Research Group Matthias Feurer

Hyperparameter optimization is crucial for obtaining peak performance of machine learning models. The standard protocol evaluates various hyperparameter configurations using a resampling estimate of the generalization error to guide optimization and select a final hyperparameter configuration. Without much evidence, paired resampling splits, i.e., either a fixed train-validation split or a fixed cross-validation scheme, are often recommended. We show that, surprisingly, reshuffling the splits for every configuration often improves the final model's generalization performance on unseen data. Our theoretical analysis explains how reshuffling affects the asymptotic behavior of the validation loss surface and provides a bound on the expected regret in the limiting regime. This bound connects the potential benefits of reshuffling to the signal and noise characteristics of the underlying optimization problem. We confirm our theoretical results in a controlled simulation study and demonstrate the practical usefulness of reshuffling in a large-scale, realistic hyperparameter optimization experiment. While reshuffling leads to test performances that are competitive with using fixed splits, it drastically improves results for a single train-validation holdout protocol and can often make holdout become competitive with standard CV while being computationally cheaper.

Machine Learning is a core building block in novel data-driven applications. Practitioners face many ambiguous design decisions while developing practical machine learning (ML) solutions. Automated machine learning (AutoML) facilitates the development of machine learning applications by providing efficient methods for optimizing hyperparameters, searching for neural architectures, or constructing whole ML pipelines (Hutter et al., 2019). Thereby, design decisions such as the choice of modelling, pre-processing, and training algorithm are crucial to obtaining well-performing solutions. By automatically obtaining ML solutions, AutoML aims to lower the barrier to leveraging machine learning and reduce the time needed to develop or adapt ML solutions for new domains or data.Highly performant software packages for automatically building ML pipelines given data, so-called AutoML systems, are available and can be used off-the-shelf. Typically, AutoML systems evaluate ML models sequentially to return a well-performing single best model or multiple models combined into an ensemble. Existing AutoML systems are typically highly engineered monolithic software developed for specific use cases to perform well and robustly under various conditions...

We warn against a common but incomplete understanding of empirical research in machine learning (ML) that leads to non-replicable results, makes findings unreliable, and threatens to undermine progress in the field. To overcome this alarming situation, we call for more awareness of the plurality of ways of gaining knowledge experimentally but also of some epistemic limitations. In particular, we argue most current empirical ML research is fashioned as confirmatory research while it should rather be considered exploratory.

Automated machine learning (AutoML) was formed around the fundamental objectives of automatically and efficiently configuring machine learning (ML) workflows, aiding the research of new ML algorithms, and contributing to the democratization of ML by making it accessible to a broader audience. Over the past decade, commendable achievements in AutoML have primarily focused on optimizing predictive performance. This focused progress, while substantial, raises questions about how well AutoML has met its broader, original goals. In this position paper, we argue that a key to unlocking AutoML’s full potential lies in addressing the currently underexplored aspect of user interaction with AutoML systems, including their diverse roles, expectations, and expertise. We envision a more human-centered approach in future AutoML research, promoting the collaborative design of ML systems that tightly integrates the complementary strengths of human expertise and AutoML methodologies.

The recently developed Prior-Data Fitted Networks (PFNs) have shown very promising results for applications in low-data regimes. The TabPFN model, a special case of PFNs for tabular data, is able to achieve state-of-the-art performance on a variety of classification tasks while producing posterior predictive distributions in mere seconds by in-context learning without the need for learning parameters or hyperparameter tuning. This makes TabPFN a very attractive option for a wide range of domain applications. However, a major drawback of the method is its lack of interpretability. Therefore, we propose several adaptations of popular interpretability methods that we specifically design for TabPFN. By taking advantage of the unique properties of the model, our adaptations allow for more efficient computations than existing implementations. In particular, we show how in-context learning facilitates the estimation of Shapley values by avoiding approximate retraining and enables the use of Leave-One-Covariate-Out (LOCO) even when working with large-scale Transformers. In addition, we demonstrate how data valuation methods can be used to address scalability challenges of TabPFN.

Data in tabular form makes up a large part of real-world ML applications, and thus, there has been a strong interest in developing novel deep learning (DL) architectures for supervised learning on tabular data in recent years. As a result, there is a debate as to whether DL methods are superior to the ubiquitous ensembles of boosted decision trees. Typically, the advantage of one model class over the other is claimed based on an empirical evaluation, where different variations of both model classes are compared on a set of benchmark datasets that supposedly resemble relevant real-world tabular data. While the landscape of state-of- the-art models for tabular data changed, one factor has remained largely constant over the years: The datasets. Here, we examine 30 recent publications and 187 different datasets they use, in terms of age, study size and relevance. We found that the average study used less than 10 datasets and that half of the datasets are older than 20 years. Our insights raise questions about the conclusions drawn from previous studies and urge the research community to develop and publish additional recent, challenging and relevant datasets and ML tasks for supervised learning on tabular data.