Polynomial neural networks explore thresholds for maximum expressiveness.
problem Understanding the limits of polynomial neural networks' expressiveness.
method Introducing activation degree threshold to measure network expressiveness and proving its existence and upper bounds.
result Polynomial neural networks with equi-width architectures achieve the maximum expressiveness.
Review of neural network expressivity and architectures.
problem Understanding neural network expressivity across different architectures.
method Comprehensive overview of approximation results for various neural network types.
result Deep neural networks offer advantages over shallow ones for specific function classes.
Two new criteria help understand the advantage of deep neural networks.
problem Understanding the advantage of deepening neural networks.
method Proposed two new criteria to evaluate the expressivity of functions computable by deep neural networks.
result Increasing layers is more effective than increasing units in improving the expressivity of deep neural networks.
PNCs balance tractability and expressiveness in probabilistic modeling.
problem Balancing tractability and expressiveness in probabilistic models.
method Introduce probabilistic neural circuits (PNCs) as a mix of Bayesian networks and neural networks.
result PNCs are powerful function approximators.
Deep polynomial neural networks measure their expressiveness by the dimension of their functional space.
problem Measuring the expressiveness of deep polynomial neural networks.
method Analyzing the algebraic variety defined by the polynomial neural network's weights and activations.
result The dimension of the algebraic variety is a precise measure of the network's expressiveness.
We survey results on neural network expressivity described in "On the Expressive Power of Deep Neural Networks". The paper motivates and develops three natural measures of expressiveness, which all display an exponential dependence on the depth of the network. In fact, all of these measures are related to a fourth quan…
Stable unactivated neurons reduce expressiveness in ReLU networks.
problem Reducing expressiveness in ReLU neural networks due to stably unactivated neurons.
method Investigated the probability of neurons being stably unactivated in ReLU networks with symmetric weight and bias distributions.
result Proved the probability of a neuron being stably unactivated in the second hidden layer of a ReLU network.
Study the expressivity and training complexity of polynomial neural networks.
problem Understanding the expressivity and training complexity of polynomial neural networks.
method Use algebraic geometry to describe neuromanifolds and neurovarieties, analyzing their dimension and learning degree.
result Characterized the dimension and learning degree of neuromanifolds, providing geometric and complexity measures.
This paper explores how enforcing equivariance constraints limits neural network expressivity and proposes compensatory model size increases.
problem The impact of enforcing equivariance constraints on the expressive power of neural networks.
method Examined 2-layer ReLU networks, analyzed boundary hyperplanes and channel vectors, and constructed upper bounds on model size required for compensation.
result Enforcing equivariance constraints reduces the expressive power of neural networks, but this can be compensated by increasing model size.
The paper analyzes deep neural networks' expressivity and training, revealing critical expressivity issues.
problem Critical expressivity issues in deep neural networks.
method Quantitative analysis using Hilbert space and Hermite polynomials for feature mapping and activation function design.
result Deep neural networks evolve to the edge of chaos, but expressivity depends on overcoming convergence.
Survey on GNNs' power and limitations.
problem Theoretical limitations of GNNs.
method Comprehensive overview of GNNs and their variants.
result Provably powerful variants of GNNs.
Paper studies the expressivity of Convolutional Neural Networks (CNNs).
problem Understanding the expressivity of CNNs and their superiority in deep learning.
method Mathematical analysis of linear regions in one-layer and multi-layer ReLU CNNs.
result Deeper CNNs and CNNs have more expressivity per parameter than fully-connected NNs.
New MPNNs match 2-WL, faster distinguishing graphs.
problem Improving graph neural network expressiveness.
method Introducing ℓ-walk MPNNs and second-order GNNs. result Walk MPNNs match 2-WL and can distinguish graphs faster.
We investigate graph neural networks for multi-relational data.
problem Understanding and improving graph neural networks for multi-relational data.
method Aligning Relational GCN and Compositional GCN with the Weisfeiler-Leman test to understand their expressive power and introduce a new k-RN architecture. result The k-RN architecture overcomes the expressiveness limitations of Relational GCN and Compositional GCN. Improved bounds on neural network expressivity.
problem Understanding neural network expressivity and approximation capabilities.
method Improved bounds on the maximal number of linear regions of ReLU-networks.
result New insights into the expressivity of neural networks.
Study on HyperGNNs' expressiveness and structural generalization.
problem Understanding HyperGNNs' ability to solve graph problems and generalize to larger graphs.
method Fine-grained analysis of expressiveness and learning properties, with theoretical and empirical support.
result HyperGNNs can solve a hierarchy of graph problems defined by hyperparameters.
Express Wavenet reduces neural network parameters to 1% of standard networks.
problem Optical neural networks with high parameter count.
method Wavelet modulation, random shift wavelets, expressway structure.
result Express Wavenet achieves high accuracy with significantly fewer parameters.
Enhances deep neural networks for MRI reconstruction by increasing expressivity.
problem Balancing network complexity and performance in deep learning MRI reconstruction.
method Geometric approach using bootstrapping and subnetwork aggregation with attention module.
result Significant improvement in MRI reconstruction performance with minimal complexity increase.
This work explores the relation between depth and expressivity in neural networks.
problem Understanding the power of depth in neural networks and its relation to gradient-based optimization.
method Depth separation argument for distributions with fractal structure, proving that deep networks can express fine details efficiently but shallow ones cannot.
result The success of learning deep networks depends on whether the distribution can be well approximated by shallower networks.
Analytic expressions for deep neural network output under stochastic training.
problem Understanding the impact of noise and hyperparameters on deep neural network performance.
method Taylor expansion of network output to derive analytical expressions for weights and output.
result Noise in training affects generalization by preventing the output from fully converging on train data, but does not provide explicit regularization.
We develop a general duality between neural networks and compositional kