- title: 'Preface to Geometry-grounded Representation Learning and Generative Modeling (GRaM) Workshop' abstract: 'The Geometry-grounded Representation Learning and Generative Modeling (GRaM) workshop at ICLR 2024 explored the concept of geometric grounding. A representation, method, or theory is grounded in geometry if it can be amenable to geometric reasoning, that is, it abides by the mathematics of geometry. This idea plays a crucial role in developing generative models that understand geometry and can aid in geometric representations. We explored many different aspects of geometric representations at the GRaM Workshop.' volume: 251 URL: https://proceedings.mlr.press/v251/vadgama24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/vadgama24a/vadgama24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-vadgama24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 1-6 id: vadgama24a issued: date-parts: - 2024 - 10 - 12 firstpage: 1 lastpage: 6 published: 2024-10-12 00:00:00 +0000 - title: 'SE(3)-Hyena Operator for Scalable Equivariant Learning' abstract: 'Modeling global geometric context while maintaining equivariance is crucial for accurate predictions in many fields such as biology, chemistry, or vision. Yet, this is challenging due to the computational demands of processing high-dimensional data at scale. Existing approaches such as equivariant self-attention or distance-based message passing, suffer from quadratic complexity with respect to sequence length, while localized methods sacrifice global information. Inspired by the recent success of state-space and long-convolutional models, in this work, we introduce SE(3)-Hyena operator, an equivariant long-convolutional model based on the Hyena operator. The SE(3)-Hyena captures global geometric context at sub-quadratic complexity while maintaining equivariance to rotations and translations. Evaluated on equivariant associative recall and n-body modeling, SE(3)-Hyena matches or outperforms equivariant self-attention while requiring significantly less memory and computational resources for long sequences. Our model processes the geometric context of 20k tokens x3.5 times faster than the equivariant transformer and allows x175 longer a context within the same memory budget.' volume: 251 URL: https://proceedings.mlr.press/v251/moskalev24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/moskalev24a/moskalev24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-moskalev24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Artem family: Moskalev - given: Mangal family: Prakash - given: Rui family: Liao - given: Tommaso family: Mansi editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 7-19 id: moskalev24a issued: date-parts: - 2024 - 10 - 12 firstpage: 7 lastpage: 19 published: 2024-10-12 00:00:00 +0000 - title: 'Topology-Informed Graph Transformer' abstract: 'Transformers, through their self-attention mechanisms, have revolutionized performance in Natural Language Processing and Vision. Recently,there has been increasing interest in integrating Transformers with Graph Neural Networks (GNNs) to enhance analyzing geometric properties of graphs by employing global attention mechanisms. A key challenge in improving graph transformers is enhancing their ability to distinguish between isomorphic graphs, which can potentially boost their predictive performance. To address this challenge, we introduce ’Topology-Informed Graph Transformer (TIGT)’, a novel transformer enhancing both discriminative power in detecting graph isomorphisms and the overall performance of Graph Transformers. TIGT consists of four components: (1) a topological positional embedding layer using non-isomorphic universal covers based on cyclic subgraphs of graphs to ensure unique graph representation, (2) a dual-path message-passing layer to explicitly encode topological characteristics throughout the encoder layers, (3) a global attention mechanism, and (4) a graph information layer to recalibrate channel-wise graph features for improved feature representation. TIGT outperforms previous Graph Transformers in classifying synthetic dataset aimed at distinguishing isomorphism classes of graphs. Additionally, mathematical analysis and empirical evaluations highlight our model’s competitive edge over state-of-the-art Graph Transformers across various benchmark datasets.' volume: 251 URL: https://proceedings.mlr.press/v251/choi24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/choi24a/choi24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-choi24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Yun Young family: Choi - given: Sun Woo family: Park - given: Minho family: Lee - given: Youngho family: Woo editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 20-34 id: choi24a issued: date-parts: - 2024 - 10 - 12 firstpage: 20 lastpage: 34 published: 2024-10-12 00:00:00 +0000 - title: 'Alignment of MPNNs and Graph Transformers' abstract: 'As the complexity of machine learning (ML) model architectures increases, it is important to understand to what degree simpler and more efficient architectures can align with their complex counterparts. In this paper, we investigate the degree to which a Message Passing Neural Network (MPNN) can operate similarly to a Graph Transformer. We do this by training an MPNN to align with the intermediate embeddings of a Relational Transformer (RT). Throughout this process, we explore variations of the standard MPNN and assess the impact of different components on the degree of alignment. Our findings suggest that an MPNN can align to RT and the most important components that affect the alignment are the MPNN’s permutation invariant aggregation function, virtual node and layer normalisation.' volume: 251 URL: https://proceedings.mlr.press/v251/nguyen24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/nguyen24a/nguyen24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-nguyen24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Bao family: Nguyen - given: Anjana family: Yodaiken - given: Petar family: Veličković editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 35-49 id: nguyen24a issued: date-parts: - 2024 - 10 - 12 firstpage: 35 lastpage: 49 published: 2024-10-12 00:00:00 +0000 - title: 'Stability Analysis of Equivariant Convolutional Representations Through The Lens of Equivariant Multi-layered CKNs' abstract: 'In this paper we construct and theoretically analyse group equivariant convolutional kernel networks (CKNs) which are useful in understanding the geometry of (equivariant) CNNs through the lens of reproducing kernel