Implicit Bias of Spectral Descent and Muon on Multiclass Separable Data
This provides a foundational theoretical framework for understanding how different optimization algorithms affect generalization in overparameterized models, which is incremental but clarifies biases in multi-class settings.
The paper tackles the problem of characterizing implicit optimization bias in gradient-based methods for multi-class linear classification, proving that p-norm normalized steepest descent and momentum steepest descent converge to max-margin solutions with respect to the classifier matrix's p-norm, with established convergence rates.
Different gradient-based methods for optimizing overparameterized models can all achieve zero training error yet converge to distinctly different solutions inducing different generalization properties. We provide the first complete characterization of implicit optimization bias for p-norm normalized steepest descent (NSD) and momentum steepest descent (NMD) algorithms in multi-class linear classification with cross-entropy loss. Our key theoretical contribution is proving that these algorithms converge to solutions maximizing the margin with respect to the classifier matrix's p-norm, with established convergence rates. These results encompass important special cases including Spectral Descent and Muon, which we show converge to max-margin solutions with respect to the spectral norm. A key insight of our contribution is that the analysis of general entry-wise and Schatten p-norms can be reduced to the analysis of NSD/NMD with max-norm by exploiting a natural ordering property between all p-norms relative to the max-norm and its dual sum-norm. For the specific case of descent with respect to the max-norm, we further extend our analysis to include preconditioning, showing that Adam converges to the matrix's max-norm solution. Our results demonstrate that the multi-class linear setting, which is inherently richer than the binary counterpart, provides the most transparent framework for studying implicit biases of matrix-parameter optimization algorithms.