"The future is already here — it's just not very evenly distributed"
- William Gibson -
Research I am fascinated by the question: what makes a "good" representation? Because machines don't have any of the incredible evolutionary gifts we developed over millions of years, we have to find ways to somehow short-circuit that learning process. My research thus far has pre-dominantly focused on unsupervised (deep) machine learning, with special interest in methods exploring (non-trivial) manifold learning.
T.R. Davidson*, L. Falorsi*, N. De Cao*, T. Kipf, J.M. Tomczak, Hyperspherical Variational Autoencoders, Oral presentation at the 34th Conference on Uncertainty in Artificial Intelligence (UAI, 2018)
[arXiv] [code] [oral]
L. Falorsi*, P. de Haan*, T.R. Davidson*, N. De Cao, M. Weiler, P. Forré, T. Cohen, Explorations in Homeomorphic Variational Auto-Encoding, ICML Workshop on Theoretical Foundations and Applications of Deep Generative Models (ICML, 2018)
L. Falorsi, P. de Haan, T.R. Davidson, P. Forré, Reparameterizing Distributions on Lie Groups, Oral presentation at the 22nd International Conference on Artificial Intelligence and Statistics (AISTATS, 2019)
T.R. Davidson, J.M. Tomczak, E. Gavves, Increasing Expressivity of a Hyperspherical VAE, NeurIPS Workshop on Bayesian Deep Learning (NeurIPS, 2019)
T.R. Davidson, The Shape of a Black Box: A Closer Look at Structured Latent Spaces, Master's Thesis, University of Amsterdam, 2019