Les réseaux de neurones vous connaissent un peu trop bien

[Tortue joviale]

 

 

JPMorgan Algorithm Knows You’re a Rogue Employee Before You Do

 

Wall Street traders are already threatened by computers that can do their jobs faster and cheaper. Now the humans of finance have something else to worry about: Algorithms that make sure they behave.

JPMorgan Chase & Co., which has racked up more than $36 billion in legal bills since the financial crisis, is rolling out a program to identify rogue employees before they go astray, according to Sally Dewar, head of regulatory affairs for Europe, who’s overseeing the effort. Dozens of inputs, including whether workers skip compliance classes, violate personal trading rules or breach market-risk limits, will be fed into the software.

“It’s very difficult for a business head to take what could be hundreds of data points and start to draw any themes about a particular desk or trader,” Dewar, 46, said last month in an interview. “The idea is to refine those data points to help predict patterns of behavior.”

 

JPMorgan’s surveillance program, which is being tested in the trading business and will spread throughout the global investment-banking and asset-management divisions by 2016, offers a glimpse into Wall Street’s future. An industry reeling from billions of dollars in fines for the actions of employees who rigged markets, cheated clients and aided criminals is turning to technology to police itself better. Failure to do so will provide ammunition for those pushing to separate trading operations from retail banks.

Surveillance Unit

At New York-based JPMorgan, the world’s biggest investment bank by revenue, the push comes after government probes into fraudulent mortgage-bond sales, the $6.2 billion London Whale trading loss, services provided to Ponzi-scheme operator Bernard Madoff and the rigging of currency and energy markets.

The company has hired 2,500 compliance workers and spent $730 million over the past three years to improve operations. Job postings show it is building a surveillance unit to monitor electronic and telephone communication in the investment bank.

E-mails, chats and telephone transcripts can be analyzed electronically to determine if employees are trying to collude or conceal intentions, said Tim Estes, chief executive officer of Digital Reasoning Systems Inc.

“We’re taking technology that was built for counter-terrorism and using it against human language, because that’s where intentions are shown,” said Estes, whose company counts Goldman Sachs Group Inc. and Credit Suisse Group AG as clients and investors, but not JPMorgan. “If you want to be proactive, you have to get people before they act.”

 

‘Slippery Slope’

Automated surveillance is necessary for Wall Street firms because billions of e-mails flow through each bank annually, overwhelming the ability of people to monitor them, according to Estes. Still, technology that predicts behavior, as in the 2002 science-fiction movie “Minority Report,” in which Tom Cruise plays a Precrime officer who hunts down murder suspects before they can act, raises ethical questions.

“What they’re trying to do is forecast human behavior,” said Mark Williams, a former Federal Reserve bank examiner who’s now a lecturer at Boston University’s Questrom School of Business. “Policing intentions can be a slippery slope. Do people get a scarlet letter for something they have yet to do?”

Care will be taken to strike the right balance in monitoring employees at JPMorgan, said Dewar, a former U.K. regulator. She’s responsible for helping executives at the investment bank implement the new controls, while Chief Control Officer Shannon Warren has oversight of the firm-wide effort.

The bank wouldn’t describe all of the inputs being used for its predictive program, which specific business it’s being tested on, or what steps will be taken if concerns are raised about an employee.

 

Legal Bills

A February memo from executives including Chief Operating Officer Matt Zames urged employees to flag compliance concerns to managers and reminded them that scandals hurt bonuses for everyone. Dedicated whistle-blower phone lines and e-mail addresses were created for workers to raise issues anonymously.

“The problem we saw last year in FX and the other unacceptable events have implications beyond just a one-time fine,” according to the memo, a copy of which was obtained by Bloomberg News. “They damage our reputation.”

New technology is half of a two-pronged effort to reduce legal bills. The other part involves a review of the firm’s culture -- reaching into every business and appointing more than 300 leaders in the investment bank -- to fix areas where lapses could occur, Dewar said. Training sessions use real JPMorgan incidents as examples so the lessons hit home, she said.

 

‘Confidence Level’

The program was hinted at in a report published in December on the bank’s website, “How We Do Business,” signed by CEO Jamie Dimon. It outlines ways the firm is improving compliance, including starting a global communications surveillance program.

“We recognized that enhancing market conduct would require using multiple preventive and detective levers in a coordinated way,” JPMorgan said in the report.

Meeting the company’s financial targets depends on reducing legal bills. The investment bank’s return on equity will rise to 13 percent from last year’s 10 percent largely by cutting legal and other expenses, according to a February presentation.

Thousands of investment bank and asset-management employees will be subject to the new predictive monitoring, said Dewar, who spent about a decade at the U.K.’s Financial Services Authority before joining JPMorgan in London in 2011.

The combination of new surveillance methods and an improved culture should lower the bank’s future legal bills, Williams said. Still, even Dewar acknowledges that the human element involves risks that can’t be eliminated.

