Math in Business Cathy O'Neil mathbabe.org. Outline of talk.

Math in Business

Cathy O'Neilmathbabe.org

Outline of talk

Outline of talk

What are the options?

Outline of talk

What are the cultural differences?

Outline of talk

What are the mathematical differences?

Outline of talk

Typical data scientist duties

Outline of talk

Ethics, and how to make theworld a better place



Working as an academic mathematician


Working at a government institution


Working as a quant in finance


Working as a data scientist

Cultural Differences


Feedback is slow in academics


Institutions are painfully bureaucratic


Finance firms are cut-throat


Startups are unstable


Outside academics, mathematicians have superpowers


Inside academics, you get more flexible hours

and summers off (!?)


Outside academics, you get rewarded for organizational skills

(punished within)


Academic freedom is awesomebut can come with insularity


You don't decide what to work onin business but the questions

can be really interesting


You can't share proprietary information with the outside world when you work

in business or for the government


On the other hand,sometimes you can and

it might make a difference


In business, more emphasis on shallower, short term results


On the other hand, you get much more feedback


As in research, you learn tools and apply them


You have to constantly be aware of the business context

(which can be good)

Mathematical Differences


Quants in finance usually come frommath and physics, data scientists

come from stats and CS


In academics the data is smallIn finance it’s mediumIn data science it’s big


In finance signal is tinyIn data science it’s big


Finance: time seriesMachine Learning: pile o’ data


Seasonality really matters(not user attributes as much

as user behavior)


In finance, can change frequencyof data to compress models -

but not in user modeling


The concept of exponential decayof signal is sacrosanct in finance


Thus online learning


Bayesian priors (two versions):generalized smoothness assumptions


Questions academics focus onseem weirdly specific


Questions academics focus onseem weirdly specific

(no offense)


Finance uses mostly linear regression for forecasting

(sometimes with trees)


Machine learners have allsorts of cool models

(a question of accuracy)


The hardest things to do probably have small signal

(or complicated relationships)



You spend time visualizing the datafor the sake of non-quants

• Reporting help


You spend time modeling(forecasting user behavior)


You spend time monitoringsignal versus noise


You spend time onbusiness development

How do I get a job like that?


Get a Ph.D.(establish your creativity)


Know your way around a computer(awk grep “rm –fr *”)


Learn python or R, MapReduce or pig


Get some domain knowledge (vocabulary)


Acquire some data visualization skills(knowing what is crucial)


Learn basic statistics


Read up on machine learning


Emphasize your communication skills and follow-through


Practice explaining what a confidence interval is

Ethics, and how to make theworld a better place

Other stuff

Data modeling is everywhere(good data modelers aren't)

Other stuff

Beware the authority of the inscrutable

Other stuff

Open source data, open source modeling

Other stuff

Modeler’s Hippocratic Oath

Other stuff

Meetups

Other stuff

Data hackathons

Other stuff

Data science for therest of us?

Math in Business Cathy O'Neil mathbabe.org. Outline of talk.

Documents