A geometric local landscape model explains PTQ basin-crossing failure at aggressive bitwidths and proves finite-time QAT recovery via straight-through estimator gradient bias under quantizer-compatibility assumptions.
Beyond the quadratic approximation: The multiscale structure of neural network loss landscapes
2 Pith papers cite this work. Polarity classification is still indexing.
2
Pith papers citing it
years
2026 2verdicts
UNVERDICTED 2representative citing papers
Heat kernel regularization ensures the regularized Hessian stays asymptotically nondegenerate near nonsmooth minimizers of the form |x|^a, making the continuation equation locally solvable for small t.
citing papers explorer
-
Understanding Quantization-Aware Training: Gradients at Quantized Weights Bias to the Low-Loss Basin
A geometric local landscape model explains PTQ basin-crossing failure at aggressive bitwidths and proves finite-time QAT recovery via straight-through estimator gradient bias under quantizer-compatibility assumptions.
-
From Nonsmooth Minima to Smooth Branches via Heat Kernel Regularization
Heat kernel regularization ensures the regularized Hessian stays asymptotically nondegenerate near nonsmooth minimizers of the form |x|^a, making the continuation equation locally solvable for small t.