Hilbert spaces (RKHSs). We then proceed to study the stability analysis of such equiv-CKNs under the action of diffeomorphism and draw a connection with equiv-CNNs, where the goal is to analyse the geometry of inductive biases of equiv-CNNs through the lens of reproducing kernel Hilbert spaces (RKHSs). Traditional deep learning architectures, including CNNs, trained with sophisticated optimization algorithms is vulnerable to perturbations, including ‘adversarial examples’. Understanding the RKHS norm of such models through CKNs is useful in designing the appropriate architecture and can be useful in designing robust equivariant representation learning models.' volume: 251 URL: https://proceedings.mlr.press/v251/chowdhury24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/chowdhury24a/chowdhury24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-chowdhury24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Soutrik Roy family: Chowdhury editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 50-64 id: chowdhury24a issued: date-parts: - 2024 - 10 - 12 firstpage: 50 lastpage: 64 published: 2024-10-12 00:00:00 +0000 - title: 'Asynchrony Invariance Loss Functions for Graph Neural Networks' abstract: 'A ubiquitous class of graph neural networks (GNNs) operates according to the message-passing paradigm, such that nodes systematically broadcast and listen to their neighbourhood. Yet, these synchronous computations have been deemed potentially sub-optimal as they could result in irrelevant information sent across the graph, thus interfering with efficient representation learning. In this work, we devise self-supervised loss functions biasing learning of synchronous GNN-based neural algorithmic reasoners towards representations that are invariant to asynchronous execution. Asynchrony invariance could successfully be learned, as revealed by analyses exploring the evolution of the self-supervised losses as well as their effect on the learned latent embeddings. Our approach to enforce asynchrony invariance constitutes a novel, potentially valuable tool for graph representation learning, which is increasingly prevalent in multiple real-world contexts.' volume: 251 URL: https://proceedings.mlr.press/v251/monteagudo-lago24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/monteagudo-lago24a/monteagudo-lago24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-monteagudo-lago24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Pablo family: Monteagudo-Lago - given: Arielle family: Rosinski - given: Andrew Joseph family: Dudzik - given: Petar family: Veličković editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 65-77 id: monteagudo-lago24a issued: date-parts: - 2024 - 10 - 12 firstpage: 65 lastpage: 77 published: 2024-10-12 00:00:00 +0000 - title: 'A Coding-Theoretic Analysis of Hyperspherical Prototypical Learning Geometry' abstract: 'Hyperspherical Prototypical Learning (HPL) is a supervised approach to representation learning that designs class prototypes on the unit hypersphere. The prototypes bias the representations to class separation in a scale invariant and known geometry. Previous approaches to HPL have either of the following shortcomings: (i) they follow an unprincipled optimisation procedure; or (ii) they are theoretically sound, but are constrained to only one possible latent dimension. In this paper, we address both shortcomings. To address (i), we present a principled optimisation procedure whose solution we show is optimal. To address (ii), we construct well-separated prototypes in a wide range of dimensions using linear block codes. Additionally, we give a full characterisation of the optimal prototype placement in terms of achievable and converse bounds, showing that our proposed methods are near-optimal.' volume: 251 URL: https://proceedings.mlr.press/v251/lindstrom24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/lindstrom24a/lindstrom24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-lindstrom24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Martin family: Lindström - given: Borja family: Rodríguez-Gálvez - given: Ragnar family: Thobaben - given: Mikael family: Skoglund editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 78-91 id: lindstrom24a issued: date-parts: - 2024 - 10 - 12 firstpage: 78 lastpage: 91 published: 2024-10-12 00:00:00 +0000 - title: '3D Shape Completion with Test-Time Training' abstract: 'This work addresses the problem of shape completion, i.e., the task of restoring incomplete shapes by predicting their missing parts. While previous works have often predicted the fractured and restored shape in one step, we approach the task by separately predicting the fractured and newly restored parts, but ensuring these predictions are interconnected. We use a decoder network motivated by related work on the prediction of signed distance functions (DeepSDF). In particular, our representation allows us to consider test-time-training, i.e., finetuning network parameters to match the given incomplete shape more accurately during inference. While previous works often have difficulties with artifacts around the fracture boundary, we demonstrate that our overfitting to the fractured parts leads to significant improvements in the restoration of eight different shape categories of the ShapeNet data set in terms of their chamfer distances.' volume: 251 URL: https://proceedings.mlr.press/v251/schopf-kuester24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/schopf-kuester24a/schopf-kuester24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-schopf-kuester24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Michael family: Schopf-Kuester - given: Zorah family: Lähner - given: Michael family: Moeller editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 92-102 id: schopf-kuester24a issued: date-parts: - 2024 - 10 - 12 firstpage: 92 lastpage: 102 published: 2024-10-12 00:00:00 +0000 - title: 'Commute-Time-Optimised Graphs for GNNs' abstract: 'We explore graph rewiring methods that optimise commute time. Recent graph rewiring approaches facilitate long-range interactions in sparse graphs, making such rewirings commute-time-optimal on average. However, when an expert prior exists on which node pairs should or should not interact, a superior rewiring would favour short commute times between these privileged node pairs. We construct two synthetic datasets with known priors reflecting realistic settings, and use these to motivate two bespoke rewiring methods that incorporate the known prior. We investigate the regimes where our rewiring improves