3 papers
cs.LG2026
Loss Landscape Features That Make Adam Stall: Definitions, Estimators, and the Preconditioned Hessian View
Rodion Podorozhny
Across implicit-neural-representation (INR) architectures and analytic benchmarks we observe that a thoroughly tuned Adam (especially its learning rate (lr), e.g. in a hyperparamet…
cs.LG2026
Blockwise Stabilized Adaptive Cubic Regularization with Subsolvers via Recurrence
Rodion Podorozhny
Cubic regularized Newton methods have the optimal global rate, but a dense subproblem solve limits the feasible block size. Scalable Cubic Newton variants r…
cs.LG2026
Physics-Guided Transformer (PGT): Physics-Aware Attention Mechanism for PINNs
Ehsan Zeraatkar, Rodion Podorozhny, Jelena TeÅ¡iÄ
Reconstructing continuous physical fields from sparse, irregular observations is a central challenge in scientific machine learning, particularly for systems governed by partial di…