TIDMAD: Time Series Dataset for Discovering Dark Matter with AI Denoising

TL;DR

TIDMAD releases an ultra-long ABRACADABRA time-series dataset with ground-truth signal injections to train and benchmark AI-based denoising for axion dark matter searches. It pairs training/validation data with a science set and provides a physics-focused benchmark (the Brazil-band TS limit) to translate denoising performance into improved dark matter limits, using $f_a = m_a/2\pi$ and a ground-truth signal model $J_{\mathrm{eff}} = g_{a\gamma\gamma} \sqrt{2 \rho_{\mathrm{DM}}} \mathbf{B}_0 \cos(m_a t)$. Eight denoising methods are evaluated, with deep-learning architectures substantially outperforming traditional filters, culminating in a best-performing FC Net that enhances sensitivity by 1–2 orders of magnitude across $m_a$ relative to the raw data. While hardware constraints prevent surpassing ABRA-10cm Run 3 limits in this dataset, the approach demonstrates a clear path to broader discovery potential and provides a foundation for generalizing time-series denoising to other wave-like dark matter searches and physics experiments. The release thus accelerates collaboration between ML researchers and the particle-astrophysics community, enabling rapid translation from denoising proficiency to publishable physics results.

Abstract

Dark matter makes up approximately 85% of total matter in our universe, yet it has never been directly observed in any laboratory on Earth. The origin of dark matter is one of the most important questions in contemporary physics, and a convincing detection of dark matter would be a Nobel-Prize-level breakthrough in fundamental science. The ABRACADABRA experiment was specifically designed to search for dark matter. Although it has not yet made a discovery, ABRACADABRA has produced several dark matter search results widely endorsed by the physics community. The experiment generates ultra-long time-series data at a rate of 10 million samples per second, where the dark matter signal would manifest itself as a sinusoidal oscillation mode within the ultra-long time series. In this paper, we present the TIDMAD -- a comprehensive data release from the ABRACADABRA experiment including three key components: an ultra-long time series dataset divided into training, validation, and science subsets; a carefully-designed denoising score for direct model benchmarking; and a complete analysis framework which produces a community-standard dark matter search result suitable for publication as a physics paper. This data release enables core AI algorithms to extract the dark matter signal and produce real physics results thereby advancing fundamental science. The data downloading and associated analysis scripts are available at https://github.com/jessicafry/TIDMAD

Figures (8)

• Figure 1: Left to right: ABRA-10cm dilution fridge with outer vacuum cans on; Coldest stage of ABRA-10cm fridge above shielded 1T superconducting torroidal magnet; Interior of ABRA-10cm magnet including pickup and calibration loop wires in the center of the magnet; Effective circuit diagram for both dark matter and injected signals.
• Figure 2: 10-millisecond snapshot of the time series in TIDMAD training dataset compared to the power spectral density of the same data snapshot.
• Figure 3: TIDMAD data flow explaining how data splits (green) result in benchmarks (blue). Rectangles correspond to provided scripts. Left to right, top to bottom the scripts are train.py, inference.py, benchmark.py, process_science_data.py, brazilband.py
• Figure 4: Left: The color bar represents the denoising score for 20s of raw data with added Gaussian noise showing that noisier data results in a lower score. Right: Denoising score and $\Lambda$ for 20s of raw data with added Gaussian noise of variable noise amplitude. The exponential behavior of $\Lambda$ can clearly be seen in contrast to the smooth linearity of the denoising score.
• Figure 5: Plotted alongside the present, state-of-the-art axion dark matter limits (grey) are the $2\sigma$ exclusion limits for ABRA-10cm Run 3 (pink) abra21, ABRA-TIDMAD Raw limit (red), and ABRA-TIDMAD Denoised limit from the trained FC net (maroon). The ABRA-TIDMAD result does not surpass ABRA-10cm Run 3 due to reduced geometric coupling and shorter data taking duration in the TIDMAD configuration. Shaded regions correspond to pairings of dark matter model parameters ($m_a$, $g_{a\gamma\gamma}$) that are ruled out by the specified experiment, and the bright yellow region indicates theoretical predictions (See Appendix \ref{['dmexp']} for details). The plotting script is modified from the publicly available AxionLimits repository ohare.