test performance on the synthetic datasets. Finally, we perform a case study on a real-world citation graph to investigate the practical implications of our work.' volume: 251 URL: https://proceedings.mlr.press/v251/sterner24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/sterner24a/sterner24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-sterner24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Igor family: Sterner - given: Shiye family: Su - given: Petar family: Veličković editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 103-112 id: sterner24a issued: date-parts: - 2024 - 10 - 12 firstpage: 103 lastpage: 112 published: 2024-10-12 00:00:00 +0000 - title: 'Self-supervised detection of perfect and partial input-dependent symmetries' abstract: 'Group equivariance can overly constrain models if the symmetries in the group differ from those observed in data. While common methods address this by determining the appropriate level of symmetry at the dataset level, they are limited to supervised settings and ignore scenarios in which multiple levels of symmetry co-exist in the same dataset. In this paper, we propose a method able to detect the level of symmetry of each input without the need for labels. Our framework is general enough to accommodate different families of both continuous and discrete symmetry distributions, such as arbitrary unimodal, symmetric distributions and discrete groups. We validate the effectiveness of our approach on synthetic datasets with different per-class levels of symmetries, and demonstrate practical applications such as the detection of out-of-distribution symmetries.' volume: 251 URL: https://proceedings.mlr.press/v251/urbano24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/urbano24a/urbano24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-urbano24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Alonso family: Urbano - given: David W. family: Romero editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 113-131 id: urbano24a issued: date-parts: - 2024 - 10 - 12 firstpage: 113 lastpage: 131 published: 2024-10-12 00:00:00 +0000 - title: 'Metric Learning for Clifford Group Equivariant Neural Networks' abstract: 'Clifford Group Equivariant Neural Networks (CGENNs) leverage Clifford algebras and multivectors as an alternative approach to incorporating group equivariance to ensure symmetry constraints in neural representations. In principle, this formulation generalizes to orthogonal groups and preserves equivariance regardless of the metric signature. However, previous works have restricted internal network representations to Euclidean or Minkowski (pseudo-)metrics, handpicked depending on the problem at hand. In this work, we propose an alternative method that enables the metric to be learned in a data-driven fashion, allowing the CGENN network to learn more flexible representations. Specifically, we populate metric matrices fully, ensuring they are symmetric by construction, and leverage eigenvalue decomposition to integrate this additional learnable component into the original CGENN formulation in a principled manner. Additionally, we motivate our method using insights from category theory, which enables us to explain Clifford algebras as a categorical construction and guarantee the mathematical soundness of our approach. We validate our method in various tasks and showcase the advantages of learning more flexible latent metric representations. The code and data are available at \url{https://github.com/rick-ali/Metric-Learning-for-CGENNs}.' volume: 251 URL: https://proceedings.mlr.press/v251/ali24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/ali24a/ali24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-ali24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Riccardo family: Ali - given: Paulina family: Kulytė - given: Haitz prefix: Sáez de family: Ocáriz Borde - given: Pietro family: Lio editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 132-145 id: ali24a issued: date-parts: - 2024 - 10 - 12 firstpage: 132 lastpage: 145 published: 2024-10-12 00:00:00 +0000 - title: 'Dirac–Bianconi Graph Neural Networks – Enabling Non-Diffusive Long-Range Graph Predictions' abstract: 'The geometry of a graph is encoded in dynamical processes on the graph. Many graph neural network (GNN) architectures are inspired by such dynamical systems, typically based on the graph Laplacian. Here, we introduce Dirac–Bianconi GNNs (DBGNNs), which are based on the topological Dirac equation recently proposed by Bianconi. Based on the graph Laplacian, we demonstrate that DBGNNs explore the geometry of the graph in a fundamentally different way than conventional message passing neural networks (MPNNs). While regular MPNNs propagate features diffusively, analogous to the heat equation, DBGNNs allow for coherent long-range propagation. Experimental results showcase the superior performance of DBGNNs over existing conventional MPNNs for long-range predictions of power grid stability and peptide properties. This study highlights the effectiveness of DBGNNs in capturing intricate graph dynamics, providing notable advancements in GNN architectures.' volume: 251 URL: https://proceedings.mlr.press/v251/nauck24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/nauck24a/nauck24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-nauck24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Christian family: Nauck - given: Rohan family: Gorantla - given: Michael family: Lindne - given: Konstantin family: Schurholt - given: Antonia S. J. S. family: Mey - given: Frank family: Hellmann editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 146-157 id: nauck24a issued: date-parts: - 2024 - 10 - 12 firstpage: 146 lastpage: 157 published: 2024-10-12 00:00:00 +0000 - title: 'Leveraging Topological Guidance for Improved Knowledge Distillation' abstract: 'Deep learning has shown its efficacy in extracting useful features to solve various computer vision tasks. However, when the structure of the data is complex and noisy, capturing effective information to improve performance is very difficult. To this end, topological data analysis (TDA) has been utilized to derive useful representations that can contribute to improving performance and robustness against perturbations. Despite its effectiveness, the requirements for large computational resources and significant time consumption in extracting topological features through TDA are critical problems when implementing it on small devices. To address this issue, we propose a framework called Topological Guidance-based