I’m about to make such a politically touchy statement that I want to preface it with a few words about my role.
A few people have tried to figure out (or think they know) whether I am pro- or anti-voucher schools. Waste of time.
I tell people that my wife doesn’t know whether I would give a thumbs up or thumbs down overall to the program that now allows almost a quarter of all the students in Milwaukee who are getting publicly funded education to attend private (almost all of them religious) schools.
I view myself like the color commentator on broadcasts of football games. I talk about how the teams are doing, how well the quarterback is playing and so on. But I don’t say whether I’m rooting for the Packers or the Bears.
This distinction is especially important when it comes to vouchers because the program has a 27-year history of being polarizing and controversial.
I admit I’m in favor of good schools and opposed to bad schools. There are quite a few good to excellent private schools in the voucher program. And there have been — and still are — bad schools. I have asked often for more than a decade whether enough was being done about them.
The 32 BC Mark Antony takedown: it began with a fake-news campaign masterminded by Octavian, complete with Tweet-like proclamations on ancient coins.
The Simon of Trent humdinger: in 1475 a prince-bishop in Italy set off a story that local Jews murdered missing 2-year-old Simon—and used his blood for rituals. Fifteen Jews burned at the stake.
The Benjamin Franklin special edition: he concocted an entire 1782 newspaper, peddling a fake story about Native Americans scalping 700 men, women, children, and infants.
In short, fake news is old news.
For all the handwringing over fake news today, BYU journalism professor Joel J. Campbell’s (BA ’87) response is more “meh.” It’s another punch for a profession that’s been in the ring for the better part of a decade. Trust in news media is at an all-time low. Revenue models are upended. Reporters are exhausted. Readers are fragmented. And that’s just a short list of jabs.
Looming larger in Campbell’s eyes are analytics-driven newsrooms and disenfranchised readers, who, flooded with content, are living in information silos or, worse, opting out altogether.
So how does one make sense of the crowded, increasingly polarized news landscape? And what’s left of journalism as we knew it?
The United Arab Emirates is on pace to contribute $20 million over the course of 2016 and 2017 to the Middle East Institute, one of Washington’s leading think tanks, according to a document obtained by The Intercept. The outsized contribution, which the UAE hoped to conceal, would allow the institute, according to the agreement, to “augment its scholar roster with world class experts in order to counter the more egregious misperceptions about the region, inform U.S. government policy makers, and convene regional leaders for discreet dialogue on pressing issues.”
The Emirates, according to the Associated Press, operate a network of torture pens in Yemen where detainees are grilled alive.
MEI was founded in 1946 and has long been an influential player in Washington foreign policy circles. It serves as a platform for many of the U.S.’s most influential figures, allowing them to regularly appear on cable news, author papers, host private briefings and appear on panels in between stints in government.
From 1965–1970, Fieldworkers for the Dictionary of American Regional English (DARE) conducted interviews with nearly 3,000 “Informants” in 1,002 communities across America. They visited native residents in all fifty states and D.C., collecting local words, phrases, and pronunciations. In addition to answering more than 1,600 questions from the DARE Questionnaire, many of the Informants, along with auxiliary speakers, agreed to be recorded by the Fieldworkers. These recordings consisted of conversational interviews as well as readings of “The Story of Arthur the Rat” (devised to elicit the essential differences in pronunciation across the country). This fieldwork data provided invaluable regional information for the Dictionary of American Regional English Volumes I–VI (1985–2013) and Digital DARE.
You’re likely familiar with the term “plagiarism.” We’re taught, early in our education, that plagiarism is copying someone else’s work and claiming it as our own. Sounds simple enough, right? The reality is, there are many ways of being dishonest about the work we produce and giving appropriate credit to whom the original work belongs. The next time you write a paper, make sure you’re keeping your citation standards high. Don’t allow yourself to succumb to any of the thirteen possible ways to plagiarize.
For a similar reference I created in 2015, please see the popular Did I Plagiarize? flowchart.
Winning goes a long way in college football. It packs stadiums, brings in money and can even lead to the glory of a national championship. But at many programs, there’s a qualifier for evaluating that winning: How much did fans have to grit their teeth and pinch their noses on their way to those victories?
This is the awkward harmony of college football. There’s what happens on the field, which grips fans like nothing else on Saturdays. Then there’s what goes on off the field, which may be the only thing capable of overshadowing the football itself.
Now the season is set to kick into full gear this weekend, with the first full slate of games for most teams in the country. Which means it’s time for The Wall Street Journal’s annual Grid of Shame, an exercise that quantifies answers to the two most important questions about your favorite team: How good are they? And how embarrassed should you be about them?
A few months ago, we noticed a notable trend in our web-traffic data: stories with a political aspect were extremely popular with readers. Perhaps this isn’t surprising; today’s news cycle - from the chaos of Brexit to the shambles in the White House, the tragedy of Grenfell to an iceberg twice the size of Luxembourg breaking away from the Antarctic ice shelf – is relentlessly political and possesses an existential urgency.
At one point, it seemed that liberal democracy was cruising towards comfortable middle age. The world order had been established and we were edging in the direction of greater freedoms and equality, some of it driven by increased access to technology. Sometimes progress was dramatic but, more often, it was simply the direction of travel, pulled inexorably in one direction by the tide of history.
Today, whether we are addressing issues of security or the environment, employment law or corporate takeovers of global organisations with vast amounts of data, the WIRED perspective of the world – one that is centred on how technology and science are shaping every aspect of society – is the norm, not an outlier.
Autism and its related conditions remain among the least understood mental health issues of our time. But one significant change that has taken place over the past few years has been a shift from perceiving the autistic mind not as disabled but as differently abled — and often impressive in its difference, as in extraordinary individuals like mathematical mastermind Daniel Tammet or architectural savant Gilles Trehin. And yet despite the stereotype of the autistic mind as a methodical computational machine, much of its magic — the kind most misunderstood — lies in its capacity for creative expression.
