HEMP: High-order Entropy Minimization for neural network comPression
Formulates the entropy of a quantized neural network as a differentiable regularizer, trainable end-to-end for optimal entropy-coded compression.
Maintainers: Enzo Tartaglione
Open-source software and code releases from the Multimedia team — multimedia frameworks, compression toolkits, training code for our published methods.
We believe in reproducible research and open code. The packages below are maintained by members of the team and are available on GitHub (or the appropriate forge). To add a new entry, drop a Markdown file into this folder.
Formulates the entropy of a quantized neural network as a differentiable regularizer, trainable end-to-end for optimal entropy-coded compression.
Maintainers: Enzo Tartaglione
Graph rewiring algorithm addressing the trade-off between over-smoothing and over-squashing in deep graph neural networks.
Maintainers: Jhony H. Giraldo
Semi-supervised moving-object segmentation in video using graph signal processing, requiring far less labeled data than deep-learning baselines.
Maintainers: Jhony H. Giraldo
Proposed extension to SVG path syntax allowing reusable, ID-referenced path segments (including reversed copies), expanded to standard SVG 1.1 at load time.
Maintainers: Jean-Claude Moissinac
Patents and patent applications naming members of the Multimedia team as inventors.