arXiv subjects

Marc Brinkmann

Publications and source records attributed to Marc Brinkmann.

6 recordsLinked to original sources

arXiv2025

Verde: Verification via Refereed Delegation for Machine Learning Programs

Machine learning programs, such as those performing inference, fine-tuning, and training of LLMs, are commonly delegated to untrusted compute providers. To provide correctness guarantees for the client, we propose adapting the cryptographic notion of refereed delegation to the machine learning setting. This approach enables a computationally limited client to delegate a program to multiple untrusted compute providers, with a guarantee of obtaining the correct result if at least one of them is honest. Refereed delegation of ML programs poses two technical hurdles: (1) an arbitration protocol to resolve disputes when compute providers disagree on the output, and (2) the ability to bitwise reproduce ML programs across different hardware setups, For (1), we design Verde, a dispute arbitration protocol that efficiently handles the large scale and graph-based computational model of modern ML programs. For (2), we build RepOps (Reproducible Operators), a library that eliminates hardware "non-determinism" by controlling the order of floating point operations performed on all hardware. Our implementation shows that refereed delegation achieves both strong guarantees for clients and practical overheads for compute providers.

cs.LG↗

arXiv2024

GEO600 beam splitter thermal compensation system: new design and commissioning

Gravitational waves have revolutionised the field of astronomy by providing scientists with a new way to observe the universe and gain a better understanding of exotic objects like black holes. Several large-scale laser interferometric gravitational wave detectors (GWDs) have been constructed worldwide, with a focus on achieving the best sensitivity possible. However, in order for a detector to operate at its intended sensitivity, its optics must be free from imperfections such as thermal lensing effects. In the GEO\,600 gravitational wave detector, the beam splitter (BS) experiences a significant thermal lensing effect due to the high power build-up in the Power Recycling Cavity (PRC) combined with a very small beam waist. This causes the fundamental mode to be converted into higher order modes (HOMs), subsequently impacting the detector's performance. To address this issue, the GEO\,600 detector is equipped with a thermal compensation system (TCS) applied to the BS. This involves projecting a spatially tunable heating pattern through an optical system onto the beam splitter. The main objective of the TCS is to counteract the thermal lens at the BS and restore the detector to its ideal operating condition. This paper presents the new beam splitter TCS in GEO\,600, its commissioning, and its effect on strain sensitivity. It also outlines the planned upgrade to further enhance the performance of the TCS.

astro-ph.IM↗

arXiv2023

First demonstration of neural sensing and control in a kilometer-scale gravitational wave observatory

Suspended optics in gravitational wave (GW) observatories are susceptible to alignment perturbations, particularly slow drifts over time, due to variations in temperature and seismic levels. Such misalignments affect the coupling of the incident laser beam into the optical cavities, degrade both circulating power and optomechanical photon squeezing and thus decrease the astrophysical sensitivity to merging binaries. Traditional alignment techniques involve differential wavefront sensing using multiple quadrant photodiodes but are often restricted in bandwidth and are limited by the sensing noise. We present the first-ever successful implementation of neural network-based sensing and control at a gravitational wave observatory and demonstrate low-frequency control of the signal recycling mirror at the GEO 600 detector. Alignment information for three critical optics is simultaneously extracted from the interferometric dark port camera images via a CNN-LSTM network architecture and is then used for MIMO control using soft actor-critic-based deep reinforcement learning. Overall sensitivity improvement achieved using our scheme demonstrates deep learning's capabilities as a viable tool for real-time sensing and control for current and next-generation GW interferometers.

physics.ins-det↗

arXiv2021

Direct limits for scalar field dark matter from a gravitational-wave detector

The nature of dark matter remains unknown to date; several candidate particles are being considered in a dynamically changing research landscape. Scalar field dark matter is a prominent option that is being explored with precision instruments, such as atomic clocks and optical cavities. Here we report on the first direct search for scalar field dark matter utilising a gravitational-wave detector, which operates beyond the quantum shot-noise limit. We set new upper limits for the coupling constants of scalar field dark matter as a function of its mass, by excluding the presence of signals that would be produced through the direct coupling of this dark matter to the beamsplitter of the GEO$\,$600 interferometer. The new constraints improve upon bounds from previous direct searches by more than six orders of magnitude, and are in some cases more stringent than limits obtained in tests of the equivalence principle by up to four orders of magnitude. Our work demonstrates that scalar field dark matter can be probed or constrained with direct searches using gravitational-wave detectors, and highlights the potential of quantum-enhanced interferometry for dark matter detection.

gr-qc↗

arXiv2020

Bilinear noise subtraction at the GEO 600 observatory

Longitudinal control signals used to keep gravitational wave detectors at a stable operating point are often affected by modulations from test mass misalignments leading to an elevated noise floor ranging from 50 to 500 Hz. Nonstationary noise of this kind results in modulation sidebands and increases the number of glitches observed in the calibrated strain data. These artifacts ultimately affect the data quality and decrease the efficiency of the data analysis pipelines looking for astrophysical signals from continuous waves as well as the transient events. In this work, we develop a scheme to subtract one such bilinear noise from the gravitational wave strain data and demonstrate it at the GEO 600 observatory. We estimate the coupling by making use of narrow-band signal injections that are already in place for noise projection purposes and construct a coherent bilinear signal by a two-stage system identification process. We improve upon the existing filter design techniques by employing a Bayesian adaptive directed search strategy that optimizes across the several key parameters that affect the accuracy of the estimated model. The scheme takes into account the possible nonstationarities in the coupling by periodically updating the involved filter coefficients. The resulting postoffline subtraction leads to a suppression of modulation sidebands around the calibration lines along with a broadband reduction of the midfrequency noise floor. The observed increase in the astrophysical range and a reduction in the occurrence of nonastrophysical transients suggest that the above method is a viable data cleaning technique for current and future generation gravitational wave observatories.

gr-qc↗

arXiv2020

First demonstration of 6 dB quantum noise reduction in a kilometer scale gravitational wave observatory

Photon shot noise, arising from the quantum-mechanical nature of the light, currently limits the sensitivity of all the gravitational wave observatories at frequencies above one kilohertz. We report a successful application of squeezed vacuum states of light at the GEO\,600 observatory and demonstrate for the first time a reduction of quantum noise up to $6.03 \pm 0.02$ dB in a kilometer-scale interferometer. This is equivalent at high frequencies to increasing the laser power circulating in the interferometer by a factor of four. Achieving this milestone, a key goal for the upgrades of the advanced detectors, required a better understanding of the noise sources and losses, and implementation of robust control schemes to mitigate their contributions. In particular, we address the optical losses from beam propagation, phase noise from the squeezing ellipse, and backscattered light from the squeezed light source. The expertise gained from this work carried out at GEO 600 provides insight towards the implementation of 10 dB of squeezing envisioned for third-generation gravitational wave detectors.

physics.ins-det↗