Three years after the original publication, New-York-based behavior analyst Jill Mullin returns with an expanded edition of Drawing Autism (public library | IndieBound) — a beautiful and thoughtful celebration of the vibrantly creative underbelly of autism, featuring contributions from more than 50 international graphic artists and children who fall somewhere on the autism spectrum, with a foreword by none other than Temple Grandin.
If one skill is much harder to learn than another, it lowers the “return on investment” of time spent learning that skill. We think STEM skills might be faster to improve than “leadership” skills, which means that they are still contenders for the best skills to learn, especially technology.
We tried to estimate how difficult the skills are to learn with the “time to learn” score. If a skill is useful in jobs that take a long-time to enter, we rated it as “hard to learn”, and vice versa.
We found that many of the most employable skills take the longest to learn, such as judgement, active learning and critical thinking. The STEM skills, however, came out mid-table for “time to learn”.
This makes sense. Programming is a concrete body of knowledge that can clearly be improved. If you go to App Academy, you’ll probably be much better at coding than you were before.
It’s much less obvious how to improve your “judgement” or “social perceptiveness”. We expect that practice and mentorship can help, but they probably partly boil down to personality traits or general mental ability. A meta-analysis of efforts to teach critical thinking skills found that while critical thinking improves during college, it’s unclear that deliberate attempts to train critical thinking have any positive long-term effects, or that some majors improve critical thinking more than others.3
This means that depending on how you weigh the difficulty of learning each skill, STEM skills could still be among the best to learn.
Relatedly, even if you do improve your soft skills, it’s hard to prove to an employer that you possess them. This also reduces the boost you get from learning them.
Similarly, when it comes to “tests and exams” for Watson’s “Data Management” and “Energy Informatics” courses, all will be “open book and open notes” and “designed to assess low level mastery of the course material” (the “Stress Reduction” section has been removed from both syllabi, but an archived version of the “Data Management” syllabus has been provided here).
Finally, for in-class presentations, Watson will allow “only positive comments” to be made, while “comments designed to improve future presentations will be communicated by email.”
This report, the third in the Technology at Work series, focuses on the automation driven by e-Commerce for physical goods. We look at the technology needed to automate order fulfillment, inventory management, and delivery when consumers shop online and examine the implications in a wide range of areas for industry, retailers, supply chains, real-estate, and transportation, looking too at the impact on labor and employment.
Growth in e-Commerce is the main driver of warehouse automation, a driver which itself will increase with broadband and mobile device penetration. In Japan, 32% of all goods bought on the Internet were bought on smartphones in 2016, up from 27% a year earlier. Millennials, those most likely to shop online, will soon enter their peak spending years. Global e-Commerce sales have grown at a compound annual growth rate of 20% over the last decade, and online retail sales have gone from ~2% of total to ~8%%.
While technology is not yet either capable or cost effective in all cases, this is likely to change. Our estimates show that that 80% of jobs in transportation, warehousing, and logistics are susceptible to automation as a consequence of the trends we observe in technology. Retail is one industry in which employment is likely to vanish, but unlike manufacturing jobs which are highly concentrated, the downfall of retail employment will affect every city and region. U.S. companies employ 2 million people just to do stock and order fulfillment work and over 90% of warehouse picking is currently done by hand. Migrating to automated picking gives productivity gains of 2x–3x that as compared to pick-to-conveyor operations and 5x–6x as compared to manual pick-to-pallet fulfillment centers.
Despite the critical importance of free speech on campus, too many universities—in policy and in practice—censor and punish students’ and faculty members’ expressive activity. One way that universities do so is through the use of speech codes: policies prohibiting speech that, outside the bounds of campus, would be protected by the First Amendment.
FIRE surveyed 449 schools for this report and found that 39.6 percent maintain severely restrictive, “red light” speech codes that clearly and substantially prohibit constitutionally protected speech. This is the ninth year in a row that the percentage of schools maintaining such policies has declined, and this year’s drop was nearly ten percentage points. (Last year, 49.3 percent of schools earned a red light rating.)
There are not many agents in the Federal Bureau of Investigation like Ed You. In a workforce that cultivates anonymity, his clean-shaven head gleams. While most of his colleagues are notoriously tight-lipped, Mr You is the chatty star of technology conferences such as South-by-Southwest and DEFCON.
He is also at the forefront of a potential dispute between the US and China, which could have implications for both commercial relations between the world’s two biggest economies and for the future of biomedical research.
The high profile that Mr You has adopted is part of an unusual FBI campaign to highlight the risks in America’s headlong pursuit to unlock the secrets of the human genome. A supervisory special agent in the bureau’s biological countermeasures unit, Mr You warns that the US is not protecting the genomic data used to create lucrative new medicines — but which can also be used to develop fearsome bioweapons.
“We don’t know how much bio data has left our shores,” he says. “Our concept for biological security needs to be broadened.”
The map above shows the percentage of residents in various European countries who are willing to fight and go to war for their country.
Full results below:
From high to low, these are the percentages by country:
“I’m a little embarrassed by the state of this room,” said Will Shortz, The New York Times’s crossword editor, as he waded through a seemingly endless array of puzzle ephemera in his upstairs library. “The problem is, I recently lent part of my collection for use at an exhibition. I got everything back, but I haven’t put it all away yet.”
He paused, looking up from a stack of old magazines.
“Well, it’s more than that, of course,” he said. “Things are just — piling up.”
Indeed they are. Mr. Shortz’s collection includes more than 25,000 puzzle books and magazines, dating to 1534, along with pamphlets, small mechanical puzzles and other ephemeral items. It overwhelms the décor of his home in Pleasantville, N.Y., where he lives and works. A clock in his office is — well, its face is a crossword puzzle. (The hands? Two stubby pencils.) A display case in the living room holds, among other treasures, the first crossword puzzle ever published — in a 1913 Sunday “Fun” section of The New York World. Even the tiled floor in the upstairs bathroom, made of small black and white squares, calls to mind a crossword grid.
