Deep convolutional neural networks have shown remarkable performance in the image classification domain. However, Deep Learning models are vulnerable to noise and redundant information encapsulated into the high-dimensional raw input images, leading to unstable and unreliable predictions. Autoencoders constitute an unsupervised dimensionality reduction technique, proven to filter out noise and redundant information and create robust and stable feature representations. In this work, in order to resolve the problem of DL models’ vulnerability, we propose a convolutional autoencoder topological model for compressing and filtering out noise and redundant information from initial high dimensionality input images and then feeding this compressed output into convolutional neural networks. Our results reveal the efficiency of the proposed approach, leading to a significant performance improvement compared to Deep Learning models trained with the initial raw images.

With the proliferation of image-based applications in various domains, the need for accurate and interpretable image similarity measures has become increasingly critical. Existing image similarity models often lack transparency, making it challenging to understand the reasons why two images are considered similar. In this paper, we propose the concept of explainable image similarity, where the goal is the development of an approach, which is capable of providing similarity scores along with visual factual and counterfactual explanations. Along this line, we present a new framework, which integrates Siamese Networks and Grad-CAM for providing explainable image similarity and discuss the potential benefits and challenges of adopting this approach. In addition, we provide a comprehensive discussion about factual and counterfactual explanations provided by the proposed framework for assisting decision making. The proposed approach has the potential to enhance the interpretability, trustworthiness and user acceptance of image-based systems in real-world image similarity applications.

Explainable machine learning is an emerging new domain fundamental for trustworthy real-world applications. A lack of trust and understanding are the main drawbacks of deep learning models when applied to real-world decision systems and prediction tasks. Such models are considered as black boxes because they are unable to explain the reasons for their predictions in human terms; thus, they cannot be universally trusted. In critical real-world applications, such as in medical, legal, and financial ones, an explanation of machine learning (ML) model decisions is considered crucially significant and mandatory in order to acquire trust and avoid fatal ML bugs, which could disturb human safety, rights, and health. Nevertheless, explainable models are more than often less accurate; thus, it is essential to invent new methodologies for creating interpretable predictors that are almost as accurate as black-box ones. In this work, we propose a novel explainable feature extraction and prediction framework applied to 3D image recognition. In particular, we propose a new set of explainable features based on mathematical and geometric concepts, such as lines, vertices, contours, and the area size of objects. These features are calculated based on the extracted contours of every 3D input image slice. In order to validate the efficiency of the proposed approach, we apply it to a critical real-world application: pneumonia detection based on CT 3D images. In our experimental results, the proposed white-box prediction framework manages to achieve a performance similar to or marginally better than state-of-the-art 3D-CNN black-box models. Considering the fact that the proposed approach is explainable, such a performance is particularly significant.

Within the field of computer vision, image segmentation and classification serve as crucial tasks, involving the automatic categorization of images into predefined groups or classes, respectively. In this work, we propose a framework designed for simultaneously addressing segmentation and classification tasks in image-processing contexts. The proposed framework is composed of three main modules and focuses on providing transparency, interpretability, and explainability in its operations. The first two modules are used to partition the input image into regions of interest, allowing the automatic and interpretable identification of segmentation regions using clustering techniques. These segmentation regions are then analyzed to select those considered valuable by the user for addressing the classification task. The third module focuses on classification, using an explainable classifier, which relies on hand-crafted transparent features extracted from the selected segmentation regions. By leveraging only the selected informative regions, the classification model is made more reliable and less susceptible to misleading information. The proposed framework’s effectiveness was evaluated in a case study on skin-cancer-segmentation and -classification benchmarks. The experimental analysis highlighted that the proposed framework exhibited comparable performance with the state-of-the-art deep-learning approaches, which implies its efficiency, considering the fact that the proposed approach is also interpretable and explainable.

Efficient and accurate vision models are essential for real-world applications such as medical imaging and deepfake detection, where both performance and computational efficiency are critical. While recent vision models achieve high accuracy, they often come with the trade-off of increased size and computational demands. In this work, we propose MobileNet-HeX, a new ensemble model based on Heterogeneous MobileNet eXperts, designed to achieve top-tier performance while minimizing computational demands in real-world vision tasks. By utilizing a two-step Expand-and-Squeeze mechanism, MobileNet-HeX first expands a MobileNet population through diverse random training setups. It then squeezes the population through pruning, selecting the top-performing models based on heterogeneity and validation performance metrics. Finally, the selected Heterogeneous eXpert MobileNets are combined via sequential quadratic programming to form an efficient super-learner. MobileNet-HeX is benchmarked against state-of-the-art vision models in challenging case studies, such as skin cancer classification and deepfake detection. The results demonstrate that MobileNet-HeX not only surpasses these models in performance but also excels in speed and memory efficiency. By effectively leveraging a diverse set of MobileNet eXperts, we experimentally show that small, yet highly optimized, models can outperform even the most powerful vision networks in both accuracy and computational efficiency.

Efficient and accurate text classification is essential for a wide range of natural language processing applications, including sentiment analysis, spam detection and machine-generated text identification. While recent advancements in transformer-based large language models have achieved remarkable performance, they often come with significant computational costs, limiting their applicability in resource-constrained environments. In this work, we propose TextNeX, a new ensemble model that leverages lightweight language models to achieve state-of-the-art performance while maintaining computational efficiency. The development process of TextNeX model follows a three-phase procedure: (i) Expansion: generation of a pool of diverse lightweight models via randomized model setups and variations of training data; (ii) Selection: application of a clustering-based heterogeneity-driven selection to retain the most complementary models and (iii) Ensemble optimization: optimization of the selected models’ contributions using sequential quadratic programming. Experimental evaluations on three challenging text classification datasets demonstrate that TextNeX outperforms existing state-of-the-art ensemble models in accuracy, robustness and computational effectiveness, offering a practical alternative to large-scale models for real-world applications.