Theoretical Analysis
Theoretical analysis in machine learning currently focuses on rigorously understanding the behavior and performance of various models and algorithms, aiming to explain observed phenomena and improve their design. Research emphasizes analyzing generalization capabilities, convergence properties, and the impact of factors like model size, data characteristics, and training procedures, often within frameworks like linear regression, Gaussian processes, and specific neural network architectures (e.g., Transformers, GNNs, DeepONets). These analyses provide crucial insights for developing more efficient, robust, and reliable machine learning systems, impacting fields ranging from robotics and natural language processing to graph analysis and federated learning.
Papers
Double Descent Meets Out-of-Distribution Detection: Theoretical Insights and Empirical Analysis on the role of model complexity
Mouïn Ben Ammar, David Brellmann, Arturo Mendoza, Antoine Manzanera, Gianni Franchi
Data Augmentations Go Beyond Encoding Invariances: A Theoretical Study on Self-Supervised Learning
Shlomo Libo Feigin, Maximilian Fleissner, Debarghya Ghoshdastidar