Zero 88

18 Dec 18
OILERSNATION
My mother is one of the most well-read people I’ve ever met. She has read over 10,000 books and earlier this year, while renovating her living room, she installed a bookcase across an entire wall. It is filled with books of various genres. It’s almost like having a library in your own home. (She actually just finished writing her first novel — more on that later.) She loves reading and learning, and while she is very much an intellect, I will always remember the advice she gave to my brother and me when we were little. In grade one, some kid kept bullying my brother at school and stealing his toys in class. “Next time he does it, you look him right in the eye and say, ‘please stop,'” mom said to my brother. “If he doesn’t, then you ask him again to stop. If he continues then you tell him, ‘If you do it one more time I will have no choice but to punch you.'” Then she added, “And if you decide to punch him, make sure you hit him first and hit him hard enough he won’t get up.” I’m sure some oversensitive parent today thinks this is bad parenting, but at some point you need to stand up for yourself, especially after two warnings, and take matters into your own hands, rather than tattle to the teacher. My brother ended up socking the kid one day, and after that he never bullied him again, so the lesson worked. I think the lesson also applies to sports. It is always better to gain an advantage early or, shall we say, land the first blow. Scoring first in the NHL is still a major advantage. The Edmonton Oilers are 12-5-2 when they score first and 6-8-1 when they give up the first goal. Scoring first doesn’t guarantee you victory, but it gives you an early advantage and often teams can carry that momentum through the game. The St.Louis Blues are struggling. They are 3-12-1 when the opposition scores first and 9-3-3 when they score first, but lately netminder Jake Allen has really struggled out of the gate. In the past four games he has allowed ten first period goals on 40 shots. The Calgary Flames scored four goals on 16 shots on Saturday and Allen didn’t return for the second period. Three games earlier the Vancouver Canucks pumped three goals past him, on only six shots, in the first 14 minutes of the game and he was done for the night. The Blues are fragile and if the Oilers can get to Allen early, you know he and his teammates will be thinking, “Here we go again.” The mental aspect of the NHL plays a major role for every player and for collective teams. Score the first goal, and a team can get an early boost, and if it is one of your best offensive players then they often have more pep in their step the remainder of the game. It can also work the other way, especially for losing teams. They give up an early goal, and doubt creeps into their mind a lot quicker than it would for a winning team. Players try to block it out, and good teams do, but it is not a surprise that since Connor McDavid arrived in the NHL (2015/2016 season), no NHL team has a winning record when they surrender the first goal. Only three teams have a winning% over .400 when allowing the first goal during the past 280 games. Washington is 60-56-16, Pittsburgh went 56-58-13 and Tampa Bay is 61-65-17. In those 280 games the Oilers are 44-87-15 when the opposition scores first, but they are 88-35-11 when they score first. I recall someone trying to suggest the second goal was equally important, but that doesn’t jive for me. When you score first you get the early advantage, and it allows you to go up 2-0 and gain a bigger advantage, but it also allows you to surrender a goal and be right back to even. If you don’t score first you have zero chance of going ahead 2-0. The Blues have been on their heels all season. They’ve never won more than two games in a row and they’ve only done that three times. They are 4-6-2 on the road and have allowed four or more goals in half of their road games. In their four victories they’ve only allowed one first period goal, and in their eight losses they’ve allowed eight first period goals. The Oilers need to land the first goal (punch) tonight against a Blues team that has proven to have a weak chin so far this season. *A quick moment for a son to brag. My mother’s first novel, I, the Woman, Planted the Tree, was released earlier this month and is now available on Amazon.  It will resonate with those interested in a deeper learning of who you are and what your dreams are trying to tell you.  “An immersion experience for seekers, healers and dreamers, this book is a journey into the dark feminine. This is a real, gut-wrenching and timeless story of woman’s search for the Divine Feminine. A surprising story of the desperation and final release from seemingly endless depression, this book is for those who have found no relief either in talk therapy, the medical establishment, pharmaceuticals, or conventional religious and cultural institutions.” Recommend it to a friend if you have a moment.