At the turn of the 20th century, according to historical estimates, the United States took over from the United Kingdom as the world’s leading economy, a rank it has sustained ever since. Before World War I, in 1913, income per capita in the U.S. was 8 percent higher than in the U.K., 52 percent higher than in all of Western Europe combined and almost 3.5 times the world average income per person.1 By the end of World War II, those gaps were much higher: In 1950, the U.S. income per capita was 38 percent higher than that in the U.K., 112 percent higher than that in Western Europe and more than 4.5 times the world average income per person.2
This global dominance by the U.S. economy can be sustained only by a superior qualification of its workers. This article compares the education of U.S. workers with that of workers in other developed countries and in emerging economies. Although American workers have historically been much better trained than their counterparts abroad, that lead has been quickly disappearing in recent years as other countries have accelerated the skill formation of their workers. Formal skill through education has become increasingly important in a knowledge-based world economy.
Then: U.S. Workers Were No. 1
Figure 1 shows the level of education in 1950 for workers in the U.S. and several other countries that are identified today as developed.3 The levels range from no formal schooling to college completed, which includes workers with education beyond an undergraduate degree. The data for this and the other figures cover males and females of all ages; in all cases, the education levels are the maximum achieved for that group.
Using stylometry one is able to compare texts to determine authorship of a particular work. Throughout the years Satoshi wrote thousands of posts and emails and most of which are publicly available. According to my source, the NSA was able to the use the ‘writer invariant’ method of stylometry to compare Satoshi’s ‘known’ writings with trillions of writing samples from people across the globe. By taking Satoshi’s texts and finding the 50 most common words, the NSA was able to break down his text into 5,000 word chunks and analyse each to find the frequency of those 50 words. This would result in a unique 50-number identifier for each chunk. The NSA then placed each of these numbers into a 50-dimensional space and flatten them into a plane using principal components analysis. The result is a ‘fingerprint’ for anything written by Satoshi that could easily be compared to any other writing.
An unintended effect of Google’s heavy-handed attempt to silence Barry Lynn and his Open Markets program at New America has been to shine a really bright light both on Google’s monopoly power and the unrestrained and unlovely ways they use it. Happily, Lynn’s group has landed on its feet, seemingly with plenty of new funding or maybe even more than it had. I got a press release from them this evening. This seems to be their new site. I’ve already seen other stories of Google bullying come out of the woodwork. Here’s one.
I think it’s great that all this stuff is coming out. But what is more interesting to me than the instances of bullying are the more workaday and seemingly benign mechanisms of Google’s power. If you have extreme power, when things get dicey, you will tend to abuse that power. It’s not surprising. It’s human nature. What’s interesting and important is the nature of the power itself and what undergirds it. Don’t get me wrong. The abuses are very important. But extreme concentrations of power will almost always be abused. The temptations are too great. But what is the nature of the power itself?
Many people who know more than I do can describe different aspects of this story. But how Google affects and dominates the publishing industry is something I know very, very well because I’ve lived with it for more than a decade. To say I’ve “lived with it” makes it sound like a chronic disease or some huge burden. That would be a very incomplete, misleading picture. Google has directly or indirectly driven millions of dollars of revenue to TPM over more than a decade. Not only that, it’s provided services that are core parts of how we run TPM. So Google isn’t some kind of thralldom we’ve lived under. It’s ubiquitous. In many ways, it makes what we do possible.
What I’ve known for some time – but which became even more clear to me in my talk with Barry Lynn on Monday – is that few publishers really want to talk about the depths or mechanics of Google’s role in news publishing. Some of this is secrecy about proprietary information; most of it is that Google could destroy or profoundly damage most publications if it wanted to. So why rock the boat?
I’m not worried about that for a few reasons: 1: We’ve refocused TPM toward much greater reliance on subscriptions. So we’re less vulnerable. 2: Most people who know these mechanics don’t write. I do. 3: We’re small and I don’t think Google cares enough to do anything to TPM. (If your subscription to Prime suddenly doubles in cost, you’ll know I was wrong about this.) What I hope I can capture is that Google is in many ways a great thing for publishers. At least it’s not a purely negative picture. If you’re a Star Trek fan you’ll understand the analogy. It’s a bit like being assimilated by the Borg. You get cool new powers. But having been assimilated, if your implants were ever removed, you’d certainly die. That basically captures our relationship to Google.
It all starts with “DFP”, a flavor of Doubleclick called DoubleClick for Publishers (DFP), one of the early “ad-serving companies” that Google purchased years ago. DFP actually started as GAM – Google Ad Manager. We were chosen to be one of the beta-users. This was I think back in 2006 or 2007. What’s DFP? DFP is the application (or software, or system – you could define it in different ways) that serves ads on TPM. I don’t know the exact market penetration. But it’s the hugely dominant player in ad serving across the web. So on TPM, Google software manages the serving of ads. Our ads all drive on Google’s roads.
Then there’s AdExchange. That’s the part of Google that buys ad inventory. A huge amount of our ads come through ad networks. AdExchange is far and away the largest of those for us – often accounting for around 15% of total revenues every month – sometimes higher. So our largest single source of ad revenue is usually Google. To be clear that’s not Google advertising itself but advertisers purchasing our ad space through Google. But every other ad we ever run runs over Google’s ad serving system too. So Google software/service (DFP) runs the ad ecosystem on TPM. And the main buyer within that ecosystem is another Google service (Adexchange).
Then there’s Google Analytics. That’s the benchmark audience and traffic data service. How many unique visitors do we have? How many page views do we serve each month? What’s the geographical distribution of our audience? That is all collected through Google Analytics. Now, that’s not our only source of audience data. We have several services we use for that in addition to our own internal systems. But we do use it for the big aggregate numbers and longterm record keeping. In many ways it’s the canonical data people on the outside look at to see how big our audience is. Do we have to share that data? No. Unless we want potential advertisers to see we have an audience.
Next there’s search. Heard of that? There’s general search and then there’s Google News, a separate bucket of search. Search tends not to be that important for us in part because we’ve never prioritized it and in part because as a site focused on iterative news coverage what we produce tends to be highly ephemeral – at least in search terms. We don’t publish a lot of evergreen stories. Still, search is important. For other publishers it’s the whole game.
One additional Google implant is Gmail, which we use to provision our corporate email. The backbone of the @talkingpointsmemo.com email addresses is gmail. Lots of companies now do this.
