Herein, we present a novel topic variation detection method that combines a topic extraction method and a change-point detection method. It extracts topics from time-series text data as the feature of each time and detects change points from the changing patterns of the extracted topics. We applied this method to analyze the valuable, albeit underutilized, text dataset containing the Japanese Prime Minister’s (PM’s) detailed daily activities for over 32 years. The proposed method and data provide novel insights into the empirical analyses of political business cycles, which is a classical issue in economics and political science. For instance, as our approach enables us to directly observe and analyze the PM’s actions, it can overcome the empirical challenges encountered by previous research owing to the unobservability of the PM’s behavior. Our empirical observations are primarily consistent with recent theoretical developments regarding this topic. Despite limitations, by employing a completely novel method and dataset, our approach enhances our understanding and provides new insights into this classic issue.
Out-of-distribution (OOD) detection is crucial to safety-critical machine learning applications and has been extensively studied. While recent studies have predominantly focused on classifier-based methods, research on deep generative model (DGM)-based methods have lagged relatively. This disparity may be attributed to a perplexing phenomenon: DGMs often assign higher likelihoods to unknown OOD inputs than to their known training data. This paper focuses on explaining the underlying mechanism of this phenomenon. We propose a hypothesis that less complex images concentrate in high-density regions in the latent space, resulting in a higher likelihood assignment in the Normalizing Flow (NF). We experimentally demonstrate its validity for five NF architectures, concluding that their likelihood is untrustworthy. Additionally, we show that this problem can be alleviated by treating image complexity as an independent variable. Finally, we provide evidence of the potential applicability of our hypothesis in another DGM, PixelCNN++.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.