Revisiting Natural Gradient for Deep Networks
read the original abstract
We evaluate natural gradient, an algorithm originally proposed in Amari (1997), for learning deep models. The contributions of this paper are as follows. We show the connection between natural gradient and three other recently proposed methods for training deep models: Hessian-Free (Martens, 2010), Krylov Subspace Descent (Vinyals and Povey, 2012) and TONGA (Le Roux et al., 2008). We describe how one can use unlabeled data to improve the generalization error obtained by natural gradient and empirically evaluate the robustness of the algorithm to the ordering of the training set compared to stochastic gradient descent. Finally we extend natural gradient to incorporate second order information alongside the manifold information and provide a benchmark of the new algorithm using a truncated Newton approach for inverting the metric matrix instead of using a diagonal approximation of it.
This paper has not been read by Pith yet.
Forward citations
Cited by 14 Pith papers
-
Adam: A Method for Stochastic Optimization
A first-order stochastic optimizer that maintains bias-corrected exponential moving averages of the gradient and its square, dividing the former by the square root of the latter to set per-parameter step sizes.
-
MAOAM: Unified Object and Material Selection with Vision-Language Models
MAOAM unifies object and material selection via a VLM with segmentation head, supporting text and click interactions through multi-task training on VLM-generated material data.
-
Layerwise LQR for Geometry-Aware Optimization of Deep Networks
Steepest descent under divergence-induced quadratic models equals an LQR problem, enabling learning of diagonal or Kronecker-factored inverse preconditioners via a global layerwise objective for scalable geometry-awar...
-
Learning reveals invisible structure in low-rank RNNs
Learning in low-rank RNNs reduces to an exact low-dimensional ODE system in overlap space, where loss-invisible overlaps encode training history without affecting function.
-
Egalitarian Gradient Descent: A Simple Approach to Accelerated Grokking
EGD equalizes gradient speeds across singular directions, eliminating or shortening grokking plateaus on modular addition and sparse parity problems.
-
Double Preconditioning (DoPr): Optimization for Test-Time Performance, not Validation Loss
Double preconditioning (DoPr) improves downstream task performance in test-time feedback settings without consistent gains in validation loss.
-
Multiparameter Maximum Information States for Coherent Diffraction Measurements
Extends maximum information states to multiparameter Fisher matrix optimization for coherent diffraction using scattering matrix, with nuisance parameter strategies, numerically tested on 2D coupled dipoles.
-
Solving Classical and Quantum Spin Glasses with Deep Boltzmann Quantum States
Deep Boltzmann Quantum States with natural-gradient optimization and annealing-like training match exact or best-known solutions for large infinite-range Ising spin glasses and solve job shop scheduling instances.
-
Learning in the Fisher Subspace: A Guided Initialization for LoRA Fine-Tuning
Fisher information from target data provides a better criterion than weight geometry for choosing LoRA subspaces, yielding consistent performance gains on downstream tasks.
-
Low Rank Based Subspace Inference for the Laplace Approximation of Bayesian Neural Networks
Derives optimal low-rank subspace for Laplace approx in BNNs, provides scalable outperforming version, and new comparison metric.
-
Local Guidance, Global Impact: Gaussian-Reshaped Trust Region Unlocks Behavior Transitions
GTR introduces a bounded non-monotonic Gaussian trust region and Mixture Gaussian Anchor to enable effective behavior transitions in non-stationary RL where standard PPO fails.
-
Second-Order Multi-Level Variance Correction for Modality Competition in Multimodal Models
Introduces ML-FOP-SOAP optimizer using Fisher-Orthogonal Projection and hierarchical folding to mitigate modality competition in multimodal autoregressive training, reporting gains over AdamW on Janus and Emu3.
-
Intelligence Inertia: Physical Isomorphism and Applications
Intelligence Inertia models the computational resistance to structural change in neural networks via a heuristic relativistic analogy, yielding a J-shaped cost curve that diverges from classical approximations.
-
Learning in the Fisher Subspace: A Guided Initialization for LoRA Fine-Tuning
Fisher information from the target data distribution supplies a task-dependent criterion for selecting LoRA directions that outperforms weight-magnitude heuristics.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.