Knowledge Distillation (TGD), which uses topological features in knowledge distillation (KD) for image classification tasks. We utilize KD to train a superior lightweight model and provide topological features with multiple teachers simultaneously. We introduce a mechanism for integrating features from different teachers and reducing the knowledge gap between teachers and the student, which aids in improving performance. We demonstrate the effectiveness of our approach through diverse empirical evaluations.' volume: 251 URL: https://proceedings.mlr.press/v251/jeon24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/jeon24a/jeon24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-jeon24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Eun Som family: Jeon - given: Rahul family: Khurana - given: Aishani family: Pathak - given: Pavan family: Turaga editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 158-172 id: jeon24a issued: date-parts: - 2024 - 10 - 12 firstpage: 158 lastpage: 172 published: 2024-10-12 00:00:00 +0000 - title: 'E(n) Equivariant Message Passing Cellular Networks' abstract: 'This paper introduces E(n) Equivariant Message Passing Cellular Networks (EMPCNs), an extension of E(n) Equivariant Graph Neural Networks to CW-complexes. Our approach addresses two aspects of geometric message passing networks: 1) enhancing their expressiveness by incorporating arbitrary cells, and 2) achieving this in a computationally efficient way with a decoupled EMPCNs technique. We demonstrate that EMPCNs achieve close to state-of-the-art performance on multiple tasks without the need for steerability, including many-body predictions and motion capture. Moreover, ablation studies confirm that decoupled EMPCNs exhibit stronger generalization capabilities than their non-topologically informed counterparts. These findings show that EMPCNs can be used as a scalable and expressive framework for higher-order message passing in geometric and topological graphs' volume: 251 URL: https://proceedings.mlr.press/v251/kovac-24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/kovac-24a/kovac-24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-kovac-24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Veljko family: Kovac̆ - given: Erik family: Bekkers - given: Pietro family: Lió - given: Floor family: Eijkelboom editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 173-186 id: kovac-24a issued: date-parts: - 2024 - 10 - 12 firstpage: 173 lastpage: 186 published: 2024-10-12 00:00:00 +0000 - title: 'A Simple and Expressive Graph Neural Network Based Method for Structural Link Representation' abstract: 'Graph Neural Networks (GNNs) have achieved state-of-the-art results in tasks like node classification, link prediction, and graph classification. While much research has focused on their ability to distinguish graphs, fewer studies have addressed their capacity to differentiate links, a complex and less explored area. This paper introduces SLRGNN, a novel, theoretically grounded GNN-based method for link prediction. SLRGNN ensures that link representations are distinct if and only if the links have different structural roles within the graph. Our approach transforms the link prediction problem into a node classification problem on the corresponding line graph, enhancing expressiveness without sacrificing efficiency. Unlike existing methods, SLRGNN computes link probabilities in a single inference step, avoiding the need for individual subgraph constructions. We provide a formal proof of our method’s expressiveness and validate its superior performance through experiments on real-world datasets. The code is publicly available1' volume: 251 URL: https://proceedings.mlr.press/v251/lachi24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/lachi24a/lachi24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-lachi24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Veronica family: Lachi - given: Francesco family: Ferrini - given: Antonio family: Longa - given: Bruno family: Lepri - given: Andrea family: Passerini editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 187-201 id: lachi24a issued: date-parts: - 2024 - 10 - 12 firstpage: 187 lastpage: 201 published: 2024-10-12 00:00:00 +0000 - title: 'Invertible Temper Modeling using Normalizing Flows and the Effects of Structure Preserving Loss' abstract: 'Advanced manufacturing research and development is typically small-scale, owing to costly experiments associated with these novel processes. Deep learning techniques could help accelerate this development cycle but frequently struggle in small-data regimes like the advanced manufacturing space. While prior work has applied deep learning to modeling visually plausible advanced manufacturing microstructures, little work has been done on data-driven modeling of how microstructures are affected by heat treatment, or assessing the degree to which synthetic microstructures are able to support existing workflows. We propose to address this gap by using invertible neural networks (normalizing flows) to model the effects of heat treatment, e.g., tempering. The model is developed using scanning electron microscope imagery from samples produced using shear-assisted processing and extrusion (ShAPE) manufacturing. This approach not only produces visually and topologically plausible samples, but also captures information related to a sample’s material properties or experimental process parameters. We also demonstrate that topological data analysis, used in prior work to characterize microstructures, can also be used to stabilize model training, preserve structure, and improve downstream results. We assess directions for future work and identify our approach as an important step towards an end-to-end deep learning system for accelerating advanced manufacturing research and development.' volume: 251 URL: https://proceedings.mlr.press/v251/howland24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/howland24a/howland24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-howland24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Sylvia family: Howland - given: Keerti-Sahithi family: Kappagantula - given: Henry family: Kvinge - given: Tegan family: Emerson editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 202-211 id: howland24a issued: date-parts: - 2024 - 10 - 12 firstpage: 202 lastpage: 211 published: 2024-10-12 00:00:00 +0000 - title: 'Topological and Dynamical Representations for Radio Frequency Signal Classification' abstract: 'Radio Frequency (RF) signals are found throughout our world, carrying