50 Fun Facts About Data Science

The Student's t-test, a staple of data analysis, was devised to monitor beer quality at which brewery?

Guinness preferred its staff to publish under pen names, so William Sealy Gosset signed his 1908 paper as Student. His test was built for judging barley and stout from very small samples.

Which famous Victorian drew rose-shaped polar area diagrams of soldiers' deaths in a Crimean War hospital?

Florence Nightingale's charts made a grim point: ten times more soldiers died of disease than of battle wounds. She went on to become the first female member of the Royal Statistical Society.

Hadoop, the big-data framework, was named by creator Doug Cutting after his young son's toy of which animal?

Elephant it was: Doug Cutting was working at Yahoo when he borrowed the toy's name. The software had begun life inside Nutch, an open-source web search project.

Scientists renamed 27 human genes by 2020 because which software kept turning symbols such as MARCH1 into dates?

Excel read MARCH1 as 1-Mar and SEPT2 as 2-Sep, a problem geneticists first reported in 2004. The naming committee finally changed MARCH1 to MARCHF1 and SEPT1 to SEPTIN1.

Kaggle's 'start here' contest for machine learning beginners asks who survived which maritime disaster?

Titanic passenger records are the raw material. The challenge, subtitled Machine Learning from Disaster, is pitched as a first project for getting familiar with the basics.

Which company's $1 million data science contest ended in 2009 when a team beat the in-house rating predictions by 10%?

Netflix paid out to a team called BellKor's Pragmatic Chaos, which bettered the company's own Cinematch system by 10.06%. The open contest had been running since 2006.

British mathematician Clive Humby, who helped launch Tesco's Clubcard, said in 2006 that data is the new what?

Oil was the comparison, and it was soon stretched further: like crude, raw data is valuable but cannot really be used until it has been refined.

Physician John Snow's 1854 dot map of a London neighbourhood traced which deadly disease to a single water pump?

Cholera cases clustered around the Broad Street pump in Soho, and Snow's evidence persuaded the local council to remove its handle. He is counted among the founders of modern epidemiology.

A famous 2013 word-vector result found that 'king' minus 'man' plus 'woman' lands closest to which word?

Queen came out nearest, which meant a model fed only raw text had picked up the male/female relationship by itself, as a consistent offset between word vectors.

Michael Lewis's book Moneyball follows which baseball team's use of statistics to compete on a small budget?

Oakland Athletics general manager Billy Beane leaned on sabermetrics to build his roster. His 2002 team spent $44 million on salaries and still competed with the $125 million Yankees.

Tyler Vigen's Spurious Correlations website famously charted divorce rates against consumption of which spread?

Margarine consumption and divorce tracked each other closely for almost ten years in his graph. Vigen, then a law student, wrote a program to mine datasets for such coincidences.

Arthur Samuel, who coined the term machine learning in 1959, wrote a famous self-learning program for which game?

Checkers was the testbed for one of the world's first successful self-learning programs. Samuel first got it running on one of the earliest commercial computers.

Nate Silver named his polling site FiveThirtyEight after the total number of what in American politics?

Electoral College votes total 538. Silver launched the site in March 2008, and his model went on to call 49 of the 50 states correctly in that year's presidential election.

Charles Minard's celebrated 1869 flow map shows which commander's army shrinking on the march to Moscow and back?

Napoleon's troop numbers form a band that narrows across the page, set against distance, direction and temperature. It has been described as possibly the best statistical graphic ever drawn.

Google launched a service in 2008 that tried to track outbreaks of which illness from people's search queries?

Influenza proved harder to track than hoped: the service consistently overestimated cases from 2011 to 2013, once predicting twice the doctors' visits the CDC recorded. It stopped publishing estimates in 2015.

Ronald Fisher's 1936 dataset, a classic first exercise in machine learning, measures petals and sepals of which flower?

Iris flowers of three species were measured by botanist Edgar Anderson, who gathered two of the species from a single pasture on Canada's Gaspé Peninsula on the same day.

