David W. Romero
Vrije Universiteit Amsterdam
EDL P16-24 P1: Deep Learning as a Service
Research assignment Data-efficient Deep Learning |
I am a 3rd year PhD Student at the Vrije Universiteit Amsterdam under supervision of Erik Bekkers (UvA), Jakub Tomczak (VU) and Mark Hoogendoorn (VU). My research is focused on the development of efficient deep learning architectures, particularly towards settings with scarce (labeled) data availability. To this end, I am specially interested in neural architectures involving (1) extensive parameter sharing (e.g. equivariant networks), (2) input-depending computing architectures, e.g., attention mechanisms, and (3) architectures that efficiently leverage long-term dependencies. |
Publications |
| Co-Attentive Equivariant Neural Networks: Focusing Equivariance on Transformations Co-Occurring in Data.D.W. Romero, M. Hoogendoorn. International Conference on Learning Representations (ICLR), 2020. Conference proceedings. Open access-green. Publicly available at ArXiv. Attentive Group Equivariant Convolutional Networks. D.W. Romero, E.J. Bekkers, J.M. Tomczak, M. Hoogendoorn. International Conference on Machine Learning (ICML), 2020. Conference proceedings. Open access-green. Publicly available at ArXiv. Wavelet Networks: Scale Equivariant Learning From Raw Waveforms. D.W. Romero, E.J. Bekkers, J.M. Tomczak, M. Hoogendoorn. ArXiv Preprint, 2020. Open access-green. Group Equivariant Stand-Alone Self-Attention For Vision. D.W. Romero, J.B. Cordonnier, International Conference on Learning Representations (ICLR), 2021. Conference proceedings, Open access-green, publicly available at ArXiv. CKConv: Continuous Kernel Convolution For Sequential Data. D.W. Romero, A. Kuzina, E.J. Bekkers, J.M. Tomczak, M. Hoogendoorn. Conference proceedings, ArXiv Preprint, Open access-green, publicly available at ArXiv. |
Personal information Website Google Scholar Github |