over-the-air information for both digital and analog uses with applications ranging from WiFi to the radio. One area of focus in RF signal analysis is determining the modulation schemes employed in these signals which is crucial in many RF signal processing domains from secure communication to spectrum monitoring. This work investigates the accuracy and noise robustness of novel Topological Data Analysis (TDA) and dynamic representation based approaches paired with a small convolution neural network for RF signal modulation classification with a comparison to state-of-the-art deep neural network approaches. We show that using TDA tools, like Vietoris-Rips and lower star filtration, and the Takens’ embedding in conjunction with a standard shallow neural network we can capture the intrinsic dynamical, geometric, and topological features of the underlying signal’s manifold, informative representations of the RF signals. Our approach is effective in handling the modulation classification task and is notably noise robust, outperforming the commonly used deep neural network approaches in mode classification. Moreover, our fusion of dynamical and topological information is able to attain similar performance to deep neural network architectures with significantly smaller training datasets.' volume: 251 URL: https://proceedings.mlr.press/v251/meyers24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/meyers24a/meyers24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-meyers24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Audum family: Meyers - given: Timothy family: Doster - given: Colin family: Olson - given: Tegan family: Emerson editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 212-221 id: meyers24a issued: date-parts: - 2024 - 10 - 12 firstpage: 212 lastpage: 221 published: 2024-10-12 00:00:00 +0000 - title: 'SCENE-Net V2: Interpretable Multiclass 3D Scene Understanding with Geometric Priors' abstract: 'In this paper, we present SCENE-Net V2, a new resource-efficient, gray-box model for multiclass 3D scene understanding. SCENE-Net V2 leverages Group Equivariant Non-Expansive Operators (GENEOs) to incorporate fundamental geometric priors as inductive biases, offering a more transparent alternative to the prevalent black-box models in the domain. This model addresses the limitations of its white-box predecessor, SCENE-Net, by expanding its applicability from pole-like structures to a wider range of datasets with detailed 3D elements. Our model achieves the sweet-spot between application and transparency: SCENE-Net V2 is a general method for object identification with interpretability guarantees. Our experimental results demonstrate that SCENE-Net V2 achieves competitive performance with a significantly lower parameter count. Furthermore, we propose the use of GENEO-based architectures as a feature extraction tool for black-box models, enabling an increase in performance by adding a minimal number of meaningful parameters. Our code is available in: https://github.com/dlavado/SCENE-Net-V2.' volume: 251 URL: https://proceedings.mlr.press/v251/lavado24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/lavado24a/lavado24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-lavado24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Diogo family: Lavado - given: Cláudia family: Soares - given: Alessandra family: Micheletti editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 222-232 id: lavado24a issued: date-parts: - 2024 - 10 - 12 firstpage: 222 lastpage: 232 published: 2024-10-12 00:00:00 +0000 - title: 'Variational Inference Failures Under Model Symmetries: Permutation Invariant Posteriors for Bayesian Neural Networks' abstract: 'Weight space symmetries in neural network architectures, such as permutation symmetries in MLPs, give rise to Bayesian neural network (BNN) posteriors with many equivalent modes. This multimodality poses a challenge for variational inference (VI) techniques, which typically rely on approximating the posterior with a unimodal distribution. In this work, we investigate the impact of weight space permutation symmetries on VI. We demonstrate, both theoretically and empirically, that these symmetries lead to biases in the approximate posterior, which degrade predictive performance and posterior fit if not explicitly accounted for. To mitigate this behavior, we leverage the symmetric structure of the posterior and devise a symmetrization mechanism for constructing permutation invariant variational posteriors. We show that the symmetrized distribution has a strictly better fit to the true posterior, and that it can be trained using the original ELBO objective with a modified KL regularization term. We demonstrate experimentally that our approach mitigates the aforementioned biases and results in improved predictions and a higher ELBO.' volume: 251 URL: https://proceedings.mlr.press/v251/gelberg24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/gelberg24a/gelberg24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-gelberg24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Yoav family: Gelberg - given: Tycho F. A. prefix: van der family: Ouderaa - given: Mark prefix: van der family: Wilk - given: Yarin family: Gal editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 233-248 id: gelberg24a issued: date-parts: - 2024 - 10 - 12 firstpage: 233 lastpage: 248 published: 2024-10-12 00:00:00 +0000 - title: 'Joint Diffusion Processes as an Inductive Bias in Sheaf Neural Networks' abstract: 'Sheaf Neural Networks (SNNs) naturally extend Graph Neural Networks (GNNs) by endowing a cellular sheaf over the graph, equipping nodes and edges with vector spaces and defining linear mappings between them. While the attached geometric structure has proven to be useful in analyzing heterophily and oversmoothing, so far the methods by which the sheaf is computed do not always guarantee a good performance in such settings. In this work, drawing inspiration from opinion dynamics concepts, we propose two novel sheaf learning approaches that (i) provide a more intuitive understanding of the involved structure maps, (ii) introduce a useful inductive bias for heterophily and oversmoothing, and (iii) infer the sheaf in a way that does not scale with the number of features, thus using fewer learnable parameters than existing methods. In our evaluation, we show the limitations of the real-world benchmarks used so far on SNNs, and design a new synthetic task –leveraging the symmetries of $n$-dimensional ellipsoids– that enables us to better assess the strengths and weaknesses