So let’s go down the list: 1) The system for running ads, 2) the top purchaser of ads, 3) the most pervasive audience data service, 4) all search, 5) our email.
But wait, there’s more! Google also owns Chrome, the most used browser for visiting TPM. Chrome is responsible for 41% of our page views. Safari comes in second at 36%. But the Safari number is heavily driven by people using iOS devices. On desktop Chrome is overwhelmingly dominant.
Related: Paying Professors: Inside Google’s Academic Influence Campaign
Company paid $5,000 to $400,000 for research supporting business practices that face regulatory scrutiny; a ‘wish list’ of topics. By Brody Mullins and Jack Nicas m
The biggest financial recession since the Great Depression dealt a particular blow to those who came of age in the US during the crisis. Many older millennials — an age group just entering the labour force for the first time when the crisis hit — suffered early career setbacks that have hindered their ability to afford aspects of the lifestyle that their parents may have enjoyed at the same age.
Compounding a lower employment rate, the cost of US higher education has grown sharply. The total cost of college tuition and fees rose 63 per cent in the decade between 2006 and 2016, according to the Bureau of Labor Statistics. Student debt now exceeds all other forms of consumer debt except mortgages.
Taken together with a home ownership rate that is lower than for previous generations at the same age, the picture that emerges is of a generation falling behind their parents in terms of net worth.
A few months ago, while dining at Veggie Grill (one of the new breed of Chipotle-class fast-casual restaurants), a phrase popped unbidden into my head: premium mediocre. The food, I opined to my wife, was premium mediocre. She instantly got what I meant, though she didn’t quite agree that Veggie Grill qualified. In the weeks that followed, premium mediocre turned into a term of art for us, and we gleefully went around labeling various things with the term, sometimes disagreeing, but mostly agreeing. And it wasn’t just us. When I tried the term on my Facebook wall, and on Twitter, again everybody instantly got the idea, and into the spirit of the labeling game.
As a connoisseur and occasional purveyor of fine premium-mediocre memes, I was intrigued. It’s rare for an ambiguous neologism like this to generate such strong consensus about what it denotes without careful priming and curation by a skilled shitlord. Sure, there were arguments at the margins, and sophisticated (well, premium mediocre) discussions about distinctions between premium mediocrity and related concepts such as middle-class fancy, aristocratic shabby, and that old classic, petit bourgeois, but overall, people got it. Without elaborate explanations.
But since the sine qua non of premium mediocrity is superfluous premium features (like unnecessary over-intellectualized blog posts that use phrases like sine qua non), let me offer an elaborate explanation anyway. It’s a good way to celebrate August, which I officially declare the premium mediocre month, when all the premium mediocre people go on premium mediocre vacations featuring premium mediocre mai tais at premium mediocre resorts paid for in part by various premium-mediocre reward programs.
It is not hard to learn to pattern-match premium mediocre. In my sample of several dozen people I roped into the game, only one had serious trouble getting the idea. Most of the examples below, and all the really good ones, came from others.
“I am juicy, gooey, hot, cheesy and heaven in your mouth. What am I?” the teacher asked.
Hands shot in the air and 10 children bounced up and down in their seats. “Pizza! Pizza! Pizza!” they squealed.
The children, ages 8 to 12, were practicing giving their teacher descriptive words about their favorite food item without saying its name.
It’s just one small piece of a curriculum created by 17-year-old Katie Eder.
Four years ago, Eder’s sister started tutoring kids in math, and she wanted to follow in her big sister’s footsteps. There was one problem — Katie is bad at math.
But the thing she is good at is writing, and Eder couldn’t find anywhere that offered tutoring for children, so she approached Milwaukee’s COA Youth and Family Center to allow her to teach creative writing.
They took a chance on the 13-year-old and agreed — and the result was Kids Tales, a program to empower children, often in low-income areas or in juvenile detention centers, to use creative writing to discover their voice and share their story.
Teenagers, and only teenagers, volunteer to teach children for a week and guide them as they write their own short story, working on brainstorming and plot and character development. Once the stories are completed, they are put into a book, making each child a published author.
Rep. Ben Ray Luján (D-N.M.) last week introduced legislation that would extend $2.5 billion in funding over five years for states combating the opioid crisis, money that would be spent on top of the $1 billion appropriated in 2016 for two years of support.
Luján noted last week in introducing the Opioid and Heroin Abuse Crisis Investment Act that national overdose deaths topped 59,000 in 2016, with deaths from synthetic opioids increasing by 73 percent. Prescription painkillers like OxyContin and Vicodin were responsible for more than 17,500 deaths, according to the announcement.
“The sad fact remains that much more must be done if we are going to ensure that all those who want help can get help,” Luján said in a statement.
The $2.5 billion would be offered through a block grant, with states receiving funding to increase access to treatment, boost prevention programs and “expand evidence-based initiatives that will help address this deadly epidemic.”
Arizona-based general surgeon Jeffrey Singer, a senior fellow at the Cato Institute, said that if lawmakers continue the status quo by “throwing money at addiction treatment centers,” they’re wasting taxpayer dollars.
“It’s not like we have shortage of rehab centers,” Singer said. “People who are addicted to this are addicted because they enjoy it, so having a rehab center available to them isn’t going to make them want to go in and sign up. I just think it’s a waste of money.”
The DOJ — despite issuing its own guidance requiring warrants for Stingrays in 2015 — argued in court earlier this year that no warrant was needed to deploy the Stingray to locate a shooting suspect. It actually recommended the court not reach a conclusion on the Fourth Amendment implications of Stingray use, as it had plenty of warrant exceptions at the ready — mainly the “exigent circumstances” of locating a suspect wanted for a violent crime.
Unfortunately for the federal government (and all other law enforcement agencies located in the court’s jurisdiction), the court declined the DOJ’s offer to look the other way on Constitutional issues. It found a Stingray’s impersonation of cell tower to obtain real-time location information is a search under the Fourth Amendment.