* LINEUP… Oilers Chiasson-McDavid-Draisaitl Rieder-Nuge-Rattie Lucic-Brodziak-Kassian Caggiula-Khaira-Puljujarvi Nurse-Larsson Gravel-Jones Garrison-Benning Talbot Hitchcock juggled his second and fourth line, and he did so in hopes of them getting more icetime. “Right now we are relying on too few players, and at the end of the game, especially an intense playoff-style game like in Vancouver, we had some people who were worn down and had played too many minutes and it showed in their play. We got to get more guys helping us in a better way, and he (Khaira) is one of them,” explained Hitchcock on the reshuffling. He also mentioned he feels Khaira and Puljujarvi have potential to give more. He didn’t say it in the way they were playing poorly, just that he feels that line along with Drake Caggiula should be able to play productive minutes. Talbot gets the start. He defeated the Blues 3-2 in a SO on December 5th, but he is only 3-4-1 on home ice. Hitchcock wants to keep both his goalies involved, and even though Koskinen has assumed the starters role, with ten starts to Talbot’s five (including tonight), he doesn’t want Talbot sitting too long between starts. If Talbot didn’t play tonight he likely wouldn’t have played until after Christmas. Because the Oilers will play Koskinen v. the league-leading Tampa Bay Lightning on Saturday. The Oilers PK had been trending in the right direction prior to Sunday’s loss in Vancouver. Two of Vancouver’s three goals came from losing faceoffs. Ironically, the same Oilers centres have been better on PK draws than they have on the PP or at EV. Nugent-Hopkins is 45.6% on the PK (26-31) and 42.5% at EV (132-175). Draisaitl is 48.7% on the PK (19-20), but only 45% on the PP (58-71). He is a solid 51.6% at EV (208-195). Kyle Brodziak is also 48.7% on the PK (19-20), and is their best EV faceoff guy at 51.9% at EV (160-148). The PK needs a bounce-back effort tonight. The Blues have the fifth best road PP in the NHL at 24.3%. It is hard for any PK to have success without solid goaltending. Talbot needs to make a few more saves on the PK. He has allowed 15 goals on 61 shots (.754sv%), while Koskinen has allowed ten goals on 81 shots (.877sv%). McDavid needs one assist to reach 200 in his career and if he does it tonight he'll be the 8th fastest player to reach 200 assists.W.Gretzky-165 gamesP. Stastny-199 M. Lemieux-213D. Savard-214 P. Forsberg-214S.Crosby- 215J.Juneau- 236Tonight is McDavid's 243rd game. #NHL — Jason Gregor (@JasonGregor) December 18, 2018 Blues Steen-O’Reilly-Tarasenko Schwartz-Schenn-Perron Maroon-Bozak-Thomas Nolan-Barbashev-Sundqvist Edmundson-Parayko Bouwmeester-Bortozzo Dunn-Butler Allen The Blues forwards are healthy with Steen and Schwartz returning. When the Oilers faced them two weeks ago, Sanford and Thomas started the game on the top line with Ryan O’Reilly. The Blues have some skilled players in their top six, and the Oilers can’t play too lose, but right now the Blues biggest weakness is in their defensive zone. TONIGHT… Photoshop: Tom Kostiuk GAME DAY PREDICTION: The Oil have won six straight at home. They won their final eight home games in 2016/2017, and they inch closer to tying that with a 4-3 victory. OBVIOUS GAME DAY PREDICTION: McDavid picks up his 200th career assist. He finishes with two points and now has 11 points in 11 career games v. the Blues. NOT-SO-OBVIOUS GAME DAY PREDICTION: Tobias Rieder scores his first goal as a member of the Oilers exactly seven years and 156 days after they drafted him 114th overall. MONTH OF GIVING Thank you to Al Prokop and the groups at Blackhawk Golf Club and restaurant XIX for the great packages and to Ryan and Scott for their great bids. DAY 12: Package #2 REALTOR FEES from Michelle Derk Up for grabs is the seller’s commission for your house. (In a regular sale half the commission goes to the seller realtor and half to the buyer. So this is $3.5% on the first $100,000 and $1.5% on the balance). On a $500,000 house that is a value of $9,500.00. On a $600,000 house that is a value of $11,000.00. and so on. PACKAGE #2: Family and Rivalry games Four seats at centre ice (section 120) row 22 for Oilers/Canucks on Thursday, December 27th. Unreal seats Four Loge seats to the Edmonton Oil Kings v. Calgary Hitmen on Friday, December 28th. You can bid by listening to TSN 1260 and calling 780.444.1260 or text 101260 between 2-6 p.m. today. Proceeds will help The Christmas Bureau and Gregor’s Grads. GDB BROUGHT TO YOU BY COLLIN BRUCE MORTGAGE TEAM Find out why so many people are using Dominion Lending Centres #1 broker in Canada. We are paid by the banks on the size of the mortgage, not the interest rate, so we try to get you as low as rate as we can. Whether it is a purchase, renewal or refinance, contact us to see what we can do to help! Recently by Jason Gregor: Pacific Division Surge Can Alex Chiasson be the next Alex Burrows? GDB 33.0: Another KoskiWin? McDavid’s 300th point puts him in rare group GDB 32.0: Goin to Winnipeg Paging Nurse to the Top Unit Source: NHL, Official Game Page, 12/18/2018 – 1:00 pm MT
18 Dec 18
The X Report

