• This is the link to the arxiv article.
• This is the link to slides for the talk given at the IMTA-7.
• This is the link to the transcript of the talk given at the IMTA-7.

In this paper we present an approach to determine the smallest possible number of perceptrons in a neural net in such a way that the topology of the input space can be learned sufficiently well. We introduce a general procedure based on persistent homology to investigate topological invariants of the manifold on which we suspect the data set. We specify the required dimensions precisely, assuming that there is a smooth manifold on or near which the data are located. Furthermore, we require that this space is connected and has a commutative group structure in the mathematical sense. These assumptions allow us to derive a decomposition of the underlying space whose topology is well known. We use the representatives of the k-dimensional homology groups from the persistence landscape to determine an integer dimension for this decomposition. This number is the dimension of the embedding that is capable of capturing the topology of the data manifold. We derive the theory and validate it experimentally on toy data sets.

Keywords: Embedding Dimension, Parameterization, Persistent Homology, Neural Networks and Manifold Learning.

@inproceedings{imta7/MelodiaL21,
  author    = {Luciano Melodia and
               Richard Lenz},
  editor    = {Del Bimbo, A.,
               Cucchiara, R.,
               Sclaroff, S.,
               Farinella, G.M.,
               Mei, T.,
               Bertini, M.,
               Escalante, H.J.,
               Vezzani, R.},
  title     = {Estimate of the Neural Network Dimension using Algebraic Topology and Lie Theory},
  booktitle = {Pattern Recognition. ICPR International Workshops and Challenges, {IMTA VII}
               2021, Milano, Italy, January 11, 2021, Proceedings},
  series    = {Lecture Notes in Computer Science},
  volume    = {12665},
  pages     = {15--29},
  publisher = {Springer},
  year      = {2021},
  url       = {https://doi.org/10.1007/978-3-030-68821-9_2},
  doi       = {10.1007/978-3-030-68821-9_2},
}

1. Invertible autoencoders autoencoderInvertible.py
   • Remove tensor elements
   • Get prime factors
   • Load example Keras datasets
   • Add gaussian noise to data
   • Crop tensor elements
   • Greate a group of convolutional layers
   • Loop over a group of convolutional layers
   • Invertible Keras neural network layer
   • Convert dimensions into 2D-convolution
   • Embedded invertible autoencoder model
2. Count representatives from homology groups countHomgroups.py
   • List differences
   • Execute experimental autoencoder training
   • Derive dimension from homology groups
3. Persistence landscapes persistenceLandscapes.py
   • Concatenated multiple persistence landscapes
   • Compute persistence landscapes
   • Compute mean persistence landscapes
4. Persistence statistics persistenceStatistics.py
   • Hausdorff intervall
   • Truncated simplex trees

take_out_element(k: tuple, r) -> tuple

A function taking out specific values.

• param k: tuple object to be processed, type tuple.
• param r: value to be removed, type int, float, string, None.
• return k2: cropped tuple object, type tuple.

primeFactors(n)

A function that returns the prime factors of an integer.

• param n: an integer, type int.
• return factors: a list of prime factors, type list.

load_data_keras(dimensions: tuple, factor: float = 255.0, dataset: str = 'mnist') -> tuple

A utility function to load datasets.

This functions helps to load particular datasets ready for a processing with convolutional or dense autoencoders. It depends on the specified shape (the input dimensions). This functions is for validation purpose and works for keras datasets only. Supported datasets are mnist (default), cifar10, cifar100 and boston_housing. The shapes: mnist (28,28,1), cifar10 (32,32,3), cifar100 (32,32,3)

• param dimensions: dimension of the data, type tuple.
• param factor: division factor, default is 255, type float.
• param dataset: keras dataset, default is mnist,type str.
• return X_train, X_test, input_image: , type tuple.

add_gaussian_noise(data: numpy.ndarray, noise_factor: float = 0.5, mean: float = 0.0, std: float = 1.0) -> numpy.ndarray

A utility function to add gaussian noise to data.

The purpose of this functions is validating certain models under gaussian noise. The noise can be added changing the mean, standard deviation and the amount of noisy points added.

• param noise_factor: amount of noise in percent, type float.
• param data: dataset, type np.ndarray.
• param mean: mean, type float.
• param std: standard deviation, type float.
• return x_train_noisy: noisy data, type np.ndarray.

crop_tensor(dimension: int, start: int, end: int) -> Callable

A utility function cropping a tensor along a given dimension.

The purpose of this function is to be used for multivariate cropping and to serve as a procedure for the invertible autoencoders, which need a cropping to make the matrices trivially invertible, as can be seen in the Real NVP architecture. This procedure works up to dimension 4.

