Empirical Analysis of the Hessian of Over-Parametrized Neural Networks
read the original abstract
We study the properties of common loss surfaces through their Hessian matrix. In particular, in the context of deep learning, we empirically show that the spectrum of the Hessian is composed of two parts: (1) the bulk centered near zero, (2) and outliers away from the bulk. We present numerical evidence and mathematical justifications to the following conjectures laid out by Sagun et al. (2016): Fixing data, increasing the number of parameters merely scales the bulk of the spectrum; fixing the dimension and changing the data (for instance adding more clusters or making the data less separable) only affects the outliers. We believe that our observations have striking implications for non-convex optimization in high dimensions. First, the flatness of such landscapes (which can be measured by the singularity of the Hessian) implies that classical notions of basins of attraction may be quite misleading. And that the discussion of wide/narrow basins may be in need of a new perspective around over-parametrization and redundancy that are able to create large connected components at the bottom of the landscape. Second, the dependence of small number of large eigenvalues to the data distribution can be linked to the spectrum of the covariance matrix of gradients of model outputs. With this in mind, we may reevaluate the connections within the data-architecture-algorithm framework of a model, hoping that it would shed light into the geometry of high-dimensional and non-convex spaces in modern applications. In particular, we present a case that links the two observations: small and large batch gradient descent appear to converge to different basins of attraction but we show that they are in fact connected through their flat region and so belong to the same basin.
This paper has not been read by Pith yet.
Forward citations
Cited by 20 Pith papers
-
When and Why SignSGD Outperforms SGD: A Theoretical Study Based on $\ell_1$-norm Lower Bounds
SignSGD provably beats SGD by a factor of d under sparse noise via matched ℓ1-norm upper and lower bounds, with an equivalent result for Muon on matrices, and this predicts faster GPT-2 pretraining.
-
AMUSE: Anytime Muon with Stable Gradient Evaluation
AMUSE is a new optimizer integrating Muon orthogonalization with Schedule-Free averaging via adaptive interpolation for schedule-free anytime training that improves Pareto frontiers on vision and LLM tasks.
-
Backdoor Channels Hidden in Latent Space: Cryptographic Undetectability in Modern Neural Networks
Backdoors can be realized as statistically natural latent directions in modern neural networks, achieving high attack success with negligible clean accuracy loss and resisting existing defenses.
-
Select-then-differentiate: Solving Bilevel Optimization with Manifold Lower-level Solution Sets
Optimistic bilevel optimization with manifold lower-level minimizers is differentiable if the optimistic selection is unique, yielding a pseudoinverse hyper-gradient and a convergent HG-MS algorithm whose rate depends...
-
Hessian Surgery: Class-Targeted Post-Hoc Rebalancing via Hessian Spike Perturbation
Spectral Surgery uses a sensitivity matrix and constrained optimization to perturb weights along Hessian spike eigenvectors, rebalancing per-class accuracy on CIFAR-10 and ISIC-2019 without retraining.
-
Hessian Surgery: Class-Targeted Post-Hoc Rebalancing via Hessian Spike Perturbation
Hessian Surgery perturbs trained model weights along Hessian spike eigenvectors via a sensitivity matrix and constrained optimization to rebalance per-class accuracy on CIFAR-10 and ISIC-2019 without retraining.
-
Fast Gauss-Newton for Multiclass Cross-Entropy
FGN is a positive semidefinite under-approximation of the multiclass GGN obtained by exact decomposition into true-vs-rest and within-competitor terms, exact for binary classification and implemented via matrix-free c...
-
The Lifecycle of the Spectral Edge: From Gradient Learning to Weight-Decay Compression
The spectral edge transitions from a gradient-driven functional direction before grokking to a perturbation-flat, ablation-critical compression axis at grokking, forming three universality classes predicted by a gap f...
-
Depth, Not Data: An Analysis of Hessian Spectral Bifurcation
Deep linear networks with balanced data covariance exhibit Hessian spectral bifurcation whose dominant-to-bulk eigenvalue ratio scales linearly with depth.
-
Weight-space symmetry in deep networks gives rise to permutation saddles, connected by equal-loss valleys across the loss landscape
Permutation symmetries generate permutation saddles and equal-loss valleys linking equivalent global minima, yielding a lower bound on symmetry-induced critical points.
-
Selection Plateau and a Sparsity-Dependent Hierarchy of Pruning Features
All rank-monotone pruning scorers converge to identical accuracy at fixed sparsity, but non-monotone features with sparsity-dependent complexity can escape this plateau, as shown by the SICS hypothesis on ViT-Small/CIFAR-10.
-
Quantum Tilted Loss in Variational Optimization: Theory and Applications
QTL unifies expectation-value minimization with CVaR and Gibbs heuristics under one tunable operator, amplifying gradients in structured cases while preserving global minima and shifting the bottleneck to measurement ...
-
Generalization at the Edge of Stability
Training at the edge of stability causes neural network optimizers to converge on fractal attractors whose effective dimension, measured via a new sharpness dimension from the Hessian spectrum, bounds generalization e...
-
Exploiting Correlations in Federated Learning: Opportunities and Practical Limitations
A correlation-based taxonomy unifies existing FL compression methods, experiments show correlation strengths vary by task and architecture, and adaptive mode-switching designs are proposed to exploit this.
-
Escape dynamics and implicit bias of one-pass SGD in overparameterized quadratic networks
In overparameterized quadratic networks, one-pass SGD escapes generalization plateaus only modestly faster and selects the initialization-closest zero-loss solution due to a conserved quantity in the overlap ODEs.
-
Spectral Edge Dynamics: An Analytical-Empirical Study of Phase Transitions in Neural Network Training
Spectral gaps in the Gram matrix of parameter updates control phase transitions such as grokking in neural network training.
-
CrispEdit: Low-Curvature Projections for Scalable Non-Destructive LLM Editing
CrispEdit edits LLMs via low-curvature projections using Bregman divergence and K-FAC approximations, achieving high edit success with under 1% average capability degradation.
-
RMNP: Row-Momentum Normalized Preconditioning for Scalable Matrix-Based Optimization
RMNP preconditions matrix updates via row-wise L2 normalization instead of Newton-Schulz iteration, reducing complexity to O(mn) while matching Muon's non-convex convergence rate and empirical performance.
-
On the Convergence Analysis of Muon
Convergence analysis shows Muon outperforms gradient descent by exploiting low-rank structure in neural network Hessians.
-
Hessian based analysis of SGD for Deep Nets: Dynamics and Generalization
Provides Hessian-based theoretical characterizations of SGD dynamics and a scale-invariant generalization bound for deep nets, backed by experiments on synthetic data, MNIST, and CIFAR-10.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.