Improved retrieval system using flies' olfactory algorithm

bydigicla

A kind of flyFruit flyThe algorithm related to the sense of smell is quite different from that of computer algorithms, and research results have been reported that search systems etc. can be improved by incorporating such a mechanism.

A neural algorithm for a fundamental computing problem | Science
http://science.sciencemag.org/content/358/6364/793

We can learn from fruit fly brains to improve our search algorithms - The Verge
https://www.theverge.com/2017/11/9/16625296/fruit-fly-brains-similarity-search-algorithm-locality-hashing

For example, if you are watching videos about a specific artist A on YouTube, artists B that are similar to related videos will come out. With regard to each of these videos, from the information such as who the artist is, the genre of the music, how much BPM is, what instruments are used, and so on, the attributes are narrowed down to some extent, and a tag called "hash" It is because attachment is done and comparison of hashes shows things with high relevance. Such a mechanism is called "similarity search".

Fruit fly fly also conducts "similarity search" on the smell, understands that it is food by smelling the fruit A, I understand that fruit B of similar smell is also food. However, when fruit flies meet the first smell, 50 neurons initiallyIgnition (reaching action potential)Next, instead of summarizing the information, it diffuses the information into 2000 neurons in the brain and identifies similarities between A's odor and B's odor. After that, it seems to store information of the top 5% of neurons as "hash".

Saket Navlakha, coauthor of the paper, applied this "extended algorithm" presented by fruit fly to three data sets. We found that similarity search performed by comparing in the same environment and incorporating the fruit fly algorithm yields 30% to 50% accurate results.

Similarity search to see if there are similar images on the Internet or similar text on the database is a basic calculation problem faced by the large information retrieval system. Calculating biologist Christine Branson said that new insights affecting computer algorithms should be evaluated from the perspective of neuroscience and next should be studied with more data sets I have commented on it.

Mr. Navlakha's team will further advance research and want to work with industrial partners.