• param dimension: the dimension of cropping, type int.
• param start: starting index for cropping, type int.
• param end: ending index for cropping, type int.
• return Lambda(func): Lambda function on the tensor, type Callable.

convolutional_group(_input: numpy.ndarray, filterNumber: int, alpha: float = 5.5, kernelSize: tuple = (2, 2), kernelInitializer: str = 'uniform', padding: str = 'same', useBias: bool = True, biasInitializer: str = 'zeros')

This group can be extended for deep learning models and is a sequence of convolutional layers.

The convolutions is a 2D-convolution and uses a LeakyRelu activation function. After the activation function batch-normalization is performed on default, to take care of the covariate shift. As default the padding is set to same, to avoid difficulties with convolution.

• param _input: data from previous convolutional layer, type np.ndarray.
• param filterNumber: multiple of the filters per layer, type int.
• param alpha: parameter for LeakyRelu activation function, default 5.5, type float.
• param kernelSize: size of the 2D kernel, default (2,2), type tuple.
• param kernelInitializer: keras kernel initializer, default uniform, type str.
• param padding: padding for convolution, default same, type str.
• param useBias: whether or not to use the bias term throughout the network, type bool.
• param biasInitializer: initializing distribution of the bias values, type str.
• return data: processed data by neural layers, type np.ndarray.

loop_group(group: Callable, groupLayers: int, element: numpy.ndarray, filterNumber: int, kernelSize: tuple, useBias: bool = True, kernelInitializer: str = 'uniform', biasInitializer: str = 'zeros') -> numpy.ndarray

This callable is a loop over a group specification.

The neural embeddings ends always with dimension 1 in the color channel. For other specifications use the parameter colorChannel. The function operates on every keras group of layers using the same parameter set as 2D convolution.

• param group: a callable that sets up the neural architecture, type Callable.
• param groupLayers: depth of the neural network, type int.
• param element: data, type np.ndarray.
• param filterNumber: number of filters as exponential of 2, type int.
• param kernelSize: size of the kernels, type tuple.
• return data: processed data by neural network, type np.ndarray.
• param useBias: whether or not to use the bias term throughout the network, type bool.
• param biasInitializer: initializing distribution of the bias values, type str.

invertible_layer(data: numpy.ndarray, alpha: float = 5.5, kernelSize: tuple = (2, 2), kernelInitializer: str = 'uniform', groupLayers: int = 6, filterNumber: int = 2, croppingFactor: int = 4, useBias: bool = True, biasInitializer: str = 'zeros') -> numpy.ndarray

Returns an invertible neural network layer.

This neural network layer learns invertible subspaces, parameterized by higher dimensional functions with a trivial invertibility. The higher dimensional functions are also neural subnetworks, trained during learning process.

• param data: data from previous convolutional layer, type np.ndarray.
• param alpha: parameter for LeakyRelu activation function, default 5.5, type float.
• param groupLayers: depth of the neural network, type int.
• param kernelSize: size of the kernels, type tuple.
• param filterNumber: multiple of the filters per layer, type int.
• param croppingFactor: should be a multiple of the strides length, type int.
• param useBias: whether or not to use the bias term throughout the network, type bool.
• param biasInitializer: initializing distribution of the bias values, type str.
• return data: processed data, type np.ndarray.

invertible_subspace_dimension2(units: int)

A helper function converting dimensions into 2D convolution shapes.

This functions works only for quadratic dimension size. It reshapes the data according to an embedding with the same dimension, represented by a 2D array.

• param units: , type int.
• return embedding: , type tuple.

invertible_subspace_autoencoder(data: numpy.ndarray, units: int, invertibleLayers: int, alpha: float = 5.5, kernelSize: tuple = (2, 2), kernelInitializer: str = 'uniform', groupLayers: int = 6, filterNumber: int = 2, useBias: bool = True, biasInitializer: str = 'zeros')

A function returning an invertible autoencoder model.

This model works only with a quadratic number as units. The convolutional embedding dimension in 2D is determined, for the quadratic matrix, as the square root of the respective dimension of the dense layer. This module is for testing purposes and not meant to be part of a productive environment.

• param data: data, type np.ndarray.
• param units: projection dim. into lower dim. by dense layer, type int.
• param invertibleLayers: amout of invertible layers in the middle of the network, type int.
• param alpha: parameter for LeakyRelu activation function, default 5.5, type float.
• param kernelSize: size of the kernels, type tuple.
• param kernelInitializer: initializing distribution of the kernel values, type str.
• param groupLayers: depth of the neural network, type int.
• param filterNumber: multiple of the filters per layer, type int.
• param useBias: whether or not to use the bias term throughout the network, type bool.
• param biasInitializer: initializing distribution of the bias values, type str.
• param filterNumber: an integer factor for each convolutional layer, type int.
• return output: an output layer for keras neural networks, type np.ndarray.

concatenate_landscapes(persLandscape1: numpy.ndarray, persLandscape2: numpy.ndarray, resolution: int) -> list

This function concatenates the persistence landscapes according to homology groups.