Which publication declared data scientist 'the sexiest job of the 21st century' in a much-quoted 2012 article?

Harvard Business Review ran the piece in its October 2012 issue. Co-author DJ Patil went on to serve as the first Chief Data Scientist of the United States, from 2015 to 2017.

Which 18th-century English minister is the namesake of the theorem for updating a probability as new evidence arrives?

Thomas Bayes never published his most famous idea. His notes were edited by Richard Price and printed only after his death.

The bell-shaped normal distribution is also named after which German mathematician, who computed the orbit of Ceres?

Gauss gives the bell curve its other name, the Gaussian distribution, though some authors credit its discovery to Abraham de Moivre, who published an early form decades before him.

Abraham Wald's wartime study of damage on bombers is the textbook example of which statistical trap?

Survivorship bias is the error of judging only from what made it through. Wald's memos estimated how vulnerable each part of an aircraft was from the damage seen on planes that returned.

Which average is favoured for incomes because a few huge salaries barely move it?

Median income is the usual yardstick because increases in the largest incomes alone have no effect on it, whereas they drag the mean upward.

Which statistical term for data points far from the rest also titles Malcolm Gladwell's 2008 book on success?

Outliers examines what lies behind unusually high levels of success. It popularised the 10,000-hour rule, though the authors of the original study dispute Gladwell's use of it.

Which machine learning problem arises when a model memorises its training data and then stumbles on unseen data?

Overfitting is what happens when a model treats noise as if it were real structure. Give it as many parameters as observations and it can predict its training data perfectly by memorising it.

Which kind of experiment did Google first run in 2000 to settle how many search results to show?

A/B testing splits users by chance between variant A and variant B to see which does better. Google went on to test dozens of hyperlink hues to find the colour that drew the most clicks.

In the computing maxim GIGO, a warning that flawed input data gives flawed results, both Gs stand for which word?

Garbage in, garbage out reached a wide public through a 1957 newspaper article about US Army mathematicians and their work with early computers.

Drew Conway's much-shared picture of data science sets hacking skills, statistics and subject expertise in which form?

Venn diagram circles place data science where all three skills meet. Conway marked the overlap of hacking skills and subject expertise, with no statistics, as the danger zone.

Which Greek letter stands for standard deviation and names the quality programme Motorola introduced in 1986?

Sigma, written σ, is the usual symbol. Motorola's Six Sigma scheme refers to the share of a normal curve lying within six standard deviations of the mean, used to represent a defect rate.

Which British political consultancy closed in 2018 after harvesting data from millions of Facebook profiles?

Cambridge Analytica got hold of details from up to 87 million profiles through a third-party app. The scandal put Mark Zuckerberg in front of the US Congress in April 2018.

Herman Hollerith's tabulating machines sped up the 1890 US census by reading data stored on what?

Punched cards were Hollerith's answer to an 1880 count that had taken about eight years to process. His concept went on to dominate data processing for nearly a century.

Which branch of AI earned three of its pioneers the 2018 Turing Award?

Deep learning takes its name from the layers its data passes through, which can number from three to several thousand. The field draws its inspiration from biological neuroscience.

Which type of AI training, driven by rewards rather than labelled examples, did DeepMind use to master Atari games?

Reinforcement learning let DeepMind's system teach itself Breakout, Pong and Space Invaders using nothing but the raw pixels on screen as input.

Which method, trademarked by Leo Breiman and Adele Cutler, combines the predictions of many decision trees?

Random forest models grow a multitude of decision trees, each seeing a random selection of features, and pool their results. Tin Kam Ho created the first such algorithm in 1995.

Which neural network design, unveiled by Google in the 2017 paper Attention Is All You Need, is the T in GPT?

Transformer is the final word of generative pre-trained transformer. OpenAI was first to apply generative pre-training to the design, releasing GPT-1 in 2018.

Which British-Canadian 'godfather of AI' shared the 2024 Nobel Prize in Physics for his work on neural networks?

