Position: Embracing Negative Results in Machine Learning

• DOI: 10.48550/arxiv.2406.03980
• Publication Date: 2024-06-06
• Authors: Florian Karl; Lukas Malte Kemeter; Gabriel Dax; Paulina Sierak
• Keywords: Position (finance)

Publications proposing novel machine learning methods are often primarily rated by exhibited predictive performance on selected problems. In this position paper we argue that alone is not a good indicator for the worth of publication. Using it as such even fosters problems like inefficiencies research community whole and setting wrong incentives researchers. We therefore put out call publication "negative" results, which can help alleviate some these improve scientific output community. To substantiate our position, present advantages publishing negative results provide concrete measures to move towards paradigm where their normalized.