Time to Spectrum
"Time to Spectrum" research explores how analyzing the spectral properties of data – essentially, the distribution of its frequencies or eigenvalues – can improve various machine learning tasks. Current efforts focus on leveraging spectral information to enhance model training efficiency (e.g., through targeted training or dimensionality reduction), improve model explainability, and evaluate dataset quality for time series and other complex data. This approach holds significant promise for advancing diverse fields, from improving the performance and resource efficiency of large language models to enhancing medical image diagnostics and accelerating scientific simulations.
Papers
Inference time LLM alignment in single and multidomain preference spectrum
Sadat Shahriar, Zheng Qi, Nikolaos Pappas, Srikanth Doss, Monica Sunkara, Kishaloy Halder, Manuel Mager, Yassine Benajiba
A Random Matrix Theory Perspective on the Spectrum of Learned Features and Asymptotic Generalization Capabilities
Yatin Dandi, Luca Pesce, Hugo Cui, Florent Krzakala, Yue M. Lu, Bruno Loureiro
Single Parent Family: A Spectrum of Family Members from a Single Pre-Trained Foundation Model
Habib Hajimolahoseini, Mohammad Hassanpour, Foozhan Ataiefard, Boxing Chen, Yang Liu
Detecting Subtle Differences between Human and Model Languages Using Spectrum of Relative Likelihood
Yang Xu, Yu Wang, Hao An, Zhichen Liu, Yongyuan Li