The computation of homology groups requires a certain resolution for each homology class. According to this resolution the direct sum of persistence landscapes has to be concatenated in a correct manner, such that the persistent homology can be plotted according to the n-dimensional persistent homology groups.

• param persLandscape1: persistence landscape, type np.ndarray.
• param persLandscape2: persistence landscape, type np.ndarray.
• return concatenatedLandscape: direct sum of persistence landscapes, type list.

compute_persistence_landscape(data: numpy.ndarray, res: int = 1000, persistenceIntervals: int = 1, maxAlphaSquare: float = 1000000000000.0, filtration: str = ['alphaComplex', 'vietorisRips', 'tangential'], maxDimensions: int = 10, edgeLength: float = 0.1, plot: bool = False, smoothen: bool = False, sigma: int = 3) -> numpy.ndarray

A function for computing persistence landscapes for 2D images.

This function computes the filtration of a 2D image dataset, the simplicial complex, the persistent homology and then returns the persistence landscape as array. It takes the resolution of the landscape as parameter, the maximum size for alphaSquare and options for certain filtrations.

• param data: data set, type np.ndarray.
• param res: resolution, default is 1000, type int.
• param persistenceIntervals: interval for persistent homology, default is 1e12,type float.
• param maxAlphaSquare: max. parameter for delaunay expansion, type float.
• param filtration: alphaComplex, vietorisRips, cech, delaunay, tangential, type str.
• param maxDimensions: only needed for VietorisRips, type int.
• param edgeLength: only needed for VietorisRips, type float.
• param plot: whether or not to plot, type bool.
• param smoothen: whether or not to smoothen the landscapes, type bool.
• param sigma: smoothing factor for gaussian mixtures, type int.
• return landscapeTransformed: persistence landscape, type np.ndarray.

compute_mean_persistence_landscapes(data: numpy.ndarray, resolution: int = 1000, persistenceIntervals: int = 1, maxAlphaSquare: float = 1000000000000.0, filtration: str = ['alphaComplex', 'vietorisRips', 'tangential'], maxDimensions: int = 10, edgeLength: float = 0.1, plot: bool = False, tikzplot: bool = False, name: str = 'persistenceLandscape', smoothen: bool = False, sigma: int = 2) -> numpy.ndarray

This function computes mean persistence diagrams over 2D datasets.

The functions shows a progress bar of the processed data and takes the direct sum of the persistence modules to get a summary of the landscapes of the various samples. Further it can be decided whether or not to smoothen the persistence landscape by gaussian filter. A plot can be created with matplotlib or as another option for scientific reporting with tikzplotlib, or both.

Information: The color scheme has 5 colors defined. Thus 5 homology groups can be displayed in different colors.

• param data: data set, type np.ndarray.
• param resolution: resolution of persistent homology per group, type int.
• param persistenceIntervals: intervals for persistence classes, type int.
• param maxAlphaSquare: max. parameter for Delaunay expansion, type float.
• param filtration: alphaComplex, vietorisRips or tangential, type str.
• param maxDimensions: maximal dimension of simplices, type int.
• param edgeLength: length of simplex edge, type float.
• param plot: whether or not to plot, type bool.
• param tikzplot: whether or not to plot as tikz-picture, type bool.
• param name: name of the file to be saved, type str.
• param smoothen: whether or not to smoothen the landscapes, type bool.
• param sigma: smoothing factor for gaussian mixtures, type int.
• return meanPersistenceLandscape: mean persistence landscape, type np.ndarray.

hausd_interval(data: numpy.ndarray, confidenceLevel: float = 0.95, subsampleSize: int = -1, subsampleNumber: int = 1000, pairwiseDist: bool = False, leafSize: int = 2, ncores: int = 2) -> float

Computation of Hausdorff distance based confidence values.

Measures the confidence between two persistent features, wether they are drawn from a distribution fitting the underlying manifold of the data. This function is based on the Hausdorff distance between the points.

• param data: a data set, type np.ndarray.
• param confidenceLevel: confidence level, default 0.95, type float.
• param subsampleSize: size of each subsample, type int.
• param subsampleNumber: number of subsamples, type int.
• param pairwiseDist: if true, a symmetric nxn-matrix is generated out of the data, type bool.
• param leafSize: leaf size for KDTree, type int.
• param ncores: number of cores for parallel computing, type int.
• return confidence: the confidence to be a persistent homology class, type float.

truncated_simplex_tree(simplexTree: numpy.ndarray, int_trunc: int = 100) -> tuple

This function return a truncated simplex tree.

A sparse representation of the persistence diagram in the form of a truncated persistence tree. Speeds up computation on large scale data sets.

• param simplexTree: simplex tree, type np.ndarray.
• param int_trunc: number of persistent interval kept per dimension, default is 100, type int.
• return simplexTreeTruncatedPersistence: truncated simplex tree, type np.ndarray.