Science and technology micro-summaries for July 16, 2019

in #rsslog3 years ago

Unstructured active learning may be important for artificial intelligence; Adversarial attacks drive AI to the next level; Reliability issues for Facebook and Instagram; Identifying people - at a distance - by their heart beat; The first-ever photograph of quantum entanglement

Straight from my RSS feed:
Links and micro-summaries from my 1000+ daily headlines. I filter them so you don't have to.


pixabay license: source.

  1. A SEPARATE KIND OF INTELLIGENCE - Another talk in the "Possible Minds" series. In the linked video and transcript, Allison Gopnik talks about applying the concept of Life History to intelligence. In general, the theory of life history is a way to explain an organism's individual characteristics by its life-cycle characteristics like the duration of active parenting, number of offspring, lifespan, etc... This theory has been used to suggest explanation for things like the immune system, and Gopnik suggests that it may also be useful to explain intelligence. In particular, there are indications that organisms with longer childhoods have more intelligence. She extends this idea by suggesting that this knowledge may lead to useful insights in the realm of AI creation. As with previous Possible Minds posts, the talk is followed by a free-wheeling discussion among a number of panelists. This series has also been covered here:

  2. The Upside of Adversarial Attacks - Machine learning is vulnerable to disruption by specially crafted alterations to its data that don't require the attacker to have access to the machine. This form of attack, known as an adversarial attack, doesn't work on humans because humans are capable of higher order interpretation. The discovery of adversarial attacks shows that AI systems that do things from autonomous driving to facial recognition and health care diagnostics can all be misled by a hostile actor. The benefit of the discovery, however, is that these attacks can also be used to improve the AI algorithms in an iterative, co-evolutionary process.

  3. Facebook and Instagram are suffering from far more downtime than rival social networks - According to data from Downdetector, Facebook and Instagram have been going down more in the last year than they did in the past, and they have also been going down more than rival social media sites like Twitter and Snapchat. Facebook declined to comment on the outage data, but issued a generic statement that they work hard to maintain availability for their billions of world-wide subscribers.

  4. The Pentagon has a laser that can identify people from a distance—by their heartbeat - Gait detection and facial recognition have both been used to identify a person at a distance, but they don't uniquely identify an individual. Now the Pentagon says that they have a laser that can operate from up to 200 meters with 95% accuracy to identify a unique cardiac signature for a person. The signature can't be disguised, and remains stable over time. Researchers report that technology improvements will make it possible to increase both the distance and accuracy of the technique. Gait analysis was previously covered in Interesting Links: May 27, 2019, on Memorial Day. h/t Bruce Schneier.

  5. STEEM The First Ever Photograph Of Quantum Entanglement - In this post, @kralizec reports on an article in the journal, Science Advances describing the first time that researchers have photographed the phenomenon of quantum entanglement. The feat was accomplished by Paul-Antoine Moreau by shooting photons through a specialized material and capturing the image with ultra-high speed cameras that are capable of photographing a photon at the moment of entanglement. (@kralizec will receive 5% of the rewards from this post.)

In order to help make Steem the go to place for timely information on diverse topics, I invite you to discuss any of these links in the comments and/or your own response post.

About this series
Note: Sharing a link does not imply endorsement or agreement, and I receive no incentives for sharing from any of the content producers.

Follow on steem: @remlaps-lite, @remlaps
If you are not on Steem yet, you can follow through RSS: remlaps-lite, remlaps.

Thanks to SteemRSS from philipkoon, doriitamar, and for the Steem RSS feeds!


I often find either reporters misunderstanding research, or researchers themselves.

"This form of attack, known as an adversarial attack, doesn't work on humans because humans are capable of higher order interpretation."

This is clearly not true. While some humans capable of some human trait sometimes express that trait, it is demonstrable that all of the people do not do so all the time. The statement should actually be '...sometimes doesn't work on humans because humans are capable...'.


Good point. Thanks for the feedback! It's good to avoid overgeneralizing. Although, to be fair, I think it would be fairly unusual to find a human that can't tell (for example) that a stop sign with some stickers on it is still a stop sign. The main exception that comes to mind being very young children. That's the sort of alteration that the article is referring to.

Well, I was thinking more about misrepresentation of facts in propaganda articles. A lot of folks accept such reports without question, after long indoctrination, and little access to better information.