Domjan Barić

We introduce the Deep Convolutional Interpreter for Time Series (DCIts), a deep-learning architecture for nonlinear multivariate time series that provides sample-specific, locally interpretable descriptions of the underlying interaction structure. Unlike standard black-box forecasters, DCIts learns a time- and lag-dependent transition tensor explicitly factorized into two components: a Focuser, which selects relevant source series and time lags via a sparse masking mechanism, and a Modeler, which assigns signed coefficients to these selected interactions. This decomposition yields a local lag-adjacency structure and signed source–lag contributions for every forecast instance, enabling direct inspection of effective connectivity; when higher-order branches are activated, the same framework yields order-resolved elementwise polynomial contributions. Architecturally, DCIts uses a diverse bank of convolutional filters to capture temporal and cross-variable dependencies, which are mapped through a bottleneck network to the transition tensor. On controlled benchmark datasets with a known interaction structure, we demonstrate that DCIts achieves competitive forecasting error relative to a strong interpretable baseline while recovering stable, signed, lag-resolved interaction patterns. The framework thus prioritizes intrinsic interpretability, using forecasting accuracy as a faithfulness constraint rather than the sole objective.

Topic modeling is a fundamental technique in natural language processing used to uncover latent themes in large text corpora, yet existing approaches struggle to jointly achieve interpretability, semantic coherence, and scalability. Classical probabilistic models such as LDA and NMF rely on bag-of-words assumptions that obscure contextual meaning, while embedding-based methods (e.g., BERTopic, Top2Vec) improve coherence at the expense of diversity and stability. Prompt-based frameworks (e.g., TopicGPT) enhance interpretability but remain sensitive to prompt design and are computationally costly on large datasets. This study introduces VISTA (Vector-Similarity Topic Analysis), a multi-view, hierarchical, and interpretable framework that integrates complementary document embeddings, mutual-nearest-neighbor hierarchical clustering with selective dimension analysis, and large language model (LLM)-based topic labeling enforcing hierarchical coherence. Experiments on three heterogeneous corpora—BBC News, BillSum, and a mixed U.S. Government agency news + Twitter dataset—show that VISTA consistently ranks among the top-performing models, achieving the highest C_UCI coherence and a strong balance between topic diversity and semantic consistency. Qualitative analyses confirm that VISTA identifies domain-relevant themes overlooked by probabilistic or prompt-based models. Overall, VISTA provides a scalable, semantically robust, and interpretable framework for topic discovery, bridging probabilistic, embedding-based, and LLM-driven paradigms in a unified and reproducible design.

We introduce an interpretable deep learning model for multivariate time series forecasting that prioritizes both predictive performance and interpretability - key requirements for understanding complex physical phenomena. Our model not only matches but often surpasses existing interpretability methods, achieving this without compromising accuracy. Through extensive experiments, we demonstrate its ability to identify the most relevant time series and lags that contribute to forecasting future values, providing intuitive and transparent explanations for its predictions. To minimize the need for manual supervision, the model is designed so one can robustly determine the optimal window size that captures all necessary interactions within the smallest possible time frame. Additionally, it effectively identifies the optimal model order, balancing complexity when incorporating higher-order terms. These advancements hold significant implications for modeling and understanding dynamic systems, making the model a valuable tool for applied and computational physicists.

Motivation: Tandem monomeric units, integral components of eukaryotic genomes, form higher-order repeat (HOR) structures that play crucial roles in maintaining chromosome integrity and regulating gene expression and protein abundance. Given their significant influence on processes such as evolution, chromosome segregation, and disease, developing a sensitive and automated tool for identifying HORs across diverse genomic sequences is essential. Results: In this study, we applied the GRMhor (Global Repeat Map hor) algorithm to analyse the centromeric region of chromosome 20 in three individual human genomes, as well as in the centromeric regions of three higher primates. In all three human genomes, we identified six distinct HOR arrays, which revealed significantly greater differences in the number of canonical and variant copies, as well as in their overall structure, than would be expected given the 99.9% genetic similarity among humans. Furthermore, our analysis of higher primate genomes, which revealed entirely different HOR sequences, indicates a much larger genomic divergence between humans and higher primates than previously recognized. These results underscore the suitability of the GRMhor algorithm for studying specificities in individual genomes, particularly those involving repetitive monomers in centromere structure, which is essential for proper chromosome segregation during cell division, while also highlighting its utility in exploring centromere evolution and other repetitive genomic regions. Availability and implementation: Source code and example binaries freely available for download at github.com/gluncic/GRM2023.

