In the office early setting up for our #NeurIPS2021 oral on Learning to Draw
10
7/7 All of these works build upon an approach to differentiable rasterisation of 2D lines, points and curves. An overview of this approach is on arxiv: arxiv.org/abs/2103.16194
7
6/7 Also in the @ML4CDworkshop workshop our artwork, "Perceptions" collaboratively generated by us and our AI will be premiered. Perceptions is a study focussed on the difference between different inductive biases applied to pre-COVID photos of travels to far away cities.
5/7 In the @ML4CDworkshop workshop we present "Physically Embodied Deep Image Optimisation", where we utilise our approach to differentiable rasterisation to generate vector strokes which in turn command a robot to draw using tradiation mediums.
4/7 In the @svrhm2021 workshop we present "Shared Visual Representations of Drawing for Communication: How do different biases affect human interpretability and intent?", where we extend the main conference paper and look in detail at inductive biases and human interpretation.
1
2
3/7 Our work in the main conference, "Learning to Draw: Emergent Communication through Sketching" explores how we can build agents that learn to draw and communicate through self-play. We show that with added inductive biases the drawing agent can communicate with humans.
1
3
Thanks @iclr_conf #ICLR2021. This was an unexpected and welcome surprise at the end of a manic week!
1
9
On the way to nyc for the acmmm tpc meeting and a few days of holiday #fooloo
Materials for the comp sci lectures and labs at the #ECSsummerschool next week all ready to go
2