of sheaf-based models. Our extensive experimentation on these novel datasets reveals valuable insights into the scenarios and contexts where basic SNNs and our proposed approaches can be beneficial.' volume: 251 URL: https://proceedings.mlr.press/v251/hernandez-caralt24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/hernandez-caralt24a/hernandez-caralt24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-hernandez-caralt24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Ferran family: Hernandez Caralt - given: Guillermo family: Bernárdez Gil - given: Iulia family: Duta - given: Pietro family: Liò - given: Eduard family: Alarcón Cot editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 249-263 id: hernandez-caralt24a issued: date-parts: - 2024 - 10 - 12 firstpage: 249 lastpage: 263 published: 2024-10-12 00:00:00 +0000 - title: 'Sheaf Diffusion Goes Nonlinear: Enhancing GNNs with Adaptive Sheaf Laplacians' abstract: 'Sheaf Neural Networks (SNNs) have recently been introduced to enhance Graph Neural Networks (GNNs) in their capability to learn from graphs. Previous studies either focus on linear sheaf Laplacians or hand-crafted nonlinear sheaf Laplacians. The former are not always expressive enough in modeling complex interactions between nodes, such as antagonistic dynamics and bounded confidence dynamics, while the latter use a fixed nonlinear function that is not adapted to the data at hand. To enhance the capability of SNNs to capture complex node-to-node interactions while adapting to different scenarios, we propose a Nonlinear Sheaf Diffusion (NLSD) model, which incorporates nonlinearity into the Laplacian of SNNs through a general function learned from data. Our model is validated on a synthetic community detection dataset, where it outperforms linear SNNs and common GNN baselines in a node classification task, showcasing its ability to leverage complex network dynamics.' volume: 251 URL: https://proceedings.mlr.press/v251/zaghen24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/zaghen24a/zaghen24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-zaghen24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Olga family: Zaghen - given: Antonio family: Longa - given: Steve family: Azzolin - given: Lev family: Telyatnikov - given: Andrea family: Passerini - given: Pietro family: Liò editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 264-276 id: zaghen24a issued: date-parts: - 2024 - 10 - 12 firstpage: 264 lastpage: 276 published: 2024-10-12 00:00:00 +0000 - title: 'Decoder ensembling for learned latent geometries' abstract: 'Latent space geometry provides a rigorous and empirically valuable framework for interacting with the latent variables of deep generative models. This approach reinterprets Euclidean latent spaces as Riemannian through a pull-back metric, allowing for a standard differential geometric analysis of the latent space. Unfortunately, data manifolds are generally compact and easily disconnected or filled with holes, suggesting a topological mismatch to the Euclidean latent space. The most established solution to this mismatch is to let uncertainty be a proxy for topology, but in neural network models, this is often realized through crude heuristics that lack principle and generally do not scale to high-dimensional representations. We propose using ensembles of decoders to capture model uncertainty and show how to easily compute geodesics on the associated expected manifold. Empirically, we find this simple and reliable, thereby coming one step closer to easy-to-use latent geometries.' volume: 251 URL: https://proceedings.mlr.press/v251/syrota24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/syrota24a/syrota24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-syrota24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Stas family: Syrota - given: Pablo family: Moreno-Muñoz - given: Søren family: Hauberg editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 277-285 id: syrota24a issued: date-parts: - 2024 - 10 - 12 firstpage: 277 lastpage: 285 published: 2024-10-12 00:00:00 +0000 - title: 'The NGT200 Dataset: Geometric Multi-View Isolated Sign Recognition' abstract: 'Sign Language Processing (SLP) provides a foundation for a more inclusive future in language technology; however, the field faces several significant challenges that must be addressed to achieve practical, real-world applications. This work addresses multi-view isolated sign recognition (MV-ISR), and highlights the essential role of 3D awareness and geometry in SLP systems. We introduce the NGT200 dataset, a novel spatio-temporal multi-view benchmark, establishing MV-ISR as distinct from single-view ISR (SV-ISR). We demonstrate the benefits of synthetic data and propose conditioning sign representations on spatial symmetries inherent in sign language. Leveraging an SE(2) equivariant model improves MV-ISR performance by 8-22 percent over the baseline.' volume: 251 URL: https://proceedings.mlr.press/v251/ranum24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/ranum24a/ranum24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-ranum24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Oline family: Ranum - given: David R. family: Wessels - given: Gomer family: Otterspeer - given: Erik J. family: Bekkers - given: Floris family: Roelofsen - given: Jari I. family: Andersen editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 286-302 id: ranum24a issued: date-parts: - 2024 - 10 - 12 firstpage: 286 lastpage: 302 published: 2024-10-12 00:00:00 +0000 - title: 'On Fairly Comparing Group Equivariant Networks' abstract: 'This paper investigates the flexibility of Group Equivariant Convolutional Neural Networks (G-CNNs), which specialize conventional neural networks by encoding equivariance to group transformations. Inspired by splines, we propose new metrics to assess the complexity of ReLU networks and use them to quantify and compare the flexibility of networks equivariant to different groups. Our analysis suggests that the current practice of comparing networks by fixing the number of trainable parameters unfairly affords models equivariant to larger groups additional expressivity. Instead, we advocate for comparisons based on a fixed computational budget—which we empirically show results in more similar levels of network flexibility. This approach allows one to better disentangle the impact of constraining networks to be equivariant from the increased expressivity they are typically granted in the literature, enabling one to obtain a more nuanced view of the impact of