A major task in ground-based gamma-ray astrophysics analyses is to
separate events caused by gamma rays from the overwhelming hadronic
cosmic-ray background. In this talk we are interested in improving
the gamma ray regime below 1 TeV, where the gamma and cosmic-ray
separation becomes more difficult. Traditionally, the separation has
been done in particle sampling arrays by selections on summary
variables which distinguish features between the gamma and
cosmic-ray air showers, though the distributions become more similar
with lower energies. The structure of the HAWC observatory, however,
makes it natural to interpret the charge deposition collected by the
detectors as pixels in an image, which makes it an ideal case for
the use of modern deep learning techniques, allowing for good
performance classifers produced directly from low-level detector
information.