Talks n' Beer IV

Dear all,

our next meetup takes place tomorrow at Found Fair.

• Richard Guillemot asks how much is a beer (see abstract below).

• Eugen Funk talks about his experiences with the eigen-sparse module he already mentioned in his last talk.

• I would like to advertise a very nice paper I recently read. It is about performance metrics for supervised learning and so will my talk be.

Please, don't forget to cancel your registration if you cannot come. There are some guys on the waiting list. You can make them happy!

I'm looking forward to meeting you all (again).

Greetings,

Alex

---

Abstract: Richard Guillemot

How much for a beer?

An entrepreneur wants to create a new mobile application, which will inform users of the price of beer around. To create such an application, he must go to each bar in the town and collect prices. Due to the large number of bars and the size of the city, this operation will be costly and time-consuming for a single person. Our entrepreneur is considering asking users to collect prices. He will repay the price of the beer to incent users to work for him. He believes naively that it will cost him the sum of the price of a beer from each bar. Unfortunately he ignores the dishonesty of some users. They try to be repaid without going into the bar. Our entrepreneur cannot go himself to each bar to check the information. This is quite a pity because the cheaters reduce the quality of the information collected and increase indefinitely the cost of the price collection. So we propose that our entrepreneur model the behavior of users, in order to estimate the cost of the data collection more realistically. Our approach is quite passive; we are suffering from the dishonesty of some user without fighting against.

We would like to ask to the members of “Algorithms & Data Challenges in Berlin” the suitability of using a data analysis algorithm to detect the cheaters and then ignore the random prices they send, stop paying them and save money. We want later to simulate each algorithm with our behavioral model. We could then compare the efficiency of each algorithm and calculate the cost of the collection more accurately.During the talk, we will present our behavioral model to provide a frame for the algorithm we are looking for.