Topological deep learning (TDL) is a rapidly evolving field that uses topological features to understand and design deep learning models. This paper posits that TDL is the new frontier for relational learning. TDL may complement graph representation learning and geometric deep learning by incorporating topological concepts, and can thus provide a natural choice for various machine learning settings. To this end, this paper discusses open problems in TDL, ranging from practical benefits to theoretical foundations. For each problem, it outlines potential solutions and future research opportunities. At the same time, this paper serves as an invitation to the scientific community to actively participate in TDL research to unlock the potential of this emerging field.

Investigation of human brain states through electroencephalograph (EEG) signals is a crucial step in human-machine communications. However, classifying and analyzing EEG signals are challenging due to their noisy, nonlinear and nonstationary nature. Current methodologies for analyzing these signals often fall short because they have several regularity assumptions baked in. This work provides an effective, flexible and noise-resilient scheme to analyze EEG by extracting pertinent information while abiding by the 3N (noisy, nonlinear and nonstationary) nature of data. We implement a topological tool, namely persistent homology, that tracks the evolution of topological features over time intervals and incorporates individual's expectations as prior knowledge by means of a Bayesian framework to compute posterior distributions. Relying on these posterior distributions, we apply Bayes factor classification to noisy EEG measurements. The performance of this Bayesian classification scheme is then compared with other existing methods for EEG signals.

Identification of objective criteria to correctly diagnose ectatic diseases of the cornea or to detect early stages of corneal ectasia is of great interest in ophthalmology and optometry. Metrics for diagnosis typically employed are curvature maps (axial/sagittal, tangential); elevation map of the anterior surface of the cornea with respect to a reference sphere; and pachymetry (thickness) map of the cornea. We present evidence that currently used curvature maps do not represent the actual curvatures (principal or mean) in a human cornea. A novel contribution of this paper is the computation of the true mean curvature over every point of a central region of the cornea. We show that the true mean curvature can accurately identify the location of the ectasia. We present a quartic smoothing spline algorithm for the simultaneous computation of elevation maps for anterior and posterior corneal surfaces, pachymetry, and true mean curvature. The input to the algorithm is data from a single measurement from imaging devices such as an anterior segment optical coherence tomographer or a Scheimpflug imager. We show that a different combination of metrics is useful for the diagnosis of existing ectasia (true mean curvature and anterior elevation map) as opposed to subclinical ectasia (pachymetry and posterior elevation map). We compare our results with existing algorithms, and present applications to a normal cornea, a forme fruste keratoconic cornea, and an advanced keratoconic cornea.

Optical coherence tomography (OCT) is a non-invasive imaging technique used to study and understand internal structures of biological tissues such as the anterior chamber of the human eye. An interesting problem is the reconstruction of the shape of the biological tissue from OCT images, that is not only a good fit of the data but also respects the smoothness properties observed in the images. A similar problem arises in Magnetic Resonance Imaging (MRI). We cast the problem as a penalized weighted least squares regression with a penalty on the magnitude of the second derivative (Laplacian) of the surface. We present a novel algorithm to construct the Kimeldorf-Wahba solution for unit ball domains. Our method unifies the ad-hoc approaches currently in the literature. Application of the theory to data from an anterior segment optical coherence tomographer is presented. A detailed comparison of the reconstructed surface using different approaches is presented.