We investigate hybrid linear-morphological networks. Recent studies highlight the inherent affinity of morphological layers to pruning, but also their difficulty in training. We propose a hybrid network structure, wherein morphological layers are inserted between the linear layers of the network, in place of activation functions. We experiment with the following morphological layers: 1) maxout pooling layers (as a special case of a morphological layer), 2) fully connected dense morphological layers, and 3) a novel, sparsely initialized variant of (2). We conduct experiments on the Magna-Tag-A-Tune (music auto-tagging) and CIFAR-10 (image classification) datasets, replacing the linear classification heads of state-of-the-art convolutional network architectures with our proposed network structure for the various morphological layers. We demonstrate that these networks induce sparsity to their linear layers, making them more prunable under L1 unstructured pruning. We also show that on MTAT our proposed sparsely initialized layer achieves slightly better performance than ReLU, maxout, and densely initialized max-plus layers, and exhibits faster initial convergence.

In this article, the beta version 0.1.0 of Opera Graeca Adnotata (OGA), the largest open-access multilayer corpus for Ancient Greek (AG) is presented. OGA consists of 1,687 literary works and 34M+ tokens coming from the PerseusDL and OpenGreekAndLatin GitHub repositories, which host AG texts ranging from about 800 BCE to about 250 CE. The texts have been enriched with seven annotation layers: (i) tokenization layer; (ii) sentence segmentation layer; (iii) lemmatization layer; (iv) morphological layer; (v) dependency layer; (vi) dependency function layer; (vii) Canonical Text Services (CTS) citation layer. The creation of each layer is described by highlighting the main technical and annotation-related issues encountered. Tokenization, sentence segmentation, and CTS citation are performed by rule-based algorithms, while morphosyntactic annotation is the output of the COMBO parser trained on the data of the Ancient Greek Dependency Treebank. For the sake of scalability and reusability, the corpus is released in the standoff formats PAULA XML and its offspring LAULA XML.

Morphological neural networks, or layers, can be a powerful tool to boost the progress in mathematical morphology, either on theoretical aspects such as the representation of complete lattice operators, or in the development of image processing pipelines. However, these architectures turn out to be difficult to train when they count more than a few morphological layers, at least within popular machine learning frameworks which use gradient descent based optimization algorithms. In this paper we investigate the potential and limitations of differentiation based approaches and back-propagation applied to morphological networks, in light of the non-smooth optimization concept of Bouligand derivative. We provide insights and first theoretical guidelines, in particular regarding initialization and learning rates.

To better appreciate how a complex organ such as the brain functions, scientists strive to accurately understand both its detailed cellular architecture and the intercellular communications taking place within it.

Over the last decade, deep learning has made several breakthroughs and demonstrated successful applications in various fields (e.g. in computer vision Krizhevsky et al. [2012], Simonyan and Zisserman [2014a], He et al. [2016a], Huang et al. [2017], object detection Redmon et al. [2016], or NLP Dai et al. [2019], Radford et al. [2019]). This success is mainly due to its automation of the feature engineering process. This success is mainly attributable to the fact that it automates the feature engineering process. Rather than manually designed features, features are learned in an end-to-end process from data. The need for improved architecture has swiftly followed the advent of deep learning. Experts now place a premium on architecture engineering in lieu of features engineering. Architecture engineering is concerned with determining the most appropriate operations for the network, their hyperparameters (e.g. the number of neurons for fully connected layers, or the number of filters or kernel size for convolutional layers), and the connectivity of all the operations. Generally, practitioners propose novel operations to validate various architectures and tasks in order to improve performance on specific tasks. As a result, developing a novel operation remains a time-consuming and costly process.

--Evaluating, explaining, and visualizing high-level concepts in generative models, such as variational autoencoders (V AEs), is challenging in part due to a lack of known prediction classes that are required to generate saliency maps in supervised learning. While saliency maps may help identify relevant features (e.g., pixels) in the input for classification tasks of deep neural networks, similar frameworks are understudied in unsupervised learning. Therefore, we introduce a new method of obtaining saliency maps for latent representations of known or novel high-level concepts, often called concept vectors in generative models. Concept scores, analogous to class scores in classification tasks, are defined as dot products between concept vectors and encoded input data, which can be readily used to compute the gradients. The resulting concept saliency maps are shown to highlight input features deemed important for high-level concepts. Our method is applied to the V AE's latent space of CelebA dataset in which known attributes such as "smiles" and "hats" are used to elucidate relevant facial features. Furthermore, our application to spatial transcriptomic (ST) data of a mouse olfactory bulb demonstrates the potential of latent representations of morphological layers and molecular features in advancing our understanding of complex biological systems. By extending the popular method of saliency maps to generative models, the proposed concept saliency maps help improve interpretability of latent variable models in deep learning. I NTRODUCTION A rapidly increasing amount of unlabeled data, such as images and molecular data, has prompted a rise of deep generative models, that can be trained without human supervision. By using a vast amount of unlabeled data, unsupervised learning models such as variational autoencoders (V AEs) [1], [2] extract low-dimensional latent spaces that compactly encode high-dimensional input data and potentially reveal hidden relationships.