The court adopts Judge Koh’s reasoning in In re Application for Telephone Information, 119 F. Supp. 3d at 1026, to hold that cell phone users have an expectation of privacy in their cell phone location in real time and that society is prepared to recognize that expectation as reasonable. While Judge Koh limited her analysis to the privacy interest in historical CSLI, the court determines that cell phone users have an even stronger privacy interest in real time location information associated with their cell phones, which act as a close proxy to one’s actual physical location because most cell phone users keep their phones on their person or within reach, as the Supreme Court recognized in Riley. In light of the persuasive authority of Lambis, and the reasoning of my learned colleagues on this court recognizing a privacy interest in historical cell site location information, the court holds that Ellis had a reasonable expectation of privacy in his real-time cell phone location, and that use of the Stingray devices to locate his cell phone amounted to a search requiring a warrant, absent an exception to the warrant requirement.
Steady improvements in American life expectancy have stalled, and more Americans are dying at younger ages. But for companies straining under the burden of their pension obligations, the distressing trend could have a grim upside: If people don’t end up living as long as they were projected to just a few years ago, their employers ultimately won’t have to pay them as much in pension and other lifelong retirement benefits.
In 2015, the American death rate—the age-adjusted share of Americans dying—rose slightly for the first time since 1999. And over the last two years, at least 12 large companies, from Verizon to General Motors, have said recent slips in mortality improvement have led them to reduce their estimates for how much they could owe retirees by upward of a combined $9.7 billion, according to a Bloomberg analysis of company filings. “Revised assumptions indicating a shortened longevity,” for instance, led Lockheed Martin to adjust its estimated retirement obligations downward by a total of about $1.6 billion for 2015 and 2016, it said in its most recent annual report.
The Crimson’s survey of more than 50 percent of incoming freshmen in Harvard College’s Class of 2021 asked them about their backgrounds and expectations for life on campus. Read Part I of The Crimson’s three-part series on the freshman survey here.
Lawyers acting for families who claim their children have been illegally excluded from St Olave’s grammar school are considering launching proceedings against a number of other London schools after being contacted by parents.
The news comes as a former St Olave’s governor complained about a lack of transparency in the governance of the school and called on the headteacher and current governing body to “right the wrong” being done to pupils.
The row at St Olave’s, in the London borough of Bromley, over sixth formers being kicked out halfway through their course has prompted a number of inquiries from families who say they are facing similar situations in other selective and non-selective London schools.
Last week I posted an article, which formed the first part in a series on Linear Algebra For Deep Learning. The response to the article was extremely positive, both in terms of feedback, article views and also more broadly on social media.
Many of you commented that there was “an appetite” for introductory mathematical content and this only confirms the results of the QuantStart 2017 Content Survey. Hence I’ve decided to write more introductory articles, not only continuing with Linear Algebra, but also on the topics of Calculus and Probability, which are fundamental topics for machine learning—and quantitative finance more broadly.
In the previous article we introduced the three basic entities that will be used in linear algebra, namely the scalar, vector and the matrix. We saw that they were all really specific versions of a more general entity known as a tensor.
In this article we are going to look at how to form operations between these entities. Such operations include addition and multiplication. While you will be very familiar with scalar addition and multiplication, the rules differ somewhat when dealing with more general tensor entities. This article will precisely define those operations and provide numeric examples to give you some intuition.
At this stage it is not likely to be clear why these operations will be useful in the context of deep learning. However, in the previous article I stated that linear algebra was the ‘language in which machine learning was written’. If we can understand the basics of the language, we’ll be in a much better position to grasp the more complex ideas that form the backbone of neural network models in later articles.
We will begin by looking at matrix addition and then consider matrix multiplication. These operations will eventually allow us to discuss a topic known as matrix inversion, which will form the basis of the next article.
What does it mean to add two matrices together? In this section we will explore such an operation and hopefully see that it is actually quite intuitive.
Matrices can be added to scalars, vectors and other matrices. Each of these operations has a precise definition. These techniques are used frequently in machine learning and deep learning so it is worth familiarising yourself with them.
Given two matrices of size m×n
, it is possible to define the matrix C=[cij]
as the matrix sum C=A+B
That is, C
is constructed by element-wise summing the respective elements of A
. This operation is only defined where the two matrices have equal size, except in the case of broadcasting below. The definition implies that C
also has size m×n
Matrix addition is commutative. This means that it doesn’t matter which way around the matrices are added:
It is also associative. This means that you get the same result if you add two matrices together first, and then another, as if you add another two together first and then the other:
Both of these results follow from the fact that normal scalar addition is itself commutative and associative, because we’re just adding the elements together.
I’m stressing commutivity and associativity for matrix addition because matrix multiplication (defined below) is certainly not commutative. We’ll see why later.
Consider two matrices A
. We can create a new matrix C
It is clear to see that the elements of C
are simply the elements added in the corresponding positions from A
It is possible to add a scalar value x
to a matrix A=[aij]
to produce a new matrix B=[bij]
. This simply means that we’re adding the same scalar value to every element of the matrix. It is written as B=x+A
Scalar-matrix addition is once again commutative and associative, because normal scalar addition is both commutative and associative.
For certain applications in machine learning it is possible to define a shorthand notation known as broadcasting. Consider A∈Rm×n
-dimensional real-valued matrix and x∈Rm
It is possible to define B=A+x
, despite the fact that the matrices A
are unequal in size. The definition of the sum takes bij=aij+xj
That is, the elements of x
are copied into each row of the matrix A
. Since the value of x
is `broadcast’ into each row the process is known as broadcasting.
The rules for matrix addition are relatively simple and intuitive. However when it comes to multiplication of matrices the rules become more complex.
In order to define certain matrix-matrix multiplication operations such as the dot-product (discussed below) it is necessary to define the transpose of a matrix. The transpose of a matrix A=[aij]m×n
of size m×n
is denoted by AT
of size n×m
and is given element-wise by the following expression:
That is, the indices i
are swapped. This has the effect of reflecting the matrix along the line of diagonal elements aii
. The operation is defined for non-square matrices, as well as vectors and scalars (which are simply 1×1
matrices). Note that a scalar is its own transpose: x=xT
. In addition the transpose of the transpose of a matrix is simply itself: ATT=A
Note here that AT
does not represent A
multiplied by itself T
times. This is an entirely different operation. However, it will usually be clear from the context whether we mean the transpose of a matrix or repeated multiplication by itself.