Raven X January 16, 2018 AFC Championship: Patriots vs. Jaguars Who in the world can honestly say that this was their projected AFC Championship before the season started? Thanks to a great offseason and improvements to their seasoned veterans, the Jacksonville Jaguars are in their first AFC Championship game since 1999 and are hoping to […]

18 Dec 18
STN Media

LISLE, Ill. — Navistar International Corporation (NAV) today announced fourth quarter 2018 net income of $188 million, or $1.89 per diluted share, compared to fourth quarter 2017 net income of $135 million, or $1.36 per diluted share. For fiscal year 2018, Navistar reported net income of $340 million, or $3.41 per diluted share vs. net […]

18 Dec 18
Golfweek

[jwplayer GyZP302V] The 1960s were a tough time for most golfers to win a PGA Tour event, not to mention a major championship. That’s because four players combined for 116 PGA Tour victories in the decade (keep clicking to see their records), including 18 of the possible 40 majors. It’s even tougher when choosing the […]

18 Dec 18

The Seoul Plaza Ice Rink, an iconic winter attraction in the city, will open Friday with upgraded facilities. The ice rink will operate through Feb. 10 for a total of 52 days, according to the Seoul Metropolitan Government on Tuesday. The rink’s design has been changed for the first time since it opened four years […]

18 Dec 18
Eurasia Future

A year of destiny In 1978, Chinese paramount leader Deng Xiaoping visited Singapore, one of Asia’s most successful countries then as it is now. Deng met Prime Minister Lee Kuan Yew and held deeply frank discussions about how China could unleash its potential by embracing a new era of economic openness that would be to […]

18 Dec 18
Maia's Mind

The human population is rapidly increasing, with predictions suggesting 9.8 billion people by 2050, demanding a 75% increase in food production (Morris 2018). Reconciling food production with biodiversity is an ongoing challenge as expansion of agricultural systems causes habitat loss (Phalan et al 2016), responsible for threatening 54% of threatened species (Koch et al 1998). […]

18 Dec 18
iabdb

This post is inspired by a happy coincidence, I was doing p2 of day 14 of advent of code for 2017 and I had never formerly encountered the flood fill algorithm. This confluence of events inspired me to come up with a new way to solve this classic problem.  You are given a boolean matrix and […]

18 Dec 18
THE INDIAN DEF BLOG

  REAL HISTORY OF THE DEAL IS NOT FROM 2001 BUT FROM MUCH BEFORE The current Rafale mess is a result of long vision less and rudderless control of aerospace and defense industry of this country and it’s leadership. This mess has it’s roots in long 4 decades tenure from 1960s to 2000. Will go […]

18 Dec 18
TNT

Cryptocurrency Weekly Update Weekly CommentWith the festive break rapidly approaching, those looking for any crumb of comfort from the crypto market or indeed broader macroeconomic news should probably just surrender to over eating, family arguments and endless re-runs early. Along with the ongoing Brexit saga and Donald J. Trump show, crypto markets continue to leach […]

