embedded in the original rankings the computer received because it also preferred these faces. But no one is sure why or how.

No comments:

Post a Comment