Beka Modrekiladze

Beka Modrekiladze

Generative Adversarial Networks (GANs) have demonstrated remarkable advancements in generative modeling; however, their training is often resource-intensive, requiring extensive computational time and hundreds of thousands of epochs. This paper proposes a novel optimization approach that transforms the training process by operating within a dual space of the initial data using invertible mappings, specifically autoencoders. By training GANs on the encoded representations in the dual space, which encapsulate the most salient features of the data, the generative process becomes significantly more efficient and potentially reveals underlying patterns beyond human recognition. This approach not only enhances training speed and resource usage but also explores the philosophical question of whether models can generate insights that transcend the human intelligence while being limited by the human-generated data.

Beka Modrekiladze

The detection of gravitational waves has opened unparalleled opportunities for observing the universe, particularly through the study of black hole inspirals. These events serve as unique laboratories to explore the laws of physics under conditions of extreme energies. However, significant noise in gravitational wave (GW) data from observatories such as Advanced LIGO and Virgo poses major challenges in signal identification. Traditional noise suppression methods often fall short in fully addressing the non-Gaussian effects in the data, including the fluctuations in noise power spectral density (PSD) over short time intervals. These challenges have led to the exploration of an AI approach that, while overcoming previous obstacles, introduced its own challenges, such as scalability, reliability issues, and the vanishing gradient problem. Our approach addresses these issues through a simplified architecture. To compensate for the potential limitations of a simpler model, we have developed a novel training methodology that enables it to accurately detect gravitational waves amidst highly complex noise. Employing this strategy, our model achieves over 99% accuracy in non-white noise scenarios and shows remarkable adaptability to changing noise PSD conditions. By leveraging the principles of transfer learning, our model quickly adapts to new noise profiles with just a few epochs of fine-tuning, facilitating real-time applications in dynamically changing noise environments.