Welcome to my site. Here you can find out who I am, link quickly to my publications and software for download, locate educational and research resources regarding learning classifier systems, and even view some of my extracurricular projects. You can expect many more additions to the website in the future. You can also follow my blog below.
Free Hit Counters
In the normal theme of my blog, this is my first posting in about a year. I should point out that I recently moved to UPenn, starting a position as a research associate in the lab of Dr. Jason Moore in the Perelman School of Medicine. I'm excited to be in a new environment, setting up new collaborations, and exploring a new place.
Today an article on ExSTraCS that focuses on demonstrating that this learning classifier system is approachable, accessible, and flexible was published in the SIGEVOlution Newsletter. The aim of this publication was to show that learning classifier systems, have evolved to a point where they are much easier to use. We demonstrate the operation of ExSTraCS using a simple 6-bit multiplexer problem. You can download the pdf article HERE.
I've posted a number of new things up on this webpage. Check out the new publication links under 'Publications', the new easily scrollable google map under 'About Me', a link to the E-Learning Module I created under 'Teaching', and links to the new ExSTraCS algorithm for download under 'Software'. Also, check out the new tab, dedicated to our ExSTraCS algorithm. Here you will find a description of the algorithm, a schematic, and links to the most recent version of the software, user's guide, and related publications. Enjoy!
It's been over a year since I first put this website together, so I think it's time to update the content and try to improve the look. Let me know if you have any ideas or feedback about how it looks, and what content you might want to see added.
I am pleased to announce that we have recently posted the second version of ExSTraCS freely available for download on sourceforge. ExSTraCS 2.0 has added a new rule specificity limit that dramatically improves the algorithm's ability to scale up to bigger problems and datasets. The addition of the rule specificity limit has let to a number of fairly dramatic changes to the underlying code, and manner in which ExSTraCS discovers new rules. Additionally, much attention has been paid to simplifying the algorithm's use. The associated users guide has been completely updated and summarizes all changes, improvements and bug fixes implemented in version 2.0 (including some further fixes made in versions 2.01 and 2.02). As of today the research paper corresponding to this software release is currently "in review". Please contact me with any questions or issues that you may have with this software as it is considered to be a Beta release.
I'm pleased to report that the first version of ExSTraCS is now available for download on Sourceforge. This code pairs with our first publication on ExSTraCS that was accepted at Parallel Problem Solving in Nature (PPSN) 2014. ExSTraCS is a Michigan-Style Learning Classifier System (or more generally, a stochastic machine learning algorithm) designed specifically for data mining, classification, prediction, and knowledge discovery tasks in noisy, complex, supervised learning problems. ExSTraCS v1.0 is flexible, able to handle discrete or continuous attributes in the dataset, and binary or multiclass endpoints. ExSTraCS integrates a number of recent successful LCS algorithmic components into a single, platform for application and further development. Specifically ExSTraCS combines, expert knowledge covering, attribute tracking, attribute feedback, rule compaction strategies, and a flexible knowledge representation scheme, along with a number of other algorithmic improvements designed to make ExSTraCS more flexible, user friendly, and powerful.
While I am looking forward to one of my favorite conferences (Genetic and Evolutionary Computation Conference or GECCO), it's shaping up to be an extremely busy trip. Right now, i'm co-organizing and running a day-long workshop on Learning Classifier Systems (LCSs) called IWLCS at which I am giving a talk on our accepted paper, as well as a secondary invited talk on our new algorithm (ExSTraCS). Additionally I am co-presenting a 2 hour tutorial on LCS with Will Brown (the driving force behind both the tutorial and an LCS textbook that we are co-authoring). And finally, I am chairing a session on classification in LCS. So much to prepare, but it should be a very interesting and exciting year.
Yesterday was a good day. I finally dug out all the bugs in the code for my new LCS algorithm (ExSTraCS). It was a beastly task, but it's finally done. I'm quite excited, because, now i can finally finish implementing and test a huge assortment of new ideas i have to improve and expand the algorithm.
For the past couple weeks i've been struggling to track down every last little bug in my new algorithm. I know it should work and yet....perhaps because I was not technically trained in computer science, it feel like i'm scrambling to assemble a 2000 piece, solid black jigsaw puzzle. So far, in my search for the underlying issue, i've stumbled across several minor bugs. While i've said this to my lab-mates several times over the last couple weeks, I think i've finally found 'THE' remaining bug causing the minor lingering issues. Going to run a complete test tonight to confirm, but if it works, i'm going to be ecstatic in the morning.
Two weeks ago, long-time hometown friend Michael Carpanzano launched a Kickstarter campaign for his 'Nuplug' invention and 2X award winning business model (The Connecticut Business Model Competition and Giants Entrepreneurship Challenge in Grand Forks, N.D). It’s incredible having witnessed Mike take one of his invention concepts and bring it within arms-reach of reality over the last couple years. I highly encourage you to at least take a look at his Kickstarter page and video below, summarizing his simple but slick and practical 'Nuplug' product. If you want to get a 'Nuplug' of your own, or you simply want to encourage a small start-up business venture, consider pledging to his Kickstarter campaign. Currently Nuplug is just shy of 60% funded. There are only two weeks left to reach the 100% goal which will allow final development and a first shipment of Nuplugs to contributors who pledge at least $29. Thanks!
To view the Nuplug Kickstarter page and make a pledge click HERE.
1-10 of 11