“We’ll have a much greater confidence level about early detection,” she said. “I don’t think you could ever say it will be 100 percent.”

 

Source: Bloomberg

[Brock]

"source:Bloomerg"

 

c'est probablement Bloomerg qui fait les softs utilises par JP Morgan

 

 

[Rincevent]

"Computer-aided gaydars" : demain près de chez vous !

 

https://psyarxiv.com/hv28a/

 

Citation

Abstract

 

We show that faces contain much more information about sexual orientation than can be perceived and interpreted by the human brain. We used deep neural networks to extract features from 35,326 facial images. These features were entered into a logistic regression aimed at classifying sexual orientation. Given a single facial image, a classifier could correctly distinguish between gay and heterosexual men in 81% of cases, and in 74% of cases for women. Human judges achieved much lower accuracy: 61% for men and 54% for women. The accuracy of the algorithm increased to 91% and 83%, respectively, given five facial images per person. Facial features employed by the classifier included both fixed (e.g., nose shape) and transient facial features (e.g., grooming style). Consistent with the prenatal hormone theory of sexual orientation, gay men and women tended to have gender-atypical facial morphology, expression, and grooming styles. Prediction models aimed at gender alone allowed for detecting gay males with 57% accuracy and gay females with 58% accuracy. Those findings advance our understanding of the origins of sexual orientation and the limits of human perception. Additionally, given that companies and governments are increasingly using computer vision algorithms to detect people’s intimate traits, our findings expose a threat to the privacy and safety of gay men and women.

 

[Bézoukhov]

Tiens, je me demande si les mecs de Tinder ont déjà fait tourner des algorithmes de scoring sur les likes mutuels.

[0100011]

Y a beaucoup de biais dans l'étude. Notamment parce que les photos viennent d'un site de rencontre. 

[Freezbee]

@Rincevent Je n'ai pas regardé les données de plus près, mais une réflexion en passant :

 

Supposons que l'algorithme ait raison dans 90% des cas, et analysons une population de 1000 individus composée à 95% d'hétéros et à 5% d'homos (voir ce lien).

Alors les résultats donneront :

855 vrais hétéros / 5 faux hétéros

45 vrais homos / 95 faux homos

 

Je te laisse en tirer la conclusion... autant dire que tout le monde est hétéro, tu risques moins de te tromper.

[Bézoukhov]

il y a 8 minutes, Freezbee a dit :

Je n'ai pas regardé les données de plus près, mais une réflexion en passant :

 

Page  12 ; ils ont samplé pour avoir 50/50.

[Mathieu_D]

7 hours ago, Freezbee said:

Je te laisse en tirer la conclusion... autant dire que tout le monde est hétéro, tu risques moins de te tromper.

Oui alors dans le métier on connaît la problématique de la modélisation de classes rares, hein.

Tu peux chercher à maximiser autre chose que le taux de biens classés (exemple: https://en.m.wikipedia.org/wiki/Matthews_correlation_coefficient )

Mots clefs "imbalanced data" pour commencer la biblio.

[Rincevent]

Chers journalistes de presse maistream : si vous ne faites pas partie, des 10-20 % meilleurs de votre professions, #Learn2Code, now !

 

https://openai.com/blog/better-language-models/

[Bézoukhov]

Chercheur : on va faire un gros réseau de neurones ;

Ingénierieur : ben on va scrapper Reuters _0_.

[Mathieu_D]

Notez je dois à mon département com' un article de blog taggé IA, je compte bien faire comme ça.

[Rincevent]

Moralité : aujourd'hui, une IA fait aussi bien qu'un journaleux moyen pour inventer une fake news qui a tout du crédible.

 

Tu m'étonnes qu'OpenIA refuse de divulguer le code...

[Mathieu_D]

8 minutes ago, Rincevent said:

Tu m'étonnes qu'OpenIA refuse de divulguer le code...

Tu peux trouver facilement des modèles à entraîner pour générer du texte, ça se code assez facilement.

C'est la base d'apprentissage qui est compliquée à construire surtout. 

[Rincevent]

il y a 6 minutes, Mathieu_D a dit :

Tu peux trouver facilement des modèles à entraîner pour générer du texte, ça se code assez facilement.

C'est la base d'apprentissage qui est compliquée à construire surtout. 

Je sais bien ; jusqu'ici, un des grands succès était constitué par les modèles qui pondaient du post-moderne à la chaîne. Là, je crois qu'on a fait un petit saut tout de même, on en arrive à partir de deux ou trois phrases d'introduction à pondre un article entier avec sources, référence à des personnes, le tout devenant impossible à distinguer d'un vrai article (pas une brève, hein), true news ou fake news.

 

Regarde donc la section "samples".

[Neomatix]

il y a 24 minutes, Mathieu_D a dit :

Notez je dois à mon département com' un article de blog taggé IA, je compte bien faire comme ça.

Faire écrire un article sur l'IA par une IA ? T'as pas peur qu'elle soit biaisée ?