17 Dec 18
Us Weekly
Here’s the problem with 50 percent off sales: We never actually end up saving any money! Instead, we use this opportunity to buy twice as many things as we ordinarily would. While our lack of self-control isn’t exactly great for our bank accounts, it’s certainly great for our closets and luckily, one of our favorite retailers is having a sale we simply can’t afford to miss! Express is one of our top places for picking up both festive looks to wear to parties, dinner dates and beyond, plus the store carries plenty of wear-everywhere options to wear, well, whenever. From chunky knit sweaters made for layering to great leather-look leggings and holiday party-appropriate dresses, we’ve scored finds for just about every style and occasion. Thankfully, the store is having a massive sale where everything — yes, everything — is 50 percent off. Get your cards, carts and closets ready! Not sure what to pick up from the massive sale? Keep scrolling for our picks! A pair of camel-hued boots that can be dressed up or down Who said only black boots can be versatile? This pretty tan pair boasts a slouchy silhouette that’s perfect for pairing with skirts and is roomy enough for tucking leggings and skinny jeans into. The toe isn’t too pointed, so we won’t feel like we’re squishing our feet in these, plus the heel isn’t so high as to make them uncomfortable to walk in. They look way pricier than they are — despite giving us serious Isabel Marant vibes, they only cost a fraction of the price. Plus, reviewers dig how easy these are to wear and how they team well with both denim and dresses. Still unsure? They’re so popular, over 95 percent of reviewers recommended them! See it: Check out the Slouch Ankle Booties for 50 percent off the original price of $70, now only $35 at Express. Not feeling it? Take a look at other on-sale boots at Express!   A pair of stretchy jeans that are as comfortable as leggings Unpopular opinion: Mid-rise jeans are a way more flattering fit than high-rise styles. They cut off at a more comfortable place on your stomach and still provide plenty of control while not cutting off your circulation in the process. This pair from Express boasts so much stretch they feel more like leggings than jeans, but still have plenty of hold and won’t bag out even after multiple washes and wears. Plus, more than 600 reviewers agree that these are the most comfortable pants they own. If that’s not high praise, we don’t know what is! See it: Snag these Mid-Rise Black Stretch Jean Leggings for 50 percent off the $80 price tag, now only $40 only at Express. Not into these? Be sure to browse other on-sale jeans at Express!   A bodysuit that looks way pricier than it is Bodysuits don’t have to be basic layering tools — they can steal the show all on their own! This Express One Eleven Twist Neck Thong Bodysuit is certainly an outfit-maker. The three-quarter sleeves create a slimming effect on the arms, while the twist detailing at the shoulder gives the design a much more expensive vibe. We can’t wait to pair it with satin skirts and tucked into skinny jeans, though we bet it would look just as chic when worn with more relaxed-fit slacks. See it: Check out the One Eleven Twist Neck Thong Bodysuit for 50 percent off the original price tag of $50, now only $25 at Express. Looking for something else? Check out other bodysuits at Express!   A jumpsuit that’s made for holiday parties, rooftops and date nights We can get tired of relying on dresses for virtually any special occasion so when we spot a jumpsuit that’s just as versatile and easy to style with anything in our closet, we take notice. This particular one from Express has a sweetheart neckline that’s molded by wires, ensuring a custom fit every time (and making sure there’s zero unsightly gaping). We can’t wait to pair this style with sparkly heels and a belt to dress up the simple silhouette. For date night, we plan on pairing it with stilettos and a glimmering necklace, along with a red lip for even more drama. See it: Scoop up the V-Wire Sweetheart Jumpsuit for 50 percent off the original price of $88, now only $44 at Express. Not feeling the design? Check out other jumpsuits at Express!   A stunning coat that makes everything we wear it with so much chicer A truly magnificent coat has a way of making everything we wear with it appear much chicer. This coat from Express does just that — the camel tone makes it both versatile and irresistibly stylish, though the dramatic length is what truly took our breath away. See it: Check out the Open Long Coat marked down 50 percent from $248 to $124, only at Express! Not loving it as much as we are? Check out other coats from Express! Check out more of our picks and deals here! This post is brought to you by Us Weekly’s Shop With Us team. The Shop With Us team aims to highlight products and services our readers might find interesting and useful. Product and service selection, however, is in no way intended to constitute an endorsement by either Us Weekly or of any celebrity mentioned in the post. The Shop With Us team may receive products free of charge from manufacturers to test. In addition, Us Weekly receives compensation from the manufacturer of the products we write about when you click on a link and then purchase the product featured in an article. This does not drive our decision as to whether or not a product or service is featured or recommended. Shop With Us operates independently from advertising sales team. We welcome your feedback at ShopWithUs@usmagazine.com. Happy shopping!
17 Dec 18
Derivative Dribble