enforcing equivariance. Interestingly, our experiments indicate that enforcing equivariance results in more complex fitted functions even when controlling for compute, despite reducing network expressivity.' volume: 251 URL: https://proceedings.mlr.press/v251/roos24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/roos24a/roos24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-roos24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Lucas family: Roos - given: Steve family: Kroon editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 303-317 id: roos24a issued: date-parts: - 2024 - 10 - 12 firstpage: 303 lastpage: 317 published: 2024-10-12 00:00:00 +0000 - title: 'Graph Convolutional Networks for Learning Laplace-Beltrami Operators' abstract: 'Recovering a high-level representation of geometric data is a fundamental goal in geometric modeling and computer graphics. In this paper, we introduce a data-driven approach to computing the spectrum of the Laplace-Beltrami operator of triangle meshes using graph convolutional networks. Specifically, we train graph convolutional networks on a large-scale dataset of synthetically generated triangle meshes, encoded with geometric data consisting of Voronoi areas, normalized edge lengths, and the Gauss map, to infer eigenvalues of 3D shapes. We attempt to address the ability of graph neural networks to capture global shape descriptors–including spectral information–that were previously inaccessible using existing methods from computer vision, and our paper exhibits promising signals suggesting that Laplace-Beltrami eigenvalues on discrete surfaces can be learned. Additionally, we perform ablation studies showing the addition of geometric data leads to improved accuracy.' volume: 251 URL: https://proceedings.mlr.press/v251/wu24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/wu24a/wu24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-wu24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Yingying family: Wu - given: Roger family: Fu - given: Yang family: Peng - given: Qifeng family: Chen editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 318-331 id: wu24a issued: date-parts: - 2024 - 10 - 12 firstpage: 318 lastpage: 331 published: 2024-10-12 00:00:00 +0000 - title: 'The Geometry of Diffusion Models: Tubular Neighbourhoods and Singularities' abstract: 'Diffusion generative models have been a leading approach for generating high-dimensional data. The current research aims to investigate the relation between the dynamics of diffusion models and the tubular neighbourhoods of a data manifold. We propose an algorithm to estimate the injectivity radius, the supremum of radii of tubular neighbourhoods. Our research relates geometric objects such as curvatures of data manifolds and dimensions of ambient spaces, to singularities of the generative dynamics such as emergent critical phenomena or spontaneous symmetry breaking.' volume: 251 URL: https://proceedings.mlr.press/v251/sakamoto24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/sakamoto24a/sakamoto24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-sakamoto24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Kotaro family: Sakamoto - given: Ryosuke family: Sakamoto - given: Masato family: Tanabe - given: Masatomo family: Akagawa - given: Yusuke family: Hayashi - given: Manato family: Yaguchi - given: Masahiro family: Suzuki - given: Yutaka family: Matsuo editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 332-363 id: sakamoto24a issued: date-parts: - 2024 - 10 - 12 firstpage: 332 lastpage: 363 published: 2024-10-12 00:00:00 +0000 - title: 'Equivariant vs. Invariant Layers: A Comparison of Backbone and Pooling for Point Cloud Classification' abstract: 'Learning from set-structured data, such as point clouds, has gained significant attention from the machine learning community. Geometric deep learning provides a blueprint for designing effective set neural networks that preserve the permutation symmetry of set-structured data. Of our interest are permutation invariant networks, which are composed of a permutation equivariant backbone, permutation invariant global pooling, and regression/classification head. While existing literature has focused on improving equivariant backbones, the impact of the pooling layer is often overlooked. In this paper, we examine the interplay between permutation equivariant backbones and permutation invariant global pooling on three benchmark point cloud classification datasets. Our findings reveal that: 1) complex pooling methods, such as transport-based or attention-based poolings, can significantly boost the performance of simple backbones, but the benefits diminish for more complex backbones, 2) even complex backbones can benefit from pooling layers in low data scenarios, 3) surprisingly, the choice of pooling layers can have a more significant impact on the model’s performance than adjusting the width and depth of the backbone, and 4) pairwise combination of pooling layers can significantly improve the performance of a fixed backbone. Our comprehensive study provides insights for practitioners to design better permutation invariant set neural networks. Our code is available at https://github.com/mint-vu/backbone_vs_pooling.' volume: 251 URL: https://proceedings.mlr.press/v251/kothapalli24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/kothapalli24a/kothapalli24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-kothapalli24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Abihith family: Kothapalli - given: Ashkan family: Shahbazi - given: Xinran family: Liu - given: Robert family: Sheng - given: Soheil family: Kolouri editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 364-380 id: kothapalli24a issued: date-parts: - 2024 - 10 - 12 firstpage: 364 lastpage: 380 published: 2024-10-12 00:00:00 +0000 - title: 'Strongly Isomorphic Neural Optimal Transport Across Incomparable Spaces' abstract: 'Optimal Transport (OT) has recently emerged as a powerful framework for learning minimal-displacement maps between distributions. The predominant approach involves a neural parametrization of the Monge formulation of OT, typically assuming the same space for both distributions. However, the setting across “incomparable spaces” (e.g., of different dimensionality), corresponding to the Gromov-Wasserstein distance, remains underexplored, with existing methods often imposing restrictive assumptions on the cost function. In this paper, we present a novel neural formulation of the Gromov-Monge (GM) problem rooted in one of its fundamental properties: invariance to strong isomorphisms. We operationalize this property by decomposing the learnable OT map into two components: (i) an approximate strong isomorphism between the source distribution and an intermediate reference distribution, and (ii) a GM-optimal map between this reference and the target distribution. Our formulation leverages and extends the Monge gap regularizer of \citet{gap_monge} to eliminate the need for complex architectural requirements of other neural OT methods, yielding a simple but practical method that enjoys favorable theoretical guarantees. Our preliminary empirical results show that our framework provides a promising approach to learn OT maps across diverse spaces.' volume: 251 URL: https://proceedings.mlr.press/v251/sotiropoulou24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/sotiropoulou24a/sotiropoulou24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-sotiropoulou24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Athina family: Sotiropoulou - given: David family: Alvarez-Melis editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 381-393 id: sotiropoulou24a issued: date-parts: - 2024 - 10 - 12 firstpage: 381 lastpage: 393 published: 2024-10-12 00:00:00 +0000 - title: 'Adaptive Sampling for Continuous Group Equivariant Neural Networks' abstract: 'Steerable networks, which process data with intrinsic symmetries, often use Fourier-based non-linearities that require sampling from the entire group, leading to a need for discretization in continuous groups. As the number of samples increases, both performance and equivariance improve, yet this also leads to higher computational costs. To address this, we introduce an adaptive sampling approach that dynamically adjusts the sampling process to the symmetries in the data, reducing the number of required group samples and lowering the computational demands. We explore various implementations and their effects on model performance, equivariance, and computational efficiency. Our findings demonstrate improved model performance, and a marginal increase in memory efficiency' volume: 251 URL: https://proceedings.mlr.press/v251/inal24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/inal24a/inal24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-inal24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Berfin family: Inal - given: Gabriele family: Cesa editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 394-419 id: inal24a issued: date-parts: - 2024 - 10 - 12 firstpage: 394 lastpage: 419 published: 2024-10-12 00:00:00 +0000 - title: 'ICML Topological Deep Learning Challenge 2024: Beyond the Graph Domain' abstract: 'This paper describes the 2nd edition of the ICML Topological Deep Learning Challenge that was hosted within the ICML 2024 ELLIS Workshop on Geometry-grounded Representation Learning and Generative Modeling (GRaM). The challenge focused on the problem of representing data in different discrete topological domains in order to bridge the gap between Topological Deep Learning (TDL) and other types of structured datasets (e.g. point clouds, graphs). Specifically, participants were asked to design and implement topological liftings, i.e. mappings between different data structures and topological domains –like hypergraphs, or simplicial/cell/combinatorial complexes. The challenge received 52 submissions satisfying all the requirements. This paper introduces the main scope of the challenge, and summarizes the main results and findings.' volume: 251 URL: https://proceedings.mlr.press/v251/bernardez24a.html PDF: https://raw.githubusercontent.com/mlresearch/v251/main/assets/bernardez24a/bernardez24a.pdf edit: https://github.com/mlresearch//v251/edit/gh-pages/_posts/2024-10-12-bernardez24a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM)' publisher: 'PMLR' author: - given: Guillermo family: Bernárdez - given: Lev family: Telyatnikov - given: Marco family: Montagna - given: Federica family: Baccini - given: Mathilde family: Papillon - given: Miquel family: Ferriol-Galmés - given: Mustafa family: Hajij - given: Theodore family: Papamarkou - given: Maria Sofia family: Bucarelli - given: Olga family: Zaghen - given: Johan family: Mathe - given: Audun family: Myers - given: Scott family: Mahan - given: Hansen family: Lillemark - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Tim family: Doster - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Katrina family: Agate - given: Nesreen K family: Ahmed - given: Pengfei family: Bai - given: Michael family: Banf - given: Claudio family: Battiloro - given: Maxim family: Beketov - given: Paul family: Bogdan - given: Martin family: Carrasco - given: Andrea family: Cavallo - given: Yun Young family: Choi - given: George family: Dasoulas - given: Matous̆ family: Elphick - given: Giordan family: Escalona - given: Dominik family: Filipiak - given: Halley family: Fritze - given: Thomas family: Gebhart - given: Manel family: Gil-Sorribes - given: Salvish family: Goomanee - given: Victor family: Guallar - given: Liliya family: Imasheva - given: Andrei family: Irimia - given: Hongwei family: Jin - given: Graham family: Johnson - given: Nikos family: Kanakaris - given: Boshko family: Koloski - given: Veljko family: Kovac̆ - given: Manuel family: Lecha - given: Minho family: Lee - given: Pierrick family: Leroy - given: Theodore family: Long - given: German family: Magai - given: Alvaro family: Martinez - given: Marissa family: Masden - given: Sebastian family: Mez̆nar - given: Bertran family: Miquel-Oliver - given: Alexis family: Molina - given: Alexander family: Nikitin - given: Marco family: Nurisso - given: Matt family: Piekenbrock - given: Yu family: Qin - given: Patryk family: Rygiel - given: Alessandro family: Salatiello - given: Max family: Schattauer - given: Pavel family: Snopov - given: Julian family: Suk - given: Valentina family: Sánchez - given: Mauricio family: Tec - given: Francesco family: Vaccarino - given: Jonas family: Verhellen - given: Frederic family: Wantiez - given: Alexander family: Weers - given: Patrik family: Zajec - given: Blaz̆ family: S̆krlj - given: Nina family: Miolane editor: - given: Sharvaree family: Vadgama - given: Erik family: Bekkers - given: Alison family: Pouplin - given: Sekou-Oumar family: Kaba - given: Robin family: Walters - given: Hannah family: Lawrence - given: Tegan family: Emerson - given: Henry family: Kvinge - given: Jakub family: Tomczak - given: Stephanie family: Jegelka page: 420-428 id: bernardez24a issued: date-parts: - 2024 - 10 - 12 firstpage: 420 lastpage: 428 published: 2024-10-12 00:00:00 +0000