We are now going to consider matrix-matrix multiplication. This is a more complex operation than matrix addition because it does not simply involve multiplying the matrices element-wise. Instead a more complex procedure is utilised, for each element, involving an entire row of one matrix and an entire column of the other.
The operation is only defined for matrices of specific sizes. The first matrix must have as many columns as the second matrix has rows, otherwise the operation is not defined.
The definition below can be a bit tricky to understand initially, so have a look at it first and then try working through the examples to see how specific numeric instances match up to the general formula.
Given a matrix A=[aij]m×n
and a matrix B=[bij]n×p
the matrix product C=AB=[cij]m×p
is defined element-wise by:
That is the elements cij
of the matrix C=AB
are given by summing the products of the elements of the i
-th row of A
with the elements of the j
-th column of B
Note that matrix-matrix multiplication is not commutative. That is:
Given the following two matrices:
It is possible to construct the product AB
of size 2×2
It is also possible to construct the product BA
of size 3×3
The above definition does not initially seem to be motivated in a simple way. Why is matrix-matrix multiplication defined like this? It has to do with a deeper result involving matrices representing linear map functions between two vector spaces. We need not worry about these certain deeper aspects of linear maps as they are beyond the scope of this article series.
However, we can briefly provide some intuition through the idea of composing functions together. It is common in mathematics to compose two functions together to produce a new function. That is the function h
can be defined as h=f∘g
, with the ∘
symbol representing function composition. This notation means that h
is equivalent to g
being carried out first, and then f
If, for example f=sin(x)
. Function composition is not commutative and as such f∘g≠g∘f
. Instead g∘f=sin2(x)
, which is a completely different function. This is why matrix-matrix multiplication is not commutative and also why it is defined as above.
Note also that this definition does not imply that the elements of the matrix C
are defined as the element-wise multiplication of those from A
. That is, the elements in each location cannot simply be multiplied together to get the new product matrix. That is an entirely different operation known as the Hadamard Product and will be discussed below.
Since a column vector is in fact a n×1
matrix it is possible to carry out matrix-vector multiplication. If the left-multiplying matrix has size m×n
then this is a valid operation that will produce another m×1
matrix (column vector) as output.
Matrix-matrix and matrix-vector multiplication are extremely common operation in the physical sciences, computational graphics and machine learning fields. As such highly optimised software libraries such as BLAS and LAPACK have been developed to allow efficient scientific computation–particularly on GPUs.
Scalar-matrix multiplication is simpler than matrix-matrix multiplication. Given a matrix A=[aij]m×n
and a scalar λ∈R
the scalar-matrix product λA
is calculated by multiplying every element of A
such that λA=[λaij]m×n
If we take two real-valued scalars λ,μ∈R
the subsequent useful relationships follow from the definition above:
The first result states that you will get the same answer if you add two matrices together, and then multiply them by a scalar, as if you individually multiplied each matrix by the scalar and then added them together.
The second result states that if you add two scalars together and then multiply the result by a matrix it gives the same answer as if you individually multiplied the matrix separately by each scalar and added the result.
The third result states that the order of scalar multiplication does not matter. If you multiply the matrix by one scalar, and then multiply the result by another scalar it gives the same answer as if you first multiplied both scalars together and then by the matrix.
All of these results follow from the simple rules of scalar multiplication and addition.
It is possible to define element-wise multiplication of matrices, which differs from the definition of matrix-matrix multiplication above. The Hadamard product of two matrices A=[aij]m×n
is denoted by A⊙B
and calculated by the following expression:
That is, the elements of the new matrix are simply given as the scalar multiples of each of the elements from the other matrices. Note that since scalar multiplication is commutative so is the Hadamard Product, unlike normal matrix-matrix multiplication.
When might it be necessary to use the Hadamard product? In fact such a process has wide applications, including correcting codes in satellite transmissions and cryptography, signal processing as well as lossy compression algorithms for images in JPEG format.
An important special case of matrix-matrix multiplication occurs between two vectors and is known as the dot product. It has a deep relationship with geometry and is useful in all areas of the physical and computational sciences.
We have to be extremely careful here in regards to notation. I am being particularly precise about this definition, which may be unusual to some of you who have it seen it written before. In particular the dot product is really only defined as a true matrix-matrix multiplication, where one of the vectors is a row “matrix” and the other a column “matrix”. However, you will often see a slight “notational abuse” where it is defined for any two vectors (row or column).
The usual notation for a dot product between two n
-dimensional vectors x,y∈Rn
, which is where the name of the operation comes from. However in more advanced textbooks (particularly the popular graduate level statistics, machine learning and deep learning texts) you will see it written as xTy
, where T
represents the transpose of x
This is because x
are usually considered to both be column vectors. A matrix-matrix multiplication between two column vectors is not defined. Hence, one of the vectors needs to be transposed into a row vector such that the matrix-matrix multiplication is properly defined. Hence you will see the xTy
notation frequently in more advanced textbooks and papers. Now we can proceed with the definition!
Given two column vectors x,y∈Rn
it is possible to define the dot product (or scalar product) between them by taking the transpose of one to form a product that is defined within the rules of matrix-matrix multiplication. Such a product produces a scalar value and is commutative:
The dot product has an important geometric meaning. It is the product of the Euclidean lengths of the two vectors and the cosine of the angle between them. In subsequent articles the concept of norms will be introduced, at which point this definition will be formalised.
Another way to think about a dot product is that if we take the dot product of a vector with itself it is the square of the length of the vector:
Hence to find the (Euclidean) length of a vector you can simply take the square root of the dot product of the vector, xTx−−−−√
The dot product is a special case of a more general mathematical entity known as an inner product. In more abstract vector spaces the inner product allows intuitive concepts such as length and angle of a vector to be made rigourous. However such spaces are beyond the scope of this article series and will not be discussed further.
This article has all been about operations applied to one or more matrices. We can now add and multiply matrices together. But what does this give us? How can we use it?