In a previous post entitled, “Using Information Theory to Create Categories with No Prior Information”, I presented an algorithm that can quickly construct categories given a linear dataset with no other prior information. In this post, I’ll present a generalization of that algorithm that can be applied to a dataset of n-dimensional vectors, again with […]

17 Dec 18
Asian Miner Shop

Digital asset prices are seeing modest improvements on Monday, Dec. 17 as the top 10 cryptocurrencies have recorded gains between 5-19% over the last 24 hours. The entire crypto-economy has captured a market valuation of $108 billion but global trade volume has remained flat, showing no immediate signs of a market trend reversal. Also read: A […]

17 Dec 18
Qu5.Wiki

Digital asset prices are seeing modest improvements on Monday, Dec. 17 as the top 10 cryptocurrencies have recorded gains between 5-19% over the last 24 hours. The entire crypto-economy has captured a market valuation of $108 billion but global trade volume has remained flat, showing no immediate signs of a market trend reversal. Also read: A […]

17 Dec 18
Pew Research Center’s Social & Demographic Trends Project
To analyze image search results for various occupations, researchers completed a four-step process. First, they created a list of U.S. occupations based on Bureau of Labor Statistics (BLS) data. Second, they translated these occupation search terms into different languages. Third, the team collected data for both the U.S. and international analysis from Google Image Search and manually verified whether or not the image results were relevant to the occupations being analyzed. Finally, researchers deployed a machine vision algorithm to detect faces within photographs, and then estimate whether those faces belong to men or women. The aggregated results of those predictions are the primary data source for this report. Constructing the occupation list Because researchers wanted to compare the gender breakdown in image results to real-world gender splits in occupations, the team’s primary goal was to match the terms used in Google Image searches with the titles in BLS as closely as possible. But the technical language of the BLS occupations sometimes led to questionable search results. For example, searches for “eligibility interviewers, government programs” returned images from a small number of specialized websites that actually used that specific phrase, biasing results toward those websites’ images. So, the research team decided to filter out highly technical terms, using Google Trends to assess relative search popularity, relative to a reference occupation (“childcare worker”). The query selection process for the U.S. analysis involved the following steps: Start with the list of BLS job titles in 2017. Exclude occupations that do not have information about the fraction of women employed. For example, “credit analysts” did not have information about the fraction of women in that occupation. Filter out occupations that do not have at least 100,000 workers in the U.S. Remove all occupations with ambiguous job functions (“all other,” “Misc.”). Split all titles with composite job functions into individual job titles (For example, “models and demonstrators” to “models,” “demonstrators”). Change plural words to singular (“models” to “model”) to standardize across occupations.[7. numoffset=”7″ For one search (“bellhops”), researchers inadvertently used the plural form of the word for the U.S. analysis.] Manually inspect the list to ensure that the occupations were comprehensible and likely to describe human workers. This involved removing terms that might not apply to humans (such as tester, sorter) based on the researchers’ review of Google results. Use Google Trends to remove unpopular or highly technical job titles. Highly technical job titles like “eligibility interviewers, government programs” are searched for less frequently than less technical titles, such as “lawyer.” Accordingly, researchers decided to remove technical terms in a systematic fashion by comparing the relative search intensity of each potential job title against that of a reasonably common job title.[8. Google Trends returns results on a scale from 0 to 100, with 100 representing the highest search intensity for the terms queried within the selected region and time frame and zero the lowest.] The research team compared the search intensity results for each occupation with the search intensity of “childcare worker” using U.S. search interest in 2017. Any terms with search intensity below “childcare worker” were removed from the list of job titles. The reference occupation “childcare worker” was selected after researchers manually inspected the relative search popularity of various job titles and decided that “childcare worker” was popular enough that using it as a benchmark would remove many highly technical search terms. The global part of the analysis uses a different list of job titles meant to capture more general descriptions of the same occupations. The steps to create that list include: Start with the list of BLS job titles that had at least 100,000 people working in the occupation in the U.S. Remove all occupations with ambiguous job functions (“all other”, “Misc.”). Split all titles with composite job functions into individual job titles (For example, “models and demonstrators” to “models,” “demonstrators”). Change plural words to singular (“models” to “model”) to standardize across occupations. Manually inspect the list to ensure that the occupations were comprehensible and likely to describe human workers. This involved removing terms that might not apply to humans (such as tester, sorter) based on manual review of Google results. Replace technical job titles with