WaveDetect: Robust Framework for Machine-Generated Text Detection via Wavelet Transform

WaveDetect is a novel framework for detecting machine-generated text by reformulating the detection task as a signal processing problem using Continuous Wavelet Transform to create learnable spectral representations. It addresses key challenges such as adversarial perturbations and cross-domain shifts, demonstrating state-of-the-art performance on datasets like RAID, EvoBench, and Domain-Shift. This approach highlights the effectiveness of spectral analysis in enhancing the robustness and accuracy of detection methods for LLM-generated texts, which is critical for practitioners working with AI-generated content.