Data arising in scientific applications often exhibits complex nonlinear structure. Identifying and parameterizing said nonlinearities can enable better modeling, enhance predictive analysis, data interpolation, denoising, and anomaly detection. In this paper, we develop theoretical and computational methods to detect and parametrize Euclidean point cloud data lying near a wedge sum of tori. Specifically, we use persistent cohomology and cup products to identify the distinct toroidal components, and on each component, we optimize over low-energy multivariate trigonometric polynomials for best data fit. The ensuing optimization problem reduces to generalized ridge regression, which can be solved efficiently via standard techniques. We demonstrate the utility of the proposed framework in both synthetic and real datasets, including interpolation of low frame-rate recurrent videos, attractor recovery in quasiperiodic dynamical systems, and the parametrization of gene-space cycle data for parasites and humans.
Figures & tables
Figure 1. ToroFit Pipeline : (a) Perform topological inference on the input data by computing persistent cohomology, identifying an appropriate scale α and computing cocycles η∙ along with their cup product partitions P∙ . (b) Each cocycle identified during topological inference provides a circular coordinate hj , and together all the persistent cohomology classes provide a toroidal coordinate h . (c) Each cocycle partition Pj corresponds to a toroidal coordinate hPj ; filtering out the low-variation points of hPj identifies a stratifying torus Sj with dimension dj=∣Pj∣ . Finally, we fit a multivariate Fourier series to each Sj to obtain a parameterization Fj . We assemble the fits into the final map F:⋁j=1JTdj→CN .
Figure 2. The fits for the clover leaf dataset with different regularizer values λ=0.1,1.0,10.0 are plotted. Higher values of the regularizer extract a smoother curve.
Figure 3. Clover Leaf dataset (gray) with persistence diagram. The most persistent class (circled) is used to compute a circular coordinate map and the corresponding fit.
Figure 4. Pairwise cup products of harmonic cocycles. In this case, {η2,η3} form one connected component in this graph, indicating that our dataset has a T2 component. Similarly, {η1} represents the S1 component.
Figure 5. Toroidal Fitting Procedure on a wedge of two spaces. First, we partition the dataset into two distinct components and then fit an appropriate dimensional torus to each of the components separately.
Figure 6. Toroidal fitting pipeline on the bunny point cloud. (a) Synthetic dataset composed of a sphere and two circles wedged at a point. The persistence diagram shows 2 highly persistent classes in H1 , namely η1 and η2 . We also see a high persistence H2 class (green) corresponding to the sphere. (b) Cup-product table for the cohomology classes η1 and η2 . We see that the cup product η1⌣η2 is zero, indicating that our dataset consists of two individual circles, S1∨S1 , as opposed to a 2-torus T2 .
Figure 7. ToroFit on Bunny Point Cloud. The two fitted maps correctly fit to the “ears” of the bunny point cloud. This example shows that the partitioning scheme is robust to the presence of non-toroidal components, in this case a sphere.
Figure 8. COIL 20: The top row shows images in the original dataset. The bottom row shows the reconstructed dataset with interpolated images.
Parameter
a
b
c
τ1
τ2
f0
a+
a−
b+
b−
κ
Value
1/180
1/120
0.9/138
1.15 months
5.75 months
1/365 day -1
1
1
1.5
0.3
1.4
Table 1. Parameters for Delayed Oscillator model ( 24 )
Figure 9. Attractor Recovery for ENSO. (a) Given the time series derived from the delayed oscillator model ( 24 ), we compute the sliding-window embedding and fit a torus. (b) Persistence diagram for the ENSO sliding-window point cloud shows two highly persistent classes, consistent with the expected toroidal structure.
Method
RMSE
Std
Size (floats)
Compression Factor
ToroFit (K=60, λ =0.04642)
6.3399e-04
1.3e-04
131,769
5.30 ×
ToroFit (K=40, λ =0.0001)
8.6519e-04
1.3e-04
59,049
11.83 ×
Toroidal coords + k-NN
8.9184e-04
1.3e-04
853,941
0.82 ×
Isomap(2) + k-NN
1.9833e-02
1.1e-03
862,941
0.81 ×
PCA(2) linear
3.3861e-02
5.6e-05
27
25,877 ×
Table 2. Best sweep configurations vs. reconstruction baselines, same CV splits. “Model size” is the number of floats needed to reconstruct held-out points (ToroFit: the fitted coefficient matrix Ω , size (2K+1)2N ; the k-NN baselines: the retained training set); “Compresion Factor” is that size relative to the raw training data matrix ( Ntrain×D=77,631×9 ).
