New machine learning algorithm produces "near-perfect" fake human faces

Face-synthesizing algorithms often struggle with facial details like eyes and teeth. These features sometimes get pinned to a fixed spot as a head turns, resulting in an uncanny valley dweller.

A new algorithm, StyleGAN2, fixes this problem and produces "eye-poppingly detailed and correct images." It can also generate never-before-seen cars, churches, and animals.

Image: YouTube Read the rest

Very fun algorithm guesses your name with surprising accuracy

Nathan Yau created this fun and fascinating name guessing algorithm. You select "male" or "female," the decade you were born, and then type in the first letter of your name. I tried more than a dozen times for people I know and it nailed it on the first letter about 80% of the time. On those that it screwed up after the first letter, it got it right after I entered a second letter. From the project description:

This is based on data from the Social Security Administration, up to 2018. It’s relatively comprehensive, but there are a few limitations. First, it’s data for the United States, so the numbers don’t really apply elsewhere. Second, the SSA doesn’t include names with fewer than five people in a year, so the chart doesn’t cover more unique names. Third, there were no Social Security Numbers before 1935, so the name counts are fuzzier for years before that.

But like I said, the data still has a wide range. I aggregated the annual data by decade and calculated percentages by dividing name counts by total number of Social Security Numbers provided.

Before you enter anything, the chart shows the most popular names for the given sex and decade. Then as you enter a name, the chart shows conditional probabilities. The more information you give it, the stronger the guess.

"Guessing Names Based on What They Start With" (FlowingData) Read the rest

Excellent video about why the SHA 256 hash algorithm is so cool and useful

SHA 256 is an algorithm that takes a digital input of any length and returns a string of 256 bits (typically converted to 64 hexadecimal digits). It's a one-way algorithm, which means there's no known way to practically retrieve the input from the output. As far as anyone knows, there has never been an instance of two different inputs having the same output, which means the hash of an input is a reliable unique digital fingerprint.

In this 6-minute video, Matthew Weathers explains why SHA 256 is "useful for digital signatures, cryptography, authentication, and is a central part of the Bitcoin protocol." Read the rest

CGP Grey explains how machines learn

YouTuber CGP Grey is known for breaking down complex subjects so that they can be understood by just about anyone. His most recent animated video is no exception. In it, he answers the question, "How do all the algorithms around us learn to do their jobs?," ie. how do machines learn? Good stuff.

He's even created a place on reddit to discuss it further.

Here's a footnote video to the first one:

Read the rest

#Elsagate: a subreddit that's sleuthing out the weird videos of Youtube Kids

Last month, James "New Aesthetic" Bridle published an influential essay exploring the prolific and disturbing video-spam that had come to dominate Youtube Kids, in which seemingly algorithmically generated videos endlessly recombined a handful of Disney characters and assorted others engaged in violent, abusive and even psychosexual conduct, over a soundtrack of a few repeated public-domain kids' songs, with all sorts of trickery designed to uprank them in Youtube's play-next, recommendation and search results -- keyword stuffing, duration-stretching and more. Read the rest

Paradise Papers reveal cozy relationship between Stubhub and Canadian botmaster/scalper kingpin

The Paradise Papers continue to expose the economically useless activity that late-stage capitalism rewards with titanic sums of money: today, it's the story of Julien Lavallée, a botmaster ticket-scalper who has harvested the lion's share of concert tickets from all over the world, laundering them for millions through a secret "top seller" program that Stubhub offers to anyone who can move more than $50,000 worth of tickets per year. Read the rest

Youtube Kids spammers rack up billions of views on disturbing, violent, seemingly algorithmic videos

James Bridle takes a deep dive into the weird world of Youtube Kids videos, whose popular (think: millions and millions of views) genres and channels include endless series of videos of children being vomited on by family members and machinima-like music videos in which stock cartoon characters meet gory, violent ends. Read the rest

UK press doesn't understand chemistry or Amazon, launches bomb-making panic

UK public broadcaster Channel 4 sparked a presswide panic with a story: "Potentially deadly bomb ingredients are ‘frequently bought together’ on Amazon." Read the rest

15 sorting algorithms visualized in 6 minutes

Timo Bingman created a demo program for sorting algorithms called "The Sound of Sorting, which both visualizes the algorithms internals and their operations, and generates sound effects from the values being compared." It's a visual and aural treat! Read the rest

Entrancing avant-garde music video generated by algorithm

Damien Henry, co-inventor of Google Cardboard, trained a machine learning algorithm using footage shot from a moving vehicle and then had the machine generate this beautiful video.

