We’ve seen both in class and from previous blog posts some of the many different applications of principal component analysis - morphable face models, face recognition, mass spectrometry, financial analysis, etc. All these involve some fairly objective goals that one would naturally go about solving using matrix decomposition techniques - mainly due to the imperatively mathematical nature and multidimensionality of the dataset.

However, upon doing some research on the internet I stumbled upon some very interesting, and certainly more subjective uses of PCA: judging human beauty and wine quality. These applications might seem unusual because the subjects of comparison usually involve a great deal of human discretionary and sensory perception.

The matter of judging human beauty with artificial intelligence is, of course, only possible if the majority of humans share a similar notion of beauty. This would allow computers to learn patterns in judging facial attractiveness by analyzing images of faces. To some degree, the authors of the given research paper have suceeded in teaching computers to reasonably judge human beauty - correctly classifying about 75-85% of the images.

According to the article, PCA was used to capture variation in pixel data across all the faces in terms of several independent components - such as overall faceshape and delineation from hair, or the finer aspects of facial features - the latter of which was found to better correspond to attractiveness ratings by humans. Some other interesting finds from the study were that average faces were not the most attractive, and that deviation from the average had some correlation to beauty.

Fig 1. Analysis using principal components associated with delineating finer aspects as opposed to gross face shape

In the case of wines, wine appellation is typically based on producers’ sensory analysis. According to several articles, modern sensory methods have been developed for industry judge panels that involve applying PCA to sensory data from industry quality experts. The full process takes into consideration components such as aroma profiling, chemical analysis, as well as perceived quality bias. The results of these studies offer a detailed analysis that accommodates the wine industry professional.

References:

http://en.wikipedia.org/wiki/Principal_components_analysis

http://www.femininebeauty.info/news.php/weblog/comments/machines-judging-beauty/

http://cat.inist.fr/?aModele=afficheN&cpsidt=3158936

http://www.ajevonline.org/cgi/content/abstract/46/1/5

http://www.blackwell-synergy.com/doi/abs/10.1111/j.1745-459X.2001.tb00302.x