Pencil: Your own creative AI

Abstract:

AI is emerging as a transformative technology that is partnering with human artists in producing great works of art. Generative adversarial networks (GANs) have made serious progress in creating pieces of music, art and literature. GANs, however, tend to generate outputs that converge to a narrow subspace within the bounds of the training data. This makes it very hard to generate creative output that does not imitate the training data.

Creativity is very subjective and is often associated with finding/generating something novel, unexpected, and yet useful. At Pencil, the World's first creative AI for generating digital content and advertising at scale, we are building deep generative models that are creative, contextual and useful. In this talk, we present how Pencil's AI uses divergence in the generator to produce creative outputs that are not mere imitations of the training data.


Biodata:

Sharmili's research focuses on computer vision and machine learning technologies for various kinds of detection and recognition tasks. Prior to this she was with the image analysis team of A*STAR-NUS Clinical Imaging Research Centre headed by Dr. David Townsend where her research was geared towards personalized radiotherapy for cancers of the pelvis such as cervical, prostrate and colorectal carcinoma using Positron Emission Tomography/ Magnetic Resonance (PET/MR) imaging. She finished her PhD under Dr. Michael S. Brown at the School of Computing of National University of Singapore. During her PhD she developed advanced tools for radiological reporting and biomedical image analysis using routine clinical data.

Her main research interests lie in the fields of computer vision, machine learning, pattern recognition and biomedical image analysis.