One of the key predictions of the "WIMP" paradigm for Dark Matter (DM) is that DM particles can annihilate into charged particles. These annihilations will proceed in e. g. Galactic subhalos such as dwarf Galaxies or, as recently pointed out, high velocity clouds such as the "Smith Cloud". In this note, we focus on the radio emission associated with DM annihilations into electrons and positrons occurring in the Smith Cloud. The phenomenology of this emission is discussed in quite some detail. We argue that the uncertainties in the propagation can be captured by the typical diffusion-loss length parameter (Syrovatskii variable) but that the angle-integrated radio fluxes are independent of the propagation. We conclude that if the Smith Cloud is indeed dominated by DM, radio signals from DM annihilation stand out amongst other messengers. Furthermore, low frequencies such as the ones observed by e. g. the Low Frequency Array (LOFAR) and the next-generation Square Kilometre Array (SKA) are optimal for searches for DM in the Smith Cloud. As a practical application, we set conservative constraints on dark matter annihilation cross section using data of continuum radio emission from the Galaxy at 22 MHz and at 1.4 GHz. Stronger constraints could be reached by background subtraction, exploiting the profile and frequency dependence of the putative DM signal. We set stronger but tentative limits using the median noise in brightness temperature from the Green Bank Telescope and the LOFAR sensitivities.

No abstract

The Internet of Things (IoT) is coined by many different standards, protocols, and data formats that are often not compatible to each other. Thus, the integration of different heterogeneous (IoT) components into a uniform IoT setup can be a time-consuming manual task. This lacking interoperability between IoT components has been addressed with different approaches in the past. However, only very few of these approaches rely on Machine Learning techniques. In this work, we present a new way towards IoT interoperability based on Deep Reinforcement Learning (DRL). In detail, we demonstrate that DRL algorithms, which use network architectures inspired by Natural Language Processing (NLP), can be applied to learn to control an environment by merely taking raw JSON or XML structures, which reflect the current state of the environment, as input. Applied to IoT setups, where the current state of a component is often reflected by features embedded into JSON or XML structures and exchanged via messages, our NLP DRL approach eliminates the need for feature engineering and manually written code for pre-processing of data, feature extraction, and decision making.