8 papers
Rethinking Pixel Mean Flows via Interval Denoiser
Alexander Zaytsev, Dmitry Baranchuk, Alexander Korotin +1
Modern diffusion and flow-based models are increasingly moving toward few-step, latent-free generation to bypass the computational overhead of multi-step sampling and the reconstru…
Loss Barcode: A Topological Measure of Escapability in Loss Landscapes
Serguei Barannikov, Daria Voronkova, Alexander Mironenko +4
Neural network training is commonly based on SGD. However, the understanding of SGD's ability to converge to good local minima, given the non-convex nature of loss functions and th…
Uncovering Challenges of Solving the Continuous Gromov-Wasserstein Problem
Xavier Aramayo Carrasco, Maksim Nekrashevich, Petr Mokrov +2
Recently, the Gromov-Wasserstein Optimal Transport (GWOT) problem has attracted the special attention of the ML community. In this problem, given two distributions supported on two…
An Optimal Transport Perspective on Unpaired Image Super-Resolution
Milena Gazdieva, Petr Mokrov, Litu Rout +4
Real-world image super-resolution (SR) tasks often do not have paired datasets, which limits the application of supervised techniques. As a result, the tasks are usually approached…
Light Unbalanced Optimal Transport
Milena Gazdieva, Arip Asadulaev, Alexander Korotin +1
While the continuous Entropic Optimal Transport (EOT) field has been actively developing in recent years, it became evident that the classic EOT problem is prone to different issue…
Barcodes as Summary of Loss Function Topology
Serguei Barannikov, Alexander Korotin, Dmitry Oganesyan +2
We propose to study neural networks' loss surfaces by methods of topological data analysis. We suggest to apply barcodes of Morse complexes to explore topology of loss surfaces. An…