more general ones when possible to simplify translations and better represent searches. For example, instead of searching for “postsecondary teacher,” the team searched for “professor,” and instead of “chief executive,” the team used “CEO.” Use Google Trends to filter unpopular job titles relative to a reference occupation (“childcare worker”), following the same procedure described above. Any terms with search intensity below the search intensity of the reference occupation were removed. Select the top 100 terms with the most popular search intensity in Google in the U.S. within the past year. Translate each job title and determine which form to use when multiple translations were available. Translations To conduct the international analysis, the research team chose to examine image results within a subset of G20 countries, which collectively account for 63% of the global economy. These countries include Argentina, Australia, Brazil, Canada, France, Germany, India, Indonesia, Italy, Japan, Mexico, Russia, Saudi Arabia, South Africa, South Korea, Turkey, the United Kingdom and the United States. The analysis excludes the European Union because some of its member states are included separately and China because Google is blocked in the country. Researchers used the official language of each country for each search (or the most popular language if there were multiple official languages), and worked with a translation service, cApStAn, to develop the specific search queries. To approximate search results for each country, researchers adjusted Google’s country and language settings. For example, to search jobs in India, job titles were queried in the Hindi language with the country set to India. Several countries in the study share the same official language; for example, Argentina and Mexico both have Spanish as their official language. In these cases, researchers executed separate queries for each language and country combination. The languages used in the searches were: Modern Standard Arabic, English, French, German, Hindi, Indonesian, Italian, Japanese, Korean, Portuguese, Russian, Spanish and Turkish. Many languages spoken in these countries have gender-specific words for each occupation term. For example, in German, adding “in” to the end of the word “musiker” (musician) gives a female connotation to the word. However, the word “musiker” may not exclusively imply “male musician,” and it is not the case that only male musicians can be referred to as “musiker.” In consultation with the translation team, researchers identified the gender form of each job that would be used when a person of unknown gender is referenced, and searched for those terms. The male version is the default choice for most languages and occupations, but the translation team recommended using the feminine form for some cases when it was more commonly used. For example, researchers searched for “nurse” in Italian using the feminine term “infermière” rather than the masculine “infirmier” on the advice of the translation team. In addition, some titles do not have a directly equivalent title in another language. For example, the job term “compliance officer” does not have an Italian equivalent. Finally, the same translated term can refer to different occupations in some languages. As a result, not all languages have exactly 20 search terms. Jobs lacking an equivalent translation in a given language were excluded. Data collection To create the master dataset used for both analyses, researchers built a data pipeline to streamline image collection, facial recognition and extraction, and facial classification tasks. To ensure that a large number of images could be processed in a timely manner, the team set up a database and analysis environment on the Amazon Web Service (AWS) cloud, which enabled the use of graphics processing units (GPUs) for faster image processing. Building this pipeline also allowed the researchers to collect additional labeled training images relatively quickly, which they leveraged to increase the diversity of the training set in advance of classifying the image search results. Search results can be affected by the timing of the queries: Some photos could be more relevant during the time the query is executed, and therefore have a higher rank in the search results compared with searches at other times. There are a number of filters users can apply to the images returned by Google. Under “Tools,” for example, users can signal to Google Image Search that they would like to receive images of different types, including “Face,” “Photo” and “Clip Art,” among other options. Users can also filter images by size and usage rights. For this study, researchers collected images using both the “photo” and “face” filter settings, but the results presented in this report use the “photo” filter only. Researchers made this decision because the “photo” filter appeared to provide more diverse kinds of images than the “face” filter, while also excluding clip art and animated representations of jobs. Removing irrelevant queries For occupations included across both the U.S. and international analysis term lists, some queries returned images that did not depict individuals engaged in the occupation being examined. Instead, they often returned images that showed clients or customers, rather than practitioners of the occupation, or depicted non-human objects. For example, the majority of image results for the term “physical therapist” showed individuals receiving care rather than individuals engaging in the duties associated with being a physical therapist. To ensure the relevance of detected faces, researchers reviewed all of the collected images for each language, country and occupation combination. For the U.S. analysis, there