"Graphics are 100% generated by an algorithm in one shot. No edit or post-processing," Henry writes. "Except the first one, all frames are calculated one by one by a prediction algorithm that tries to predict the next frame from the previous one."

The soundtrack is the Steve Reich masterpiece "Music for 18 Musicians."

Read the rest

Messy: When automated anti-disaster systems make things worse, and what to do about it

"Undercover Economist" Tim Harford (previously) has a new book out, Messy, which makes a fascinating and compelling case that we are in real danger from the seductive neatness of computers, which put our messes out of sight, where they grow into great catastrophes. Read the rest

Weapons of Math Destruction: invisible, ubiquitous algorithms are ruining millions of lives

I've been writing about the work of Cathy "Mathbabe" O'Neil for years: she's a radical data-scientist with a Harvard PhD in mathematics, who coined the term "Weapons of Math Destruction" to describe the ways that sloppy statistical modeling is punishing millions of people every day, and in more and more cases, destroying lives. Today, O'Neil brings her argument to print, with a fantastic, plainspoken, call to arms called (what else?) Weapons of Math Destruction.

What if school was out, forever?

Today a future without schools. Instead of gathering students into a room and teaching them, everybody learns on their own time, on tablets and guided by artificial intelligence.

Flash Forward: RSS | iTunes | Twitter | Facebook | Web | Patreon | Reddit

In this episode we talk to a computer scientist who developed an artificially intelligent TA, folks who build learning apps, and critics who wonder if all the promises being made are too good to be true. What do we gain when we let students choose their own paths? What do we lose when we get rid of schools?

Illustration by Matt Lubchansky.

▹▹ Full show notes Read the rest

Video of 15 sorting algorithms, with "audibilization"

It's fascinating to see and hear the distinctive personalities of the different sorting algorithms in this 5-minute video. My favorite is the bogo sort at the end, which sounds the best but seems to do a poor job of sorting

Visualization and "audibilization" of 15 Sorting Algorithms in 6 Minutes.

Sorts random shuffles of integers, with both speed and the number of items adapted to each algorithm's complexity.

The algorithms are: selection sort, insertion sort, quick sort, merge sort, heap sort, radix sort (LSD), radix sort (MSD), std::sort (intro sort), std::stable_sort (adaptive merge sort), shell sort, bubble sort, cocktail shaker sort, gnome sort, bitonic sort and bogo sort (30 seconds of it).

Sorting videos are popular on YouTube. I like these ones that show robotos competing to sort balls from darkest to lightest:

Read the rest

Snowshoeing: small-batch spam that's less targeted than spear-phishing

Snowshoe spam has a "small footprint" -- it is sent is small, semi-targeted batches intended to sit below the trigger threshold for cloud-email spam filters, which treat floods of identical (or near-identical) messages as a solid indicator of spam. Read the rest

Stolen-card crime sites use "cop detection" algorithms to flag purchases

Cops covertly buy stolen cards from underground sites to figure out where they came from, and so these sites implement security measures that try to figure out whether a purchaser is an undercover cop, and refuse to sell to them if they trip a positive result. Read the rest

Computer-generated camouflage for the physical world

MIT researchers have demonstrated an algorithm that analyzes photos of a real world scene and then generates an incredibly-effective camouflage pattern to wrap objects later placed in that location. From MIT News:

According to Andrew Owens, an MIT graduate student in electrical engineering and computer science and lead author on the new paper, the problem of disguising objects in a scene is, to some degree, the inverse of the problem of object detection, a major area of research in computer vision.

"Often these algorithms work by searching for specific cues — for example they might look for the contours of the object, or for distinctive textures." Owens says. "With camouflage, you want to avoid these cues — you don't want the object's contours to be visible or for its texture to be very distinctive. Conceptually, a cue that would be good for detecting an object is something that you want to remove.”

"Custom Camouflage" Read the rest

More posts