Rethinking Parameter Counting in Deep Models: Effective Dimensionality Revisited

Abstract

Neural networks appear to have mysterious generalization properties when using parameter counting as a proxy for complexity. Indeed, neural networks often have many more parameters than there are data points, yet still provide good generalization performance. Moreover, when we measure generalization as a function of parameters, we see double descent behaviour, where the test error decreases, increases, and then again decreases. We show that many of these properties become understandable when viewed through the lens of effective dimensionality, which measures the dimensionality of the parameter space determined by the data. We relate effective dimensionality to posterior contraction in Bayesian deep learning, model selection, width-depth tradeoffs, double descent, and functional diversity in loss surfaces, leading to a richer understanding of the interplay between parameters and functions in deep models. We also show that effective dimensionality compares favourably to alternative norm- and flatness- based generalization measures.

References in corpus (17)

• Reconciling modern machine learning practice and the bias-variance trade-off
• Neural Tangent Kernel: Convergence and Generalization in Neural Networks
• Benign Overfitting in Linear Regression
• GPyTorch: Blackbox Matrix-Matrix Gaussian Process Inference with GPU Acceleration
• Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer
• Composable Effects for Flexible and Accelerated Probabilistic Programming in NumPyro
• The generalization error of random features regression: Precise asymptotics and double descent curve
• Bayesian Deep Learning and a Probabilistic Perspective of Generalization
• A Simple Baseline for Bayesian Uncertainty in Deep Learning
• Fantastic Generalization Measures and Where to Find Them
• Functional Variational Bayesian Neural Networks
• Eigenvalues of the Hessian in Deep Learning: Singularity and Beyond
• An Investigation into Neural Net Optimization via Hessian Eigenvalue Density
• Non-Vacuous Generalization Bounds at the ImageNet Scale: A PAC-Bayesian Compression Approach
• Measuring the Intrinsic Dimension of Objective Landscapes
• The Full Spectrum of Deepnet Hessians at Scale: Dynamics with SGD Training and Sample Size
• Understanding Generalization through Visualizations

Cited by in corpus (15)

• GShard: Scaling Giant Models with Conditional Computation and Automatic Sharding
• Quantum Machine Learning: from physics to software engineering
• Quantum Transfer Learning for Real-World, Small, and High-Dimensional Datasets
• Laplace Redux -- Effortless Bayesian Deep Learning
• Learning Invariances in Neural Networks
• Simplifying Hamiltonian and Lagrangian Neural Networks via Explicit Constraints
• Generalization bounds for deep learning
• Bayesian Deep Learning via Subnetwork Inference
• Robustness to Pruning Predicts Generalization in Deep Neural Networks
• Deep Learning is Singular, and That's Good
• Seq2Tens: An Efficient Representation of Sequences by Low-Rank Tensor Projections
• Loss Surface Simplexes for Mode Connecting Volumes and Fast Ensembling
• Why flatness does and does not correlate with generalization for deep neural networks
• Optimization Variance: Exploring Generalization Properties of DNNs
• Leveraging Active Subspaces to Capture Epistemic Model Uncertainty in Deep Generative Models for Molecular Design