were a total of 239 sets of images to review. For the international analysis, there were 1,800 sets of images to review. Queries were categorized into one of four categories based on the contents of the collected images. “Pass”: More than half of collected images depict only individuals employed in the queried occupation. Overall, 44% of jobs in the U.S. analysis and 43% of jobs in the global analysis fell into this category. “Fail”: The majority of collected images do not depict any face or depict faces irrelevant to the desired occupation. In many languages, the majority of collected images for the occupation “barber” depict only people who have been to a barber, rather than the actual barber. In the analysis of international search results, this includes queries that return images of an occupation different from that initially defined by the English translation. For example, the Arabic translation of “janitor” returns images of soccer goalies when queried in Saudi Arabia. Because the faces depicted in these images are not representative of the desired occupation, we categorize these queries as “fail.” A total of 31% of jobs in the U.S. analysis and 37% of jobs in the global analysis fell into this category. “Complicated”: The majority of collected images depict multiple people, some of whom are engaged in the queried occupation and some of whom are not. For example, the term “preschool teacher” and its translations often return images that feature not only a teacher but also students. These queries are categorized as “complicated” because of the difficulty in isolating the relevant faces. A total of 23% of jobs in the U.S. analysis and 17% of jobs in the global analysis fell into this category. “Ambiguous”: Some queries do not fall into the other categories, as there is no clear majority of image type or it is unclear whether the people depicted in the collected images are engaged in the occupation of interest. This may occur if the term has many definitions, such as “trainer,” which can refer to a person who trains athletes or various training equipment, or if the term has other usage in popular culture, such as the surname of a public figure (“baker”) or the name of a popular movie (“taxi driver”). Just 2% of jobs in the U.S. analysis and 3% of jobs in the global analysis fell into this category. To minimize any error caused by irrelevance of detected faces in collected images, we remove all queries categorized as “fail,” “complicated” or “ambiguous” and only retain those queries categorized as “pass.” Machine vision for gender classification Researchers used a method called “transfer learning” to train a gender classifier, rather than using machine vision methods developed by an outside vendor. In some commercial and noncommercial alternative classifiers, “multitask” learning methods are used to simultaneously perform face detection, landmark localization, pose estimation, gender recognition and other face analysis tasks. The research team’s classifier achieved high accuracy for the gender classification task, while allowing the research team to monitor a variety of important performance metrics. Face detection Researchers used the face detector from the Python library dlib to identify all faces in the image. The program identifies four coordinates of the face: top, right, bottom and left (in pixels). This system achieves 99.4% accuracy on the popular Labeled Faces in the Wild dataset. The research team cropped the faces from the images and stored them as separate files. Many images collected do not contain any individuals at all. For example, all images returned by Google for the German word “Barmixer” are images of a cocktail shaker product, even with the country search parameter set to Germany. To avoid drawing inference based on a small number of images, researchers included only queries that have at least 80 images downloaded and 50 images with at least one face detected in the analysis. Across different countries, the number of faces detected in the images varied. Hindi and Indonesian had the most detected faces. This means that their images tend to feature more people in them than other languages. The table below summarizes the number of queries, number of images and number of faces detected. Overall, researchers were able to collect over 95% of the top 100 images that we sought to download. Training the model Recently, research has provided evidence of algorithmic bias in image classification systems from a variety of high-profile vendors.[9. See Buolamwini, Joy and Timnit Gebru. 2018. “Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification.” Proceedings of Machine Learning Research.] This problem is believed to stem from imbalanced training data that often overrepresents white men. For this analysis, researchers decided to train a new gender classification model using a more balanced image training set. However, training an image classifier is a daunting task because collecting a large labeled dataset for training is very time and labor intensive, and often is too computationally intensive to actually execute. To avoid these challenges, the research team relied on a technique called “transfer learning,” which involves recycling large pretrained neural networks (a popular class of machine learning models) for more specific classification tasks. The key innovation of this technique is that lower layers of the pretrained neural networks often contain features that are useful across different image classification tasks. Researchers can reuse these pretrained lower layers and fine-tune the top layers for their specific application – in this case, the gender classification task. The specific pretrained network researchers used is VGG16, implemented in the popular deep learning Python package Keras. The VGG network architecture was introduced by Karen Simonyan and Andrew Zisserman in their 2014 paper “Very Deep Convolutional Networks for Large Scale Image Recognition.” The model is trained using ImageNet, which has over 1.2 million images and 1,000 object categories. Other common pretrained models include ResNet and Inception. VGG16 contains 16 weight layers that include several convolution and fully connected layers. The VGG16 network has achieved a 90% top-5 accuracy in ImageNet classification.