Dynamical Implicit Neural Representations

2025/11/26 by Park, Yesom, Kan, Kelvin, Flynn, Thomas +4
Computer Science · Engineering · Neuroscience · #FOS: Computer and information sciences #Face Recognition and Perception #Ferroelectric and Negative Capacitance Devices #Generative Adversarial Networks and Image Synthesis #Machine Learning (cs.LG)

paper · doi:10.48550/arxiv.2511.21787

openalex publication_date 2025/11/26 · openalex created_date 2025/12/03 · openalex updated_date 2026/07/28

Abstract

Implicit Neural Representations (INRs) provide a powerful continuous framework for modeling complex visual and geometric signals, but spectral bias remains a fundamental challenge, limiting their ability to capture high-frequency details. Orthogonal to existing remedy strategies, we introduce Dynamical Implicit Neural Representations (DINR), a new INR modeling framework that treats feature evolution as a continuous-time dynamical system rather than a discrete stack of layers. This dynamical formulation mitigates spectral bias by enabling richer, more adaptive frequency representations through continuous feature evolution. Theoretical analysis based on Rademacher complexity and the Neural Tangent Kernel demonstrates that DINR enhances expressivity and improves training dynamics. Moreover, regularizing the complexity of the underlying dynamics provides a principled way to balance expressivity and generalization. Extensive experiments on image representation, field reconstruction, and data compression confirm that DINR delivers more stable convergence, higher signal fidelity, and stronger generalization than conventional static INRs.

Citations

• Optimal Control for Transformer Architectures: Enhancing Generalization, Robustness and Efficiency
• End-to-End Implicit Neural Representations for Classification
• Neural Experts: Mixture of Experts for Implicit Neural Representations
• Inductive Gradient Adjustment For Spectral Bias In Implicit Neural Representations
• Towards Croppable Implicit Neural Representations
• Grounding and Enhancing Grid-based Models for Neural Fields
• From Activation to Initialization: Scaling Insights for Optimizing Neural Fields
• Rademacher Complexity of Neural ODEs via Chen-Fliess Series
• Continuous Field Reconstruction from Sparse Observations with Implicit Neural Networks
• Sequence Length Independent Norm-Based Generalization Bounds for Transformers
• SuperBench: A Super-Resolution Benchmark Dataset for Scientific Machine Learning
• Generalization bounds for neural ordinary differential equations and deep residual networks
• Regularize implicit neural representation by itself
• Towards Scalable Neural Representation for Diverse Videos
• WIRE: Wavelet Implicit Neural Representations
• Taming Hyperparameter Tuning in Continuous Normalizing Flows Using the JKO Scheme
• Bounding the Rademacher Complexity of Fourier neural operators
• Sobolev Training for Implicit Neural Representations with Approximated Image Derivatives
• Instant Neural Graphics Primitives with a Multiresolution Hash Encoding
• A Structured Dictionary Perspective on Implicit Neural Representations
• Beyond Periodicity: Towards a Unifying Framework for Activations in Coordinate-MLPs
• Neural Fields in Visual Computing and Beyond
• Neural Fields in Visual Computing and Beyond
• Learning Quantile Functions without Quantile Crossing for Distribution-free Time Series Forecasting
• Learning Continuous Image Representation with Local Implicit Image Function
• Learned Initializations for Optimizing Coordinate-Based Neural Representations
• Tensor Programs II: Neural Tangent Kernel for Any Architecture
• Fourier Features Let Networks Learn High Frequency Functions in Low Dimensional Domains
• Infinite attention: NNGP and NTK for deep attention networks
• Implicit Neural Representations with Periodic Activation Functions
• The ERA5 global reanalysis
• Dissecting Neural ODEs
• How to train your neural ODE: the world of Jacobian and kinetic regularization
• Towards Understanding the Spectral Bias of Deep Learning
• Finite Depth and Width Corrections to the Neural Tangent Kernel
• Potential Flow Generator with L₂ Optimal Transport Regularity for Generative Models
• Data-dependent Sample Complexity of Deep Neural Networks via Lipschitz Augmentation
• On Exact Computation with an Infinitely Wide Neural Net
• DeepSDF: Learning Continuous Signed Distance Functions for Shape Representation
• Occupancy Networks: Learning 3D Reconstruction in Function Space
• Gradient Descent Provably Optimizes Over-parameterized Neural Networks
• Rademacher Complexity for Adversarially Robust Generalization
• FFJORD: Free-form Continuous Dynamics for Scalable Reversible Generative Models
• Neural Ordinary Differential Equations
• Understanding Batch Normalization
• Stronger generalization bounds for deep nets via a compression approach
• Deep Neural Networks as Gaussian Processes
• The Implicit Bias of Gradient Descent on Separable Data
• Spectrally-normalized margin bounds for neural networks
• On the Expressive Power of Deep Neural Networks
• A vector-contraction inequality for Rademacher complexities
• Deep Residual Learning for Image Recognition
• Adam: A Method for Stochastic Optimization

Related

• COSMO-INR: Complex Sinusoidal Modulation for Implicit Neural Representations
• I-INR: Iterative Implicit Neural Representations
• FreSh: Frequency Shifting for Accelerated Neural Representation Learning
• High-Frequency First: A Two-Stage Approach for Improving Image INR
• NSTR: Neural Spectral Transport Representation for Space-Varying Frequency Fields