GeometricFlux.jl
GeometricFlux.jl copied to clipboard
FluxML
Geometric Deep Learning for Flux
> Ref. [Multi-directional geodesic neural networks via equivariant convolution](https://dl.acm.org/doi/abs/10.1145/3272127.3275102)
> Ref. [Gauge Equivariant Convolutional Networks and the Icosahedral CNN](http://proceedings.mlr.press/v97/cohen19d.html)
> [QUVA-Lab/escnn](https://github.com/QUVA-Lab/escnn) > Ref. [A Program to Build E(N)-Equivariant Steerable CNNs](https://openreview.net/forum?id=WE4qe9xlnQw)
> [e3nn/e3nn](https://github.com/e3nn/e3nn) > Ref. [Euclidean Neural Networks](https://arxiv.org/abs/2207.09453)
- [ ] ` SteerableConv` layer - [ ] steerable CNN example > Ref. [Steerable CNNs](https://arxiv.org/abs/1612.08498)
Attempt to fix #292. Other examples and docs may need to follow suite.
runs in GPU now, reshape() function didn't work with ODE output, so I just get first element of the array. The training is stuck at accuracy of 0.24, not sure...
- [ ] Graph Matching Networks > Ref. [Graph Matching Networks for Learning the Similarity of Graph Structured Objects](https://arxiv.org/abs/1904.12787)
> Ref. [Pure Transformers are Powerful Graph Learners](https://arxiv.org/abs/2207.02505)
> Ref. [GraphWorld: Fake Graphs Bring Real Insights for GNNs](https://arxiv.org/abs/2203.00112), [GraphWorld: Advances in Graph Benchmarking](http://ai.googleblog.com/2022/05/graphworld-advances-in-graph.html)