The IARPA (Intelligence Advanced Research Projects Activity), an agency of the US Director’s Office, has a mission. ” Pushing the boundaries of science to develop solutions that enable the IC (intelligence community) to do its job better and more effectively for national security.. HIATUS is a research program that aims to authenticate authors of a text and ensure their privacy through human-interpretable algorithms.
IARPA initiates research activities and the results communicate with its IC clients who themselves establish innovative technologies as a result. The four main areas of research where it invests are artificial intelligence, quantum computing, machine learning and synthetic biology.
The HIATUS program, using the underlying framework, is a descriptive feature of the text man
Whether spoken or written, linguistic elements differ from one person to another, words, the organization of sentences, their content can express who uttered or wrote them.
Timothy McKinnon, the Heights program manager, told NextGov in an interview:
“For a little context, if you have 100 different people, and you ask them to describe something simple – like how to open a door – in two sentences or one sentence, you can probably get about 100 different answers. Contains potentially used by the author’s feature system.
Every day, lots of texts are written by anonymous writers, people or machines. Timothy McKinnon noted that most of these documents contain linguistic elements that can be used to identify who wrote the information or to protect the authors’ identities if the attribution puts them at risk.
“With attribution, we identify stylistic features. So it can detect things like word placement and syntax who wrote a given text. Think of it as your written fingerprint. What features make your writing unique? In this way, the technology will be able to detect these fingerprints in contrast to a corpus of other documents and compare them if they come from the same author. In terms of privacy, the technology will find ways to change the text so that it no longer looks like a person’s handwriting. A
Currently, there are three ways to authenticate the author of a text: Language experts can do it by analyzing the text, anyone can use Machine learning, Including logistical regression or the use of a Bayesian model, but according to Timothy McKinnon, these methods will not be valid for all texts. The third option is to use a neural language model but they are not sufficiently explanatory.
“The problem with these models is that although they work very, very fast and very well, we don’t really understand what’s going on inside. They’re very complex.
And so what HIATUS wants to do, among other things, reveals some of the reasons behind the behavior of these patterns, so that when we do authorship attribution or privacy, we really understand why the system behaves like that, and will be able to verify that It is not detecting false information and is doing the right thing. A
The goal of the HIATUS program is to develop new human-usable systems to protect authorship features and authors’ privacy by identifying and exploiting linguistic fingerprints that can be interpreted and acted upon in different languages. It should last 42 months, from 30 September, 2022 to 29 March, 2026, approximately, BAA (call for proposals was published 25 February.