|I’m an experienced Data Science Leader and Researcher with over a decade of experience in modeling and analytics, specializing in both applied and theoretical machine learning, as well as experimental design. I’m a quantitative storyteller, and passionate about making data science more accessible through intuitive communication of technical concepts. I’ve worked on a number of business challenges, including sales forecasting, marketing problems, image processing, recommendation systems, and many others. Previously, I worked as a computational neuroscience researcher at Washington University in St. Louis, where I modeled the dynamics of working memory neural circuits.|
On this page you will find a few examples of some of the projects I’ve worked on. If you have any questions about anything on this page or need to reach me for any reason, you can contact me at firstname.lastname@example.org
artificial neural networksto answer many interesting questions and tell many fascinating stories. Below are some examples of artificial neural networks I've built and used.
artificial recurrent line attractor neural networkto model how information is remembered. The network uses recurrent connectivity to maintain the memory of an object even after that object has been removed from the network inputs.
neural network with back propagationto model these findings, The network can
learn how to compute the reaching to objects in space through trainingon a set of training data of objects and reaches. When we examine the neurons in the network we find that, like what we see in the actual brain, the artificial neural network uses the response gain of neurons to compute the distance between the eye position and the hand position.
technicalone and a
conceptualone. Take a look at the one you feel is most appropriate for you. I personally prefer the conceptual one because I think it's more fun, and contains just as much of the story as the technical version.
These are some reviews of this work:
“[This work] is
lucidand provides an
excellent integration of neural data and theoretical modeling.The
insightthat apparently contradictory neural and behavioral aspects of the [data] can be reconciled by shifting movement fields is
terrificjob. There were some really tough things there, and you did a darn good job of addressing them all.”
“[This work] takes a psychophysical phenomenon,
tests a reasonable hypothesisfor the underlying neural basis, finds that the data conflict with that hypothesis, and
offers a reasonable alternative explanation.[The] findings have
major implicationsfor our understanding of …”
“The documentation of the spatial and temporal features of residual memory activity in FEF is
very carefully done and valuable.”
“[This work] to my view represents
a particularly nice coupling of theory and experiment.”
time seriesof action potentials and local field potentials from the brain in both the
timedomains to help users uncover how these different brain signals relate to each other and contribute to driving behavior.
right experiment. Below is a document outlining one of my experimental designs to study working memory. This design won a highly competitive
N.R.S.A grant (~$150,000 in funding) from the National Institutes of Health.
These are some of the reviews of the design:
Very clever set of experimentsto determine whether decay of behavioral spatial working memory can be related to changes in the correlated tuning curves of neurons in DLPC with overlapping receptive fields.”
This is a very well developed research planaimed at testing complex cortical attractor networks in the extinction of spatial working memory…”
project managerto provide
analysis and data-based insightsfor a number of companies in various industries. Below you can see a couple of examples of the types of projects I worked on. Due to NDAs and other confidentiality agreements, most of the content has been redacted, but you can get a general idea of the types of projects and work I've done.
I was nominated for an award for my work on some of these projects.
“I am writing to let you know that you were recently nominated for the
Outstanding Consultant awardfor The BALSA Group. […] your nomination is in recognition of your
hard workon your most recent project and for BALSA as a whole. Though you were not eligible to win [because you are] a current Project Manager, we appreciate your continued dedication to our organization.”
do something about that.I created a new measure, the
probability to win. This measure takes into account all of the relevant factors, and condenses them into the single, intuitive measure.
training a Random Forest Classifierthat is trained on ~250,000 games. It is trained on all of the complicated features that define the state of each game. The model makes a prediction on whether a team will win after looking at all the features. The more confident the model is that a team will win, the higher the
probability to winmetric will be. The model can make this prediction at any point in time during the game and performs with very high accuracy.
give advice to novice playersby analyzing their most recent games. It uses an
event-triggered averageof player behaviors to determine what good players are doing before and after game-critical events. It then compares user teams to these metrics and gives them advice on how to improve.
Recurrent Neural Networks. I wanted to play with a different method. I thought that music may be thought of as a language, so I used a model that makes use of
Latent Dirichlet Allocationto extract
topicsfrom songs, and uses these topics to determine what song a short song-sample came from, and which composer wrote the song. It worked surprisingly well!