Franz Rothlauf

Transformer Semantic Genetic Programming (TSGP) is a semantic search approach that uses a pre-trained transformer model as a variation operator to generate offspring programs with high semantic similarity to a given parent. Unlike other semantic GP approaches that rely on fixed syntactic transformations, TSGP aims to learn diverse structural variations that lead to solutions with similar semantics. We find that a single transformer model trained on millions of programs is able to generalize across symbolic regression problems of varying dimension. Evaluated on 24 real-world and synthetic datasets, TSGP significantly outperforms standard GP, SLIM_GSGP, Deep Symbolic Regression, and Denoising Autoencoder GP, achieving an average rank of 1.58 across all benchmarks. Moreover, TSGP produces more compact solutions than SLIM_GSGP, despite its higher accuracy. In addition, the target semantic distance is able to effectively adjust the step size in the semantic space: small values enable consistent improvement in fitness but often lead to larger programs, while larger values promote faster convergence and compactness. Thus, the target semantic distance provides an effective mechanism for balancing exploration and exploitation.

Background: Cancer registries collect information on cancer patients from medical institutions. Standardized data processing at these institutions and registries, which produces comparable data, is crucial for subsequent data analysis. Otherwise, processing artifacts can hide treatment effects. To ensure comparable data, coding guidelines exist, but they can be incomplete, outdated, open to interpretation, or applied inconsistently. These discrepancies can introduce systematic differences in the reported data. Current evaluations of data comparability assess whether institutions follow coding guidelines rather than whether the reported data is comparable. This approach can miss systematic differences even when institutions conform to guidelines.Methods: We propose a data-driven approach combining machine learning (ML) and statistical methods to detect systematic differences between datasets from reporting institutions. To identify task-agnostic differences, we predict the reporting institution using classifiers; good discriminability indicates multivariate differences between the datasets. Statistical tests identify significant differences among variables, and comparisons of empirical distributions reveal differences in variable values. To identify task-specific differences, we train an ML model on data from one institution and evaluate it on data from both institutions. A drop in performance indicates distributional differences between the institutions' datasets. Findings: We validated our data-driven approach using data from the Cancer Registry Rhineland-Palatinate (CR-RLP, Germany) and the Netherlands Cancer Registry (NCR). For the CR-RLP, we identified systematic differences across five coding variables, with one reporting institution reporting substantially higher proportions of unspecified or imprecise values than the other. For the NCR, we observed differences in morphology subtypes. Interpretation: Our data-driven approach reveals potential coding differences between datasets, variables, and variable values. These differences could stem from decentralized coder training, allowing greater variability in documentation, or responsibility for data reporting. Our approach can help improve processes, inform data pooling, and identify reporting institutions where data coding could benefit from reviews.

Electronic health records (EHR) mostly consist of highly sensitive data, whose sharing is highly regulated due to privacy and security concerns. To address privacy protections, de-identification and anonymization techniques offer a more secure way of data transmission to ensure patient’s privacy. Principal components analysis (PCA) has emerged as an anonymization technique. The applicability of more diverse and complex dimensionality reduction (DR) techniques such as Factor Analysis of Mixed Data (FAMD), or Autoencoder (AE), remains largely unexplored. The objective of this study is an investigation into the potential of FAMD and AE as anonymizers for EHR. The goal of this study is to apply advanced DR methods on three EHR datasets of varying sizes. Subsequently, a supervised prediction task is performed on the anonymized EHR using an artificial neural network (ANN), and the performance of the prediction task is compared to the performance of the original, non-anonymized EHR. The findings indicate that the use of FAMD and AE in the anonymization of EHR results in a comparable performance to that of the original, non-anonymized EHR in a downstream prediction task. This study thus indicates the potential of advanced DR as an anonymization technique for EHR, thereby underscoring the necessity for further investigation.

Informed down-sampling (IDS) is known to improve performance in symbolic regression when combined with various selection strategies, especially tournament selection. However, recent work found that IDS's gains are not consistent across all problems. Our analysis reveals that IDS performance is worse for problems containing outliers. IDS systematically favors including outliers in subsets which pushes GP towards finding solutions that overfit to outliers. To address this, we introduce ROIDS (Robust Outlier-Aware Informed Down-Sampling), which excludes potential outliers from the sampling process of IDS. With ROIDS it is possible to keep the advantages of IDS without overfitting to outliers and to compete on a wide range of benchmark problems. This is also reflected in our experiments in which ROIDS shows the desired behavior on all studied benchmark problems. ROIDS consistently outperforms IDS on synthetic problems with added outliers as well as on a wide range of complex real-world problems, surpassing IDS on over 80% of the real-world benchmark problems. Moreover, compared to all studied baseline approaches, ROIDS achieves the best average rank across all tested benchmark problems. This robust behavior makes ROIDS a reliable down-sampling method for selection in symbolic regression, especially when outliers may be included in the data set.

In today’s online world, bias is ubiquitous. In a Video-on-Demand (VOD) context, there are two main sources for bias: (1) content providers and their methods that generate recommendations; (2) user preferences. We use real world data from one of the largest German Public Service Media (PSM) platforms to study both possible causes by examining content recommendations and content consumption (classified as Education, Information, Culture, and Entertainment according to the regulations governing PSM operations). We find that Education and Entertainment content to be recommended disproportionately. Algorithmic personalized recommendations lead to a bias towards both Education and Entertainment for the recommendations. Editorial content recommendations favor Entertainment content. Additionally, we find that the bias introduced by the recommender systems to be larger than the one introduced by the viewer. We recommend testing an increased use of algorithmic recommendations as these might lead to a lower bias towards Entertainment content.

Despite the continuous proposal of new anomaly detection algorithms and extensive benchmarking efforts, progress seems to stagnate, with only minor performance differences between established baselines and new algorithms. In this position paper, we argue that this stagnation is due to limitations in how we evaluate anomaly detection algorithms. Current benchmarking does not, for example, sufficiently reflect the diversity of anomalies in applications ranging from predictive maintenance to scientific discovery. Consequently, we need to rethink benchmarking in anomaly detection. In our opinion, anomaly detection should be studied using scenarios that capture the relevant characteristics of different applications. We identify three key areas for improvement: First, we need to identify anomaly detection scenarios based on a common taxonomy. Second, anomaly detection pipelines should be analyzed end-to-end and by component. Third, evaluating anomaly detection algorithms should be meaningful regarding the scenario's objectives.