In the next article we are going to look at one of the most fundamental topics in linear algebra—inverting a matrix. Matrix inversion allows us to solve matrix equations, in exactly the same way that scalar algebra allows us to solve scalar equations.
This is a widespread operation in the physical and computational sciences and will be indispensible in our studies of deep learning.
Scalars, Vectors, Matrices and Tensors – Linear Algebra for Deep Learning (Part 1)
Matrix Algebra – Linear Algebra for Deep Learning (Part 2)
 Blyth, T.S. and Robertson, E.F. (2002) Basic Linear Algebra, 2nd Ed., Springer
 Strang, G. (2016) Introduction to Linear Algebra, 5th Ed., Wellesley-Cambridge Press
 Goodfellow, I.J., Bengio, Y., Courville, A. (2016) Deep Learning, MIT Press
In the previous decade, there has been a considerable rise in the usage of smartphones.Due to exorbitant advancement in technology, computational speed and quality of image capturing has increased considerably. With an increase in the need for remote fingerprint verification, smartphones can be used as a powerful alternative for fingerprint authentication instead of conventional optical sensors. In this research, wepropose a technique to capture finger-images from the smartphones and pre-process them in such a way that it can be easily matched with the optical sensor images.Effective finger-image capturing, image enhancement, fingerprint pattern extraction, core point detection and image alignment techniques have been discussed. The proposed approach has been validated on FVC 2004 DB1 & DB2 dataset and the results show the efficacy of the methodology proposed. The method can be deployed for real-time commercial usage.
It was clear from the start — even to my partial eye — that many of the 1,000 applicants were going to be catastrophes in the classroom. One chief executive of a consultancy firm applied, claiming that he had a strong urge to teach. The following day he sent an email withdrawing his application. He had told his wife over supper what he was planning to do. She pointed out the flaw in his scheme: he didn’t like children very much — not even his own.
It was clear from the start — even to my partial eye — that many of the 1,000 applicants were going to be catastrophes in the classroom. One chief executive of a consultancy firm applied, claiming that he had a strong urge to teach. The following day he sent an email withdrawing his application. He had told his wife over supper what he was planning to do. She pointed out the flaw in his scheme: he didn’t like children very much — not even his own.
Many of the applicants had not set foot in a school since they attended one themselves 30 or 40 years earlier, and so were sent off for a week’s immersion. This weeded out all those who had a fond vision of themselves as Robin Williams in Dead Poets Society. It also got rid of those unsuited to the rigidity of school life. One man was told to leave after his first day — he had sat at the back of class checking his emails and then proceeded to go to sleep.
But for many others, time in school had the reverse effect. Richard Lewis, a 64-year-old consultant, emailed exultantly: “This is the best fun I have had since I bought my new motorbike . . . and I’ve only been here for four lessons. I want to do this all the time!”
Those who survived the week were put through the same assessment as any 22-year-old entering the profession. I sat in on some of the early interviews, wincing as former corporate titans failed to jump through hoops set out for them by people three decades their junior. A senior partner of a magic circle law firm was asked to think of a time when he had received negative feedback and explain how it had made him feel. This floored him. “Gosh”, he replied. “That’s a hard one. I haven’t received any feedback at all in living memory. It’s me who gives it to others . . . ” He didn’t make it. Lots of others didn’t make it, either — they came over as too arrogant, too inflexible or entirely out of touch.
Kind of. In a fascinating new paper published this summer, five economists, Raj Chetty, John Friedman, Emmanuel Saez, Nicholas Turner, and Danny Yagan, call into question higher education’s role in promoting upward mobility. The centerpiece of the paper is “mobility report cards” for each college in America. The researchers considered 30 million students between 1999 and 2014 and compared their parents’ incomes to their own post-college earnings, by school. With this data, they could see exactly which colleges helped the most students rise from the bottom of the earnings ladder to the top.
If the U.S. Constitution can’t protect free speech on campus, what can?
Campus speech codes are hard to understand and hard to follow for the neurodivergent, as I argued in my previous article. In principle, this problem could be reduced by rewriting speech codes to be more concrete and detailed, with complete lists of prohibited words, forbidden ideas, banned images, and unwelcome mating tactics. The neurodivergent could simply memorize these lists and feel a little more confident that they understand what they are not allowed to say or do. But no public university would dare to print such lists of communication taboos, since the First Amendment violations would be all too conspicuous, and the lawyers from the Foundation for Individual Rights in Education (FIRE) would sue for prior restraint.
Sperm count in men from North America, Europe, Australia and New Zealand declined by 50-60% between 1973 and 2011, according to a new study from the Hebrew University of Jerusalem. Surprisingly, the study, which analysed data on the sperm counts of 42,935 men, found no decline in sperm counts in men from Asia, Africa and South America, although there was limited data from these areas.
Overall, this is a very disturbing report. There has been a longstanding debate among scientists as to whether sperm counts have decreased or not. But what’s different about this study is the quality of the analysis. It was done in a systematic manner, accounting for several of the problems that had affected previous studies, such as the method used to count sperm and comparing studies performed sometimes decades apart. As such, most experts agree that the data presented is of a high quality and that the conclusions, although alarming, are reliable.
So what is going on? There has been concern for a number of years about an increase in abnormalities in male reproductive health, such as testicular cancer. The decline in sperm counts is consistent with these increases and this adds weight to the concept that male reproductive health is under attack and is declining rapidly.
In a few weeks, a new intake of students will arrive, all fresh-faced and excited, at universities around the country. They’ll be thrilled at the prospect of escaping the wagging finger of mum and dad, eager to absorb new ideas. But I’m afraid they are in for a rude awakening. Unless they’re very fortunate, they will soon find themselves enveloped in a world that’s more censorious than stimulating and taught not to question ideas but to learn by heart the progressive creed. It will take a brave and resilient youngster to survive university with their intellectual curiosity intact.
Every aspect of campus life, from what you can say to how you should party, is minutely policed by what I called the Stepford Students in this magazine three years ago. ‘No Platform’ policies strictly govern who can speak on campus. Anybody, no matter what their political background or supposedly liberal credentials, can find themselves shunted off campus for having the wrong opinion in the eyes of the Stasi of student politics.