[10. The top-5 accuracy is calculated by counting the times a predicted label matched the target label, divided by the number of data points evaluated for the five categories with the highest probabilities.] Researchers began with the classic architecture of the VGG16 neural network as a base, then added one fully connected layer, one dropout layer and one output layer. The team conducted two rounds of training – one for the layers added for the gender classification task (the custom model), and subsequently one for the upper layers of the VGG base model. Researchers froze the VGG base weights so that they could not be updated during the first round of training, and restricted training during this phase to the custom layers. This choice reflects the fact that weights for the new layers are randomly initialized, so if we allowed the VGG weights to be updated it would destroy the information contained within them. After 20 epochs of training on just the custom model, the team then unfroze four top layers of the VGG base and began a second round of training. For the second round of training, researchers implemented an early-stopping function. Early stopping checks the progress of the model loss (or error rate) during training, and halts training when validation loss value ceases to improve. This serves as both a timesaver and keeps the model from overfitting to the training data. In order to prevent the model from overfitting to the training images, researchers randomly augmented each image during the training process. These random augmentations included rotations, shifting of the center of the image, zooming in/out, and shearing the image. As such, the model never saw the same image twice during training. Selecting training images Image classification systems, even those that draw on pretrained models, require a substantial amount of training and validation data. These systems also demand diverse training samples if they are to be accurate across demographic groups. To ensure that the model was accurate when it came to classifying the gender of people from diverse backgrounds, researchers took a variety of steps. First, the team located existing datasets used by researchers for image analysis. These include the “Labeled Faces in the Wild” (LFW) and “Bainbridge 10K U.S. Adult Faces” datasets. Second, the team downloaded images of Brazilian politicians from a site that hosts municipal-level election results. Brazil is a racially diverse country, and that is reflected in the demographic diversity in its politicians. Third, researchers created original lists of celebrities who belong to different minority groups and collected 100 images for each individual. The list of minority celebrities focused on famous black and Asian individuals. The list of famous blacks includes 22 individuals: 11 men and 11 women. The list of famous Asians includes 30 individuals: 15 men and 15 women. Researchers then compiled a list of the most-populous 100 countries and downloaded up to 100 images of men and women for each nation-gender combination, respectively (for example, “French man”). This choice helped ensure that the training data included images that feature people from a diverse set of countries, balancing out the over-representativeness of white people in the training dataset. Finally, researchers supplemented this list with a set of 21 celebrity seniors (11 men and 10 women) to help improve model accuracy on older individuals. This allowed researchers to easily build up a demographically diverse dataset of faces with known gender and racial profiles. Some images feature multiple people. To ensure that the images were directly relevant, a member of the research team reviewed each face manually and removed irrelevant or erroneous faces (e.g., men in images with women). Researchers also removed images that were too blurry, too small, and those where much of the face was obscured. In summary, the training data consist of 14,351 men and 12,630 women in images. The images belong to seven different datasets. Model performance To evaluate whether the model was accurate, researchers applied it to a subset equivalent to 20% of the image sources: a “held out” set which was not used for training purposes. The model achieved an overall accuracy of 95% on this set of validation data. The model was also accurate on particular subsets of the data, achieving 0.96 positive predictive value on the black celebrities subset, for example. As a final validation exercise, researched used an online labor market to create a hand-coded random sample of 996 faces. This random subset of images overrepresented men – 665 of the images were classified as male. Each face was coded by three online workers. For the 924 faces that had consensus across the three coders, the overall accuracy of this sample is 88%. Using the value 1 for “male” and 0 for “female,” the precision and recall of the model were 0.93 and 0.90, respectively. This suggests that the model performs slightly worse for female faces, but that the rates of false positives and negatives was relatively low. Researchers found that many of the misclassified images were blurry, smaller in size, or obscured. These results are also available in downloadable form here.