Geoffrey Hinton used the attention to call for urgent research into AI safety, saying people must work out how to control systems smarter than themselves. He shared the prize with John Hopfield.

The Python library pandas is named partly after which econometrics term for records of the same subjects over time?

Panel data follows the same individuals across several time periods. Wes McKinney began building the library in 2008 at the investment firm AQR Capital to analyse financial figures.

SQL, the standard language for querying relational databases, was first developed in the 1970s at which company?

IBM researchers Donald Chamberlin and Raymond Boyce first called it SEQUEL, short for Structured English Query Language, which is why many people still pronounce SQL that way.

Which online learning company did Stanford professors Andrew Ng and Daphne Koller launch in 2012?

Coursera followed fast: Ng and Koller began offering their Stanford courses online in autumn 2011 and soon left the university to launch the company.

Which simulation method is named after a casino, in a nod to mathematician Stanislaw Ulam's gambling uncle?

Monte Carlo was picked as a code name because the work, done with John von Neumann on nuclear weapons at Los Alamos, was secret.

Which table judges a classifier by setting each predicted class against the actual one, row by column?

Confusion matrix cells on the diagonal count the cases a model got right, so everything off the diagonal shows which classes it mixes up. It is also known as an error matrix.

Which effect, famously seen in Berkeley's graduate admissions, makes a trend reverse when groups are combined?

Simpson's paradox takes its name from the statistician who described it in a 1951 paper, though Karl Pearson and Udny Yule had noted similar effects around the turn of the century.

Which 1973 set of small datasets shares near-identical summary statistics yet looks wildly different when graphed?

Anscombe's quartet was built by statistician Francis Anscombe to show why data should be graphed before it is analysed, against the belief that calculations are exact but graphs are rough.

Which American statistician coined the word bit, invented the box plot and championed exploratory data analysis?

John Tukey is also credited with the first published use of the word software. He divided his time between Princeton University and Bell Laboratories.

Which magazine wrongly forecast Roosevelt's defeat in 1936 after polling its readers, car owners and phone users?

The Literary Digest mailed ten million ballots and got 2.38 million back, but its lists skewed wealthy. George Gallup called the result correctly from a sample of just 50,000.

The ROC curve used to judge classifiers was first developed in World War II to assess operators of which technology?

Radar gave the curve its full name, receiver operating characteristic. After Pearl Harbor the US military studied how well operators could pick out Japanese aircraft from their signals.

Which law, a favourite of fraud investigators, says about 30% of real-world numbers begin with the digit 1?

Benford's law also predicts that 9 leads less than 5% of the time. Simon Newcomb had stated it in 1881, well before physicist Frank Benford did in 1938.

Andrey Markov illustrated his famous chains in 1913 by counting vowels and consonants in which Pushkin verse novel?

Eugene Onegin supplied 20,000 letters, which Markov treated purely as a string of symbols. The transitions between vowels and consonants let him show that chains of dependent events could be modelled rigorously.

Which optimisation method, first suggested by Cauchy in 1847, is often likened to walking down a mountain in fog?

Gradient descent takes repeated steps in the steepest downhill direction, using only local information. A simple extension of it is the basic algorithm used to train most modern neural networks.

Which clustering algorithm, first proposed by Stuart Lloyd of Bell Labs in 1957, groups points around centroids?

k-means began as a technique for pulse-code modulation, and Lloyd's method went unpublished in a journal until 1982. James MacQueen first used the name in 1967.

Which Danish Turing Award winner proposed data science as a better name for computer science in a 1974 book?

Peter Naur made the case in his Concise Survey of Computer Methods, wanting a name that reflected the growing emphasis on data-driven methods. He is also the Naur in Backus–Naur form.

Which open-source machine learning library did the Google Brain team build for internal use and then release in 2015?

TensorFlow is written in Python, C++ and CUDA, and it came out under the Apache License 2.0, which leaves anyone free to use it.