There are, broadly speaking, two kinds of futurology, the utopian and the apocalyptic. In Homo Deus, Yuval Noah Harari, like the Book of Revelation, offers a bit of both. And why not? The function of imaginary futures is to deliver us from banality. The present, like the past, may be a disappointing muddle, but the future had better be very good or very bad, or it won’t sell.
Harari, an Oxford-educated Israeli historian who teaches in Jerusalem, is the author of Sapiens (2015), a provocative, panoramic view of human evolution and history upward from apedom. It became an international bestseller, recommended by the likes of Mark Zuckerberg, Bill Gates, and Barack Obama. Harari’s style is breezy and accessible, sprinkled with allusions to pop culture and everyday life, but his perspective is coolly detached and almost Machiavellian in its unflinching realism about power, the role of elites, and the absence of justice in history. He is an unapologetic oracle of Darwin and data. And he is clearly a religious skeptic, but he practices a form of Buddhist meditation, and among the best things in his new book, like his previous one, are his observations on the varieties of religious experience.
Harari begins by assuring us that humanity is on a winning streak. Famine and plague, two historical scourges, are disappearing, and a third, war, is no longer routine statecraft. For the first time in history, more people die of eating too much than eating too little. More people succumb to ailments related to old age than to infectious diseases. Victims of all kinds of violence are, as percentages of the population, at historical lows in most places. The next stop, presumably, is Utopia.
For years, Misti Boackle had watched several cities break away from the Jefferson County school district that includes this Birmingham suburb, each forming what she considers superior school systems. So in 2012, she joined other residents to do the same for Gardendale.
“I felt it was the best thing for our family and our community,” said Ms. Boackle, a white mother of three children, two of whom attend Gardendale High School.
This spring, after years of battles, a court granted Gardendale a roadmap to having its own schools, though legal appeals have delayed it. Like many of the other area municipalities that have created separate school districts, Gardendale is a mostly white city while the county district is predominantly black.
The effort is one of a growing number of attempted school-district secessions—in states including California, Georgia and Wisconsin—highlighting deep divisions nationwide over race, class, and the role of desegregation orders six decades after the U.S. Supreme Court’s Brown v. Board of Education ruling. That decision declared racially separated schools unconstitutional.
Supporters of the moves say they are a way for communities to exert control over education policy and retain property-tax revenue for local benefit. Opponents say the separations can resegregate schools and exacerbate income disparities by breaking off wealthier, whiter areas.
In the hours after European antitrust regulators levied a record $2.7 billion fine against Google in late June, an influential Washington think tank learned what can happen when a tech giant that shapes public policy debates with its enormous wealth is criticized.
The New America Foundation has received more than $21 million from Google; its parent company’s executive chairman, Eric Schmidt; and his family’s foundation since the think tank’s founding in 1999. That money helped to establish New America as an elite voice in policy debates on the American left.
But not long after one of New America’s scholars posted a statement on the think tank’s website praising the European Union’s penalty against Google, Mr. Schmidt, who had chaired New America until 2016, communicated his displeasure with the statement to the group’s president, Anne-Marie Slaughter, according to the scholar.
The statement disappeared from New America’s website, only to be reposted without explanation a few hours later. But word of Mr. Schmidt’s displeasure rippled through New America, which employs more than 200 people, including dozens of researchers, writers and scholars, most of whom work in sleek Washington offices where the main conference room is called the “Eric Schmidt Ideas Lab.” The episode left some people concerned that Google intended to discontinue funding, while others worried whether the think tank could truly be independent if it had to worry about offending its donors.
Telecom companies, too, have their own favorite think tanks. And for years Google’s regulatory agenda seemed aligned with the public interest in a relatively uncomplicated way. As a dominant web search and web advertising company, they benefitted financially from lots of people having affordable high-quality internet access, and thus supported approaches to spectrum and wireline internet regulation that were likely to produce a consumer-friendly outcome.
But as Google itself grew, it came to be the target of possible regulatory action rather than simply the beneficiary of a competitive market in allied industries.
The trend is “unique in American history,” said Richard Vedder, director of the Center for College Affordability and Productivity and an economics professor at Ohio University.
“Even in the Great Depression, enrollments went up,” Vedder said. “There’s an increasing skepticism on the part of the public that college produces the bang for the buck that it claims to.”
Attorney General Maura Healey, who opened a unit in her office dedicated to assisting debt-addled students, said college’s traditional bargain is in question.
“It’s not the case anymore that a four-year, liberal arts education is going to be the ticket to economic mobility in today’s economy,” Healey said. “As I talk to employers in this state, I know there are certain jobs that are open, but they’re looking for a certain skill set that I don’t think we have done as good a job filling.”
Graduates with bachelor’s degrees still earn appreciably more than high school graduates, a median weekly pay of $1,156 compared to $692 for high school grads, according to the U.S. Bureau of Labor Statistics.
About 10 years ago, Tim Wu, the Columbia Law professor who coined the term network neutrality, made this prescient comment: “To love Google, you have to be a little bit of a monarchist, you have to have faith in the way people traditionally felt about the king.”
Wu was right. And now, Google has established a pattern of lobbying and threatening to acquire power. It has reached a dangerous point common to many monarchs: The moment where it no longer wants to allow dissent.
This summer, a small team of well-respected researchers and journalists, the Open Markets team at the New America think tank (where I have been a fellow since 2014), dared to speak up about Google, in the mildest way. When the European Union fined Google for preferring its own subsidiary companies to its rival companies in search results, it was natural that Open Markets, a group dedicated to studying and exposing distortions in markets, including monopoly power, would comment. The researchers put out a 150-word statement praising the E.U.’s actions. They wrote, “By requiring that Google give equal treatment to rival services instead of privileging its own, [the E.U.] is protecting the free flow of information and commerce upon which all democracies depend.” They called upon the Federal Trade Commission and Department of Justice and state attorneys general to apply the traditional American monopoly law, which would require separate ownership of products and services and the networks that sell products and services.