Aims: Heart failure with reduced ejection fraction (HFrEF) poses significant challenges for clinicians and researchers, owing to its multifaceted aetiology and complex treatment regimens. In light of this, artificial intelligence methods offer an innovative approach to identifying relationships within complex clinical datasets. Our study aims to explore the potential for machine learning algorithms to provide deeper insights into datasets of HFrEF patients.Methods: To this end, we analysed a cohort of 386 HFrEF patients who had been initiated on sodium‐glucose co‐transporter‐2 inhibitor treatment and had completed a minimum of a 6‐month follow‐up.Results: In traditional frequentist statistical analyses, patients receiving the highest doses of beta‐blockers (BBs) (chi‐square test, P = .036) and those newly initiated on sacubitril‐valsartan (chi‐square test, P = .023) showed better outcomes. However, none of these pharmacological features stood out as independent predictors of improved outcomes in the Cox proportional hazards model. In contrast, when employing eXtreme Gradient Boosting (XGBoost) algorithms in conjunction with the data using Shapley additive explanations (SHAP), we identified several models with significant predictive power. The XGBoost algorithm inherently accommodates non‐linear distribution, multicollinearity and confounding. Within this framework, pharmacological categories like ‘newly initiated treatment with sacubitril/valsartan’ and ‘BB dose escalation’ emerged as strong predictors of long‐term outcomes.Conclusions: In this manuscript, we not only emphasize the strengths of this machine learning approach but also discuss its potential limitations and the risk of identifying statistically significant yet clinically irrelevant predictors.

Background: Due to recent changes in breast cancer treatment strategy, significantly more patients are treated with neoadjuvant systemic therapy (NST). Radiological methods do not precisely determine axillary lymph node status, with up to 30% of patients being misdiagnosed. Hence, supplementary methods for lymph node status assessment are needed. This study aimed to apply and evaluate machine learning models on clinicopathological data, with a focus on patients meeting NST criteria, for lymph node metastasis prediction. Methods: From the total breast cancer patient data (n = 8381), 719 patients were identified as eligible for NST. Machine learning models were applied for the NST-criteria group and the total study population. Model explainability was obtained by calculating Shapley values. Results: In the NST-criteria group, random forest achieved the highest performance (AUC: 0.793 [0.713, 0.865]), while in the total study population, XGBoost performed the best (AUC: 0.762 [0.726, 0.795]). Shapley values identified tumor size, Ki-67, and patient age as the most important predictors. Conclusion: Tree-based models achieve a good performance in assessing lymph node status. Such models can lead to more accurate disease stage prediction and consecutively better treatment selection, especially for NST patients where radiological and clinical findings are often the only way of lymph node assessment.

The adaptation of deep learning models within safety-critical systems cannot rely only on good prediction performance but needs to provide interpretable and robust explanations for their decisions. When modeling complex sequences, attention mechanisms are regarded as the established approach to support deep neural networks with intrinsic interpretability. This paper focuses on the emerging trend of specifically designing diagnostic datasets for understanding the inner workings of attention mechanism based deep learning models for multivariate forecasting tasks. We design a novel benchmark of synthetically designed datasets with the transparent underlying generating process of multiple time series interactions with increasing complexity. The benchmark enables empirical evaluation of the performance of attention based deep neural networks in three different aspects: (i) prediction performance score, (ii) interpretability correctness, (iii) sensitivity analysis. Our analysis shows that although most models have satisfying and stable prediction performance results, they often fail to give correct interpretability. The only model with both a satisfying performance score and correct interpretability is IMV-LSTM, capturing both autocorrelations and crosscorrelations between multiple time series. Interestingly, while evaluating IMV-LSTM on simulated data from statistical and mechanistic models, the correctness of interpretability increases with more complex datasets.