Mimetic Neural Networks: A unified framework for Protein Design and Folding

Eliasof, Moshe; Boesen, Tue; Haber, Eldad; Keasar, Chen; Treister, Eran

doi:10.48550/arxiv.2102.03881

Cited by 2 publications

(4 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…( 1). On the other hand, applications that require conservation such as volume/distance preservation as in the dense shape correspondence task [41] and protein folding [10], are typically better treated using a hyperbolic equation as in Eq. (2).…”

Section: Partial Differential Equations On Manifoldsmentioning

confidence: 99%

“…2 suggest that just as in classical works [39,48], problems like nodeclassification obtain better performance with an anisotropic diffusion like in Eq. ( 6), and for problems involving dense-correspondences like in [41,10] that tend to conserve the energy of the underlying problem, a hyperbolic equation type of PDE as in Eq. ( 7) is more appropriate.…”

Section: Learning Pde Network Dynamicsmentioning

confidence: 99%

“…In recent years, Graph Convolutional Networks (GCNs) [1,2,3] have drawn the attention of researchers and practitioners in a variety of domains and applications, ranging from computer vision and graphics [4,5,6,7] to computational biology [8,9,10], recommendation systems [11] and social network analysis [12,13]. However, GCNs still suffer from two main problems.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

PDE-GCN: Novel Architectures for Graph Neural Networks Motivated by Partial Differential Equations

Eliasof¹,

Haber²,

Treister³

2021

Preprint

Self Cite

View full text Add to dashboard Cite

Graph neural networks are increasingly becoming the go-to approach in various fields such as computer vision, computational biology and chemistry, where data are naturally explained by graphs. However, unlike traditional convolutional neural networks, deep graph networks do not necessarily yield better performance than shallow graph networks. This behavior usually stems from the over-smoothing phenomenon. In this work, we propose a family of architectures to control this behavior by design. Our networks are motivated by numerical methods for solving Partial Differential Equations (PDEs) on manifolds, and as such, their behavior can be explained by similar analysis. Moreover, as we demonstrate using an extensive set of experiments, our PDE-motivated networks can generalize and be effective for various types of problems from different fields. Our architectures obtain better or on par with the current state-of-the-art results for problems that are typically approached using different architectures.Preprint. Under review.

show abstract

Section: Partial Differential Equations On Manifoldsmentioning

confidence: 99%

Section: Learning Pde Network Dynamicsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

PDE-GCN: Novel Architectures for Graph Neural Networks Motivated by Partial Differential Equations

Eliasof¹,

Haber²,

Treister³

2021

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…G RAPH convolutional networks (GCNs) have been shown to be highly successful when applied to a wide array of problems and domains, including social analysis [1], [2], recommendation systems [3], computational biology [4] and computer vision and graphics [5]- [7]. Conceptually, GCNs are an unstructured version of standard convolutional neural networks (CNNs) where instead of a 2D or a 3D grid, we have an unstructured graph or a mesh [7].…”

Section: Introductionmentioning

confidence: 99%

Haar Wavelet Feature Compression for Quantized Graph Convolutional Networks

Eliasof¹,

Bodner²,

Treister³

2021

Preprint

Self Cite

View full text Add to dashboard Cite

Graph Convolutional Networks (GCNs) are widely used in a variety of applications, and can be seen as an unstructured version of standard Convolutional Neural Networks (CNNs). As in CNNs, the computational cost of GCNs for large input graphs (such as large point clouds or meshes) can be high and inhibit the use of these networks, especially in environments with low computational resources. To ease these costs, quantization can be applied to GCNs. However, aggressive quantization of the feature maps can lead to a significant degradation in performance. On a different note, Haar wavelet transforms are known to be one of the most effective and efficient approaches to compress signals. Therefore, instead of applying aggressive quantization to feature maps, we propose to utilize Haar wavelet compression and light quantization to reduce the computations and the bandwidth involved with the network. We demonstrate that this approach surpasses aggressive feature quantization by a significant margin, for a variety of problems ranging from node classification to point cloud classification and part and semantic segmentation.

show abstract

Mimetic Neural Networks: A unified framework for Protein Design and Folding

Cited by 2 publications

References 32 publications

PDE-GCN: Novel Architectures for Graph Neural Networks Motivated by Partial Differential Equations

PDE-GCN: Novel Architectures for Graph Neural Networks Motivated by Partial Differential Equations

Haar Wavelet Feature Compression for Quantized Graph Convolutional Networks

Contact Info

Product

Resources

About