Welcome to my personal website here at the University of South Florida (USF). I am a faculty member in the Muma College of Business with a focus on data analytics and background in computer science. For my official USF webpage, see Don Berndt at USF Muma College. These webpages are place for me to organize and present some of my current projects, as well as highlight some selected projects from the past. This is also a chance to thank some of the really interesting mentors and colleagues (who were sometimes former students) that helped with the projects most important to me (somehow changing my life for the better in the process).
Here are some selected papers from more recent projects (with a couple highlights from past work). I am particularly excited about our recent work using agent-based modeling and simulation to better understand the dynamics of financial markets (see below for more details). The link below is for a paper on our US corporate bond market model, with an interesting look at liquidity and fire-sale dynamics. The journal paper is: Berndt, D.J.; Boogers, D.; Chakraborty, S.; McCart, J. Using Agent-Based Modeling to Assess Liquidity Mismatch in Open-End Bond Funds. Systems 2017, 5, 54. Please see the links below.
I have been fortunate to be part of a long-term research effort at the Veterans Health Administration, focused on data and text mining the VA electronic health record system. The papers below report our findings from a project on fall-related injuries. Falls can be traumatic events that start a downward spiral in health outcomes. Using machine learning to recognize risks early and possibly trigger staff interventions would be a wonderful application of technology. From a data science perspective, the experimental results reported in our case study regarding the robustness of machine learning methods still amaze me.
My most recent stream of research uses agent-based modeling and simulation (ABMS), with a focus on modeling financial market-based systems. Agent-based modeling is based on the interactions of autonomous agents typically implemented with simple, but important behaviors encoded as algorithms. The challenge of uncovering the most appropriate behaviors, then distilling those behaviors down to the most essential, make this research rewarding but difficult. When done well, the interactions of many agents making simple but impactful decision leads to interesting emergent outcomes exhibited by complex systems. For instance, we are building agent-based models of the US corporate bond market (see below) to explore market dynamics, as well as assess regulatory approaches and policies. We are also working on agent-based models of the inter-bank loan market and continue to look for other interesting sectors. Trying to analyze and understand the interactions and data generated by agent-based simulations is a challenge in itself, so we have also been looking at data visualization and sonification techniques (check out our preliminary efforts to generate music from market movements).
To learn more, see the project-specific website at GSRisk.org.
Learn More: An Agent-Based Model of the US Corporate Bond Market (BND)
Learn More: An Agent-Based Model of the Federal Funds Market (FED)
Learn More: Visualizing (and Sonifying) ABMS
Personal Note: This is an area I have been interested in since studying for my MS in computer science at Stony Brook University. I stumbled upon the Santa Fe Institute (SFI) and read some of the technical reports on complex systems (starting in the 1980s). I continued reading research on the topic, even programming some artificial life experiments, but never really had the opportunity to apply agent-based approaches to a significant problem. I was lucky to meet David Boogers, a finance practitioner, and now a great friend and colleague on a project at USF. We were fortunate to receive a grant from the Office of Financial Research (OFR) to explore computational approaches for better understanding systemic risk using granular data (administered by the NSF). I am truly excited to finally have a chance to work in this area. To learn more, see our grant-related summary and the official NSF Award 1445403 page.
Under Construction
My interest in data warehousing came from teaching the database course for our MS program at USF (which I did from the very start of our program). I first did a module on data warehsousing and became hooked after reading Ralph Kimball's Data Warehouse Toolkit (first edition), which is a case-oriented book that I still use as a reference in class. I had the good fortune to get a large grant with several other faculty members early on during my tenure at USF.
Under Construction
Under Construction
Personal Note: As noted below, Dr. Herbert Gelernter and my time working on the SYNCHEM project changed my life. However, serendipity continued to open new doors. While at Stony Brook, I met two other graduate students, David Gelernter and Nicholas Carriero. David Gelernter (Herb's son) and a PhD student at the time, went on to become an influential researcher and Yale University faculty member. Nick Carriero (my MS student office mate) went on to sudy with David at Yale and become a long-term collaborator on the Linda project and other research projects. After working as a Lisp programmer at Cognitive Systems, a New Haven-based start-up founded by Roger Schank (then a Yale faculty member), I was able to work for a couple years on the Linda project. I will always be grateful for the opportunity to work with both David and Nick, truly innovative thinkers whose influence on me continues to this day. Our research group was full of talented colleagues and friends, such as Mauricio Arango, that made these years uniquely rewarding. The ideas around distributed computing I learned on that project keep re-appearing in my work and I continue to introduce their work to my own students.
Learn More: SYNCHEM
The SYNCHEM system was a large knowledge-based heuristic problem-solving program in synthetic organic chemistry. SYNCHEM would discover synthetic routes for complex molecules without relying on human guidance. The system included an inference engine and large knowledge-base of generalized chemical reactions that were applied using graph algorithms to match potential reaction sites. The SYNCHEM system went through several generations, with a continually refined interface that could render chemical compounds and graphically present reaction pathways. The components were developed over time and implemented in many programming languages, such as Fortran, PL/I, a bit of Prolog and a lot of C/C++. This project was a wonderful way to move from graduate student to professional programmer, under the direction of a true scholar and mentor, as well as researchers like Gerald (Jerry) Miller (a long-standing chemist on the project and great colleague). The development of SYNCHEM continued for more than two decades and is probably one of the most complex AI systems ever built.
For me, one of the most interesting challenges was working with graph algorithms. The chemical compounds were represented as graphs, with reaction templates that could be embedded in many ways to find potential reaction sites. Each reaction represented another step in a set of expanding retro-synthetic pathways from complex compounds to simpler starting materials. We used SYNCHEM for lots of challenges such as finding new ways to make important compounds to simulating the release of chemicals in natural environments. The SYNCHEM project was my earliest exposure to research and publishing, including this article (thanks to my student colleague Joseph Benstock): Benstock_DAM1988SYNCHEM.pdf.
Personal Note: Life takes some truly unexpected turns and this project was the start of my graduate studies (and career) in computer science. I was finishing up some undergraduate courses in biochemistry with the hopes of going on to veterinary school. Professor Herbert Gelernter, the prinicpal investigator on the SYNCHEM project, came into our chemistry class looking for students with programming experience. I was one of the few to raise my hand (even though I had only taken some elective programming courses). After an interview, Dr. Gelernter offered me the opportunity to study for my MS in computer science, as well as a stipend from a research grant. That opportunity completely changed my life and I will always be indebted to him for my initial graduate studies, my first professional programming job, and his genuine kindness and insightful guidance while introducing me to the world of academic research. He almost single-handedly started my career—thank you again. Please read David Gelernter's article "A Life that Made Sense" for the essence of the man, as well as the biographical essay by his grandson Josh Gelernter for a look at Herbert Gelernter's remarkable story.
Technical Note: These web pages were developed using Brackets, a different type of editor with features such as quick edit and live preview, as well as being an open and extensible platform.