Mike Pekala's Projects
Numerical experiments related to AE detection.
This repository hosts simple examples of using deep networks to segment electron microscropy and other biological image data.
Software for generating dense (i.e. per-pixel) labels using semantic segmentation [lon2015]. The current implementation is based on the U-Net architecture [ron2015].
This repository hosts a small example for calibrating binary classifiers.
A simple bare-bones implementation of a conformal predictor for use in a "batch" setting.
An experiment with the "magic points" mechanism for developing interpolation-based surrogates.
This repository hosts an experiment for segmenting EM images, with the goal of working more quickly than exhaustive application of CNNs on a per-pixel basis.
Initial investigations into Gaussian processes for AM.
Deep Learning library for Python. Runs on TensorFlow, Theano, or CNTK.
An empirical study comparing MAXFUN to other feature pooling techniques.
Codes for the NIPS 2017 adversarial example competition.
Scattering transforms with shearlet-based filters.