Group items tagged

The blind application of machine learning runs the risk of amplifying biases present in data. Such a danger is facing us with word embedding, a popular framework to represent text data as vectors which has been used in many machine learning and natural language processing tasks. We show that even word embeddings trained on Google News articles exhibit female/male gender stereotypes to a disturbing extent. This raises concerns because their widespread use, as we describe, often tends to amplify these biases. Geometrically, gender bias is first shown to be captured by a direction in the word embedding. Second, gender neutral words are shown to be linearly separable from gender definition words in the word embedding. Using these properties, we provide a methodology for modifying an embedding to remove gender stereotypes, such as the association between between the words receptionist and female, while maintaining desired associations such as between the words queen and female. We define metrics to quantify both direct and indirect gender biases in embeddings, and develop algorithms to "debias" the embedding. Using crowd-worker evaluation as well as standard benchmarks, we empirically demonstrate that our algorithms significantly reduce gender bias in embeddings while preserving the its useful properties such as the ability to cluster related concepts and to solve analogy tasks. The resulting embeddings can be used in applications without amplifying gender bias.

"Sikoryak's latest graphic novel, Terms and Conditions, is like a great piece of conceptual art: dazzling to behold, if more than a little perplexing. Its panels of text and dialogue are word-for-word true to the 20,669-word terms and conditions, published by iTunes in 2015 (Apple has since adopted a lighter 7,000-word version)."

"This is one of the minor dangers of owning an Amazon Echo. Most anyone can activate the device by just saying the wake word, including Cartman. And in the season opener of South Park, that's exactly what happened. If someone watched that episode of South Park in the same room as an Echo, their Amazon shopping list was filled with random, gross items. This example shows the potential danger of having a voice-activated shopping assistant. It's easy to imagine a potential rogue advertisement, online or elsewhere, that could, in theory, say the right words to order a particular product - like a South Park box set."

"Google CEO Eric Schmidt says privacy isn't important, and if you want to keep something private, "maybe you shouldn't be doing it in the first place" (in other words, "innocent people have nothing to hide.")"

Last week, security news site KrebsOnSecurity went dark for more than 24 hours following what was believed to be a record 620 gigabit-per-second denial of service attack brought on by an ensemble of routers, security cameras, or other so-called Internet of Things devices. Now, there's word of a similar attack on a French Web host that peaked at a staggering 1.1 terabits per second, more than 60 percent bigger.

Fair and secure google docs