Scholar
Alexander Schwing
Google Scholar ID: 3B2c31wAAAAJ
University of Illinois at Urbana-Champaign
Citations & Impact
Citations
18,366
H-index
63
i10-index
134
Publications
20
Co-authors
0
Publications
Resume
Academic Achievements
- Proposed 'Dualing GANs', reformulating the GAN min-max objective into a minimization problem using duality, revealing connections between GANs and moment matching.
- Developed 'Sliced Wasserstein GAN', removing the Kantorovich-Rubinstein duality via one-dimensional projections.
- Introduced 'Max-Sliced Distance' to reduce computational cost and improve training stability.
- Investigated challenges in backpropagating through GANs to latent space for image inpainting, proposing solutions using annealed importance sampling with Hamiltonian Monte Carlo.
- Created and released the SAILVOS dataset for amodal instance-level video object segmentation.
- Released open-source code for multiple projects, including Sliced Wasserstein GAN and backpropagation tools.