Changepoint Detection in the Presence of Outliers

arXiv.org Machine Learning

Many traditional methods for identifying changepoints can struggle in the presence of outliers, or when the noise is heavy-tailed. Often they will infer additional changepoints in order to fit the outliers. To overcome this problem, data often needs to be pre-processed to remove outliers, though this is difficult for applications where the data needs to be analysed online. We present an approach to changepoint detection that is robust to the presence of outliers. The idea is to adapt existing penalised cost approaches for detecting changes so that they use loss functions that are less sensitive to outliers. We argue that loss functions that are bounded, such as the classical biweight loss, are particularly suitable -- as we show that only bounded loss functions are robust to arbitrarily extreme outliers. We present an efficient dynamic programming algorithm that can find the optimal segmentation under our penalised cost criteria. Importantly, this algorithm can be used in settings where the data needs to be analysed online. We show that we can consistently estimate the number of changepoints, and accurately estimate their locations, using the biweight loss function. We demonstrate the usefulness of our approach for applications such as analysing well-log data, detecting copy number variation, and detecting tampering of wireless devices.


Wimbledon 2017: Adrian Mannarino penalised for barging ball boy

BBC News

Adrian Mannarino receives a code violation for bumping into a ball boy during his Wimbledon second-round win against Yuichi Sugita.


Firms face 20million fines for losing your private data

Daily Mail - Science & tech

Companies will face fines of up to 20million if they lose customers' personal data in cyber-attacks. A damning report by MPs called for watchdogs to be given the ability to hammer firms in the pocket if they fail to safeguard sensitive information. Bosses should also be penalised if their business suffers a data breach โ€“ with their own pay and perks linked to effective online security, the culture, media and sport select committee has said. And criminals who hack and sell private information โ€“ including names, addresses, phone numbers and bank details โ€“ should be jailed for up to two years, according to the cross-party panel. The far-ranging recommendations were included in a report, dubbed a'giant wake-up call', which was triggered by a series of huge data losses at communications giant TalkTalk.


Shana Grice murder trial: Teenager penalised by police

BBC News

A teenager complained about her ex-boyfriend to police several times, but was penalised for wasting their time before he went on to murder her, Lewes Crown Court has been told. Jurors heard Shana Grice contacted the Sussex force over months - but at one stage was given a fixed penalty notice. Miss Grice, 19, was found with her throat slit in her bedroom, which was set alight in Portslade, East Sussex, last August. The prosecution gave a timeline of when police were contacted. Jurors heard Miss Grice told police in February she was being stalked and Mr Lane had hid outside her home, sent unwanted flowers, and left a note on her new boyfriend's car which said "Shana will always cheat on you."


Heather Watson penalised for swearing in Wimbledon doubles defeat

BBC News

Heather Watson was docked a point for swearing as she and Tatjana Maria lost 3-6 7-6 (7-5) 6-4 to Katerina Siniakova and Barbora Krejcikova in the women's doubles last eight. It followed an earlier warning for hitting the ball off the court. But Jamie Murray stayed on course to defend his mixed doubles title with new partner Victoria Azarenka. Scotsman Murray and former singles world number one Azarenka moved into the quarter-finals with a 7-6 (8-6) 6-3 win over Matwe Middelkoop and Johanna Larsson. Watson, who lost in the mixed doubles on Tuesday, said after her defeat against the third seeds: "I asked everybody behind me out loud, 'What did you hear me say?' "I just think it's really pathetic.