Table 3. Genes at the extremes of cyclic expression variation across the lytic cycle. Genes marked † are named in the literature but are annotated as hypothetical proteins in the reference used by Lou et al. [ 19 ] (ToxoDB ME49, release 59)
Figure 10. Lytic Cycle: Top-5 and Bottom-5 genes extracted using ToroFit
Figure 11. Pseudotime Analysis for Lytic Cycles. Different representations of the scRNA data.
Figure 12. Arrest and Proliferative Trajectories: 3D PHATE representation with fitted trajectories. Arrest trajectory is depicted in red while the proliferative trajectory is depicted in blue.
Figure 13. Feature expression curves for pRB and DNA. Arrest trajectory is depicted in red while the proliferative trajectory is depicted in blue.
Figure 14. Feature expression for Cdt1, p21, CDK2 (cyto), and cycD1 along the Arrest trajectory (red) and Proliferative trajectory (blue). The rank refers to the rank of each feature according to the L2 -difference between proliferative and arrest feature trajectories.
Appendix figures & tables1 asset
Supplementary material from the paper’s appendix.
Appendix
Table 4. All 40 features ordered by L2 metric between proliferative and arrest cycles, with trajectory thumbnails.
This paper presents an algorithm for detecting and fitting multiple ellipsoids in 2D, 3D and higher-dimensional data. The representation is based on tangents to the ellipsoid together with their contact points, rather than on points alone. A tangent contributes d independent equations where a point contributes one, so each hypothesis needs fewer samples, which in turn reduces the number of iterations required to recover every ellipsoid. The problem is posed as sets of linear equations whose solutions -- the ellipsoid parameters -- are recovered robustly by clustering in parameter space. The algorithm works in all dimensions and outperforms the state of the art in 2D. The approach applies in general to the robust solution of linear equations, and so has utility beyond ellipsoid fitting.
Efrat Dahan, Michael Werman
School of Computer Science, The Hebrew University of Jerusalem, Israel
Tensor Ring (TR) decomposition is a powerful tool for high-order data modeling, but is inherently restricted to discrete forms defined on fixed meshgrids. In this work, we propose a TR functional decomposition for both meshgrid and non-meshgrid data, where factors are parameterized by Implicit Neural Representations (INRs). However, optimizing this continuous framework to capture fine-scale details is intrinsically difficult. Through a frequency-domain analysis, we demonstrate that the spectral structure of TR factors determines the frequency composition of the reconstructed tensor and limits the high-frequency modeling capacity. To mitigate this, we propose a reparameterized TR functional decomposition, in which each TR factor is a structured combination of a learnable latent tensor and a fixed basis. This reparameterization is theoretically shown to improve the training dynamics of TR factor learning. We further derive a principled initialization scheme for the fixed basis and prove the Lipschitz continuity of our proposed model. Extensive experiments on image inpainting, denoising, super-resolution, and point cloud recovery demonstrate that our method achieves consistently superior performance over existing approaches. Code is available at https://github.com/YangyangXu2002/RepTRFD.
Yangyang Xu, Junbo Ke, You-Wei Wen +1
Key Laboratory of Computing and Stochastic Mathematics (Ministry of Education), School of Mathematics and Statistics, Hunan Normal University · Department of Statistics and Data Science, Southern University of Science and Technology
This paper proposes a method for estimating a surface that contains a given set of points from noisy measurements. More precisely, by assuming that the surface is described by the zero set of a function in the span of a given set of features and a parametric description of the distribution of the noise, a computationally efficient method is described that estimates both the surface and the noise distribution parameters. In the provided examples, polynomial and sinusoidal basis functions were used. However, any chosen basis that satisfies the outlined conditions mentioned in the paper can be approximated as a combination of trigonometric, exponential, and/or polynomial terms, making the presented approach highly generalizable. The proposed algorithm exhibits linear computational complexity in the number of samples. Our approach requires no hyperparameter tuning or data preprocessing and effectively handles data in dimensions beyond 2D and 3D. The theoretical results demonstrating the convergence of the proposed algorithm have been provided. To highlight the performance of the proposed method, comprehensive numerical results are conducted, evaluating our method against state-of-the-art algorithms, including Poisson Reconstruction and the Neural Network-based Encoder-X, on 2D and 3D shapes. The results demonstrate the superiority of our method under the same conditions.
Omar M. Sleem, Sahand Kiani, Constantino M. Lagoa
Kyocera International, Inc. · Pennsylvania State University