Computational Personalization
Robin van Emden, Davide Iannuzzi, and I just presented some of our work on lock-in feedback at Physics@Veldhoven. Here are a few pictures. First, the poster:
And second, our demo setup, with two “Heroes of Physics”:
For a more elaborate description of the setup check out: http://pavlov.tech/2017/01/21/flappy/
Cheers.
Hi,
We just released a new version of the oFMLR package for fitting (finite) mixtures of logistic regression in data streams. You can find the code here: https://github.com/MKaptein/ofmlr.
It is very easy to use:
library(devtools) install_github(“MKaptein/ofmlr”) library(ofmlr) ?ofmlr
suffices to get started.
Have fun, and let us know if anything does not work. Note that a publication describing the package in more detail including examples is coming up shortly.
Hi all,
First of all, a great 2017 to all of you! We are working on some awesome research projects in the coming year, so I am really looking forward to it!
However, for now I wanted to share our joint work that is presented at HICSS. Unfortunately I could not travel there myself, but our lab co-organizes a minitrack their and published some new work. We have the following contributions:
Finally, note that we are organizing a seminar series at JADS. For details see: http://www.jads.nl/jads-seminars.html
Hi all,
Hope you are doing well! In the (probably) last post in 2016 I want to take the opportunity to thank you for reading our updates and to wish you an awesome 2017!
Here are a few minor updates:
• Robin van Emden, Davide Iannuzzi and I finally submitted a more thorough version of our recent Arxiv paper (https://arxiv.org/abs/1607.08108); we hope to be able to announce more soon!
• Together with Reza Mohammadi we are working on implementing novel sampling methods for additive regression tree models. See our response at http://projecteuclid.org/euclid.ba/1472829062. We would kindly like to thank Matthew Pratola for sharing his code with us!
• We have finally written up a full paper about StreamingBandit; we hope to be able to share it soon! (for now, see http://projecteuclid.org/euclid.ba/1472829062)
• We will be presenting new work during “Physics@Veldhoven”; see also https://www.fom.nl/agenda/physicsatveldhoven/press/
• At JADS we are starting our seminar series. On the 24th of January we will have Frank Buytendijk speaking, and on the 31st Martijn Willemsen will take the stage; we are lookking forward to the inspiring talks (http://www.jads.nl/jads-seminars.html).
Well, we will keep it at this for now. Just one final note: we are looking for PhD candidates. If you are interested, send me a mail with your resume at maurits [at] maurits kaptein [dot] com.
Thanks!
Hi all,
A quick one to share this new paper: http://dx.doi.org/10.1027/1614-2241/a000116
It’s pretty cool, so have a look!
Hi all,
I am happy to announce that the work Prof. Davide Iannuzzi, Robin van Emden, and I did on the decoy effect (using lock-in feedback for optimal placements of decoys) has now been published! You can find the work at: http://www.palgrave-journals.com/articles/palcomms201682
Besides this note that we have pushed another update of StreamingBandit quite recently (https://github.com/MKaptein/streamingbandit) and hope to publish an overview article about the development and use shortly.
Cheers!
In a recent project we used Lock in Feedback to “converge” to the most beautiful face using sequential observations. The measurement tool we used was developed by Robin van Emden and is quite cool. He is now sharing it here: http://pavlov.tech/2016/10/21/facial-aspect-ratio-explorer/
Cheers.
Hi All,
The last few weeks have been busy teaching a new course called “Data Statistics”; I am teaching the course together with Prof. Dr. Edwin van den Heuvel and its pretty awesome. We will try to share the lecture notes here soon.
Also, good news, our (Robin van den Emden, Davide Iannuzzi, and Maurits Kaptein) paper on the decoy effect has been accepted by Palgrave Communications; we are awaiting the proofs and will than share a copy.
Finally, I have submitted an ERC starting grant application; here is the abstract:
“In 2011 Amazon.com listed the book The Making of a Fly for a mere $23,698,655.93 (plus $3.99 shipping). The incident is a famous example of autonomous decision making by computers gone awry. As humorous as this glitch may be, erroneous decisions by computers could well be far less amusing; what if your digital health assistant gives you the fully automated and fully personalized, but also quite deadly, recommendation to take 15 grams of paracetamol?
On the other hand, computers could be extremely helpful for making sequential decisions in the face of uncertainty. It is not far-fetched that a computer could learn the exact dosage of paracetamol that is best for you personally, outperforming your general practitioner. These algorithms already exist in a narrow sense; in online selling, for example, the prices of products are often automatically determined and personalized. However, these algorithms are not feasible outside the online selling domain since their success depends on assumptions that are violated in other domains. Furthermore, existing methods can be regarded as a “black box,” because it is unclear why a given result is produced. When similar methods are used in impactful domains such as healthcare, this lack of information is unacceptable: we need to keep the human in the loop to inspect, and possibly alter, computer-suggested decisions. Finally, when these algorithms are used to select personalized treatments, we need to ensure the privacy of the people involved.
Combining my expertise in human-computer interaction (HCI) and methods for sequential decision making, I have designed a research program to develop responsible methods for shared sequential decision making by computers and humans. The program is developed in four projects in which my team and I will relax current assumptions, allow humans to inspect and alter computer-generated decisions, and develop methods to ensure privacy when sequential decision algorithms are used to personalize treatments.”
Let’s see how that goes….
Finally, our data science program is alive and kicking with many cool events (check http://www.jads.nl), and my new book is coming up…
Hi,
After some silent weeks in summer, a little update. Over summer a number of articles that we worked on before summer have been accepted / published. Let’s do a little list:
That’s it for updates on publishing (although a very cool paper using LiFt was rejected as well last month — a pre-print can be found here: https://arxiv.org/abs/1607.08108).
Also, there was some coverage of the best paper that Jorrit Siebelink, Peter van der Putten, and Maurits Kaptein wrote for Intetain. See it here: http://blog.eai.eu/understandi
Finally, note that JADS has officially started! (www.jads.nl). We hope to see you all soon in Den Bosch where we are working on a bunch of new research projects! And, please do join us for http://www.jads.nl/big-data–dance.html.
Cheers,
Maurits
Hi all,
We just posted a new paper on arXiv; its a cool combination of Lock-in Feedback, and the beauty of avatars. The abstract reads:
“Lock-in feedback circuits are routinely used in physics laboratories all around the world to extract small signals out of a noisy environment. In a recent paper (M. Kaptein, R. van Emden, and D. Iannuzzi, paper under review), we have shown that one can adapt the algorithm exploited in those circuits to gain insight in behavioral economics. In this paper, we extend this concept to a very subjective socio-philosophical concept: the concept of beauty. We run an experiment on 7414 volunteers, asking them to express their opinion on the physical features of an avatar. Each participant was prompted with an image whose features were adjusted sequentially via a lock-in feedback algorithm driven by the opinion expressed by the previous participants. Our results show that the method allows one to identify the most attractive features of the avatar.”
Give it a try here: https://arxiv.org/abs/1607.08108
Cheers!
Maurits
ps. It might be quite over summer / August, we will post more in September.
