About the deep neural net and distinguishing an eyebrow from a face, why robots can be perceived as having a free will, and a future where our computers know us butter than we know ourselves, and will inform us when others are bored, inspired, hurt, grateful – or just not on the same page.
This Week in Context
Your Weekly Update on All Things Context, June 6 2014
There’s a flaw lurking in every deep neural net. A recent paper by Google’s deep learning team unveiled two flaws in the currently accepted understanding of how deep neural nets work. If you want to know whether they really do distinguish an eyebrow from a face, or how closely you resemble an open street, then have a look at this review of the publication. (via David)
As humans link free will to the mind, not the soul, robots as well could be perceived as having a free will. This Brown University study suggests that people could come to regard non-humans as having free will, if they come to believe that those actors (a sufficiently sophisticated robot, for example) have the capacity of independent, intentional choice.
Insurance company Allstate, which is already using their Drivewise hardware to measure and reward positive and safe driving behaviour, started experimenting with a – much cheaper – smartphone app to monitor driver behaviour. Available for iPhone and Android, but only in the US.
Mashable writes that Microsoft will launch a new smart watch that will not only be compatible with iOS, Android and Windows Phone devices, but will also have continuous heart monitoring capabilities. Great news for those in the eHealth and, of course, mood industries.
“And when our computers know us better than we know ourselves, they will help us to communicate better with one another. They will monitor our conversations and inform us when others are bored, inspired, hurt, grateful, or just not on the same page. They will encourage us to speak up when we are shy, stop us from sending e-mails when we are angry, and remind us of what makes us happier when we are sad.”
Interesting papers and research
‘Practical Machine Learning: A New Look At Anomaly Detection’ (O’Reilly Publishing) is an ebook by Ted Dunning and Ellen Friedman, available for free on the mapr website.
Interactive Collaborative Filtering (paper) – Traditional recommender systems used by Amazon or Netflix (i.e. collaborative filtering) are trained offline to discover relations between your past preferences and the preferences of other users. However, these systems have difficulties to cope with newly released items (e.g. a movie or book that has not been ‘liked’ by anyone yet). The dilemma of recommending new items to gain knowledge about user preferences, and not showing these to avoid annoying the user, can be approached from a game-theoretical point of view. (via Vincent S)
We’ve recently welcomed both Vincent Spruyt and Vincent Jocquet to the company. There’s more tech talent joining us soon, keep an eye on the Argus blog for the upcoming Q&A with Joren van Severen, who is joining Argus Labs as Data Scientist.
If you’re interested in being a part of such a diverse dream team and fit one of the profiles we’re still looking for, don’t hesitate to mail email@example.com with your resumé and a riveting cover letter. We’re checking our mail!
Enjoy the weekend!
If you’d like to get the highlights of what is moving and shaking in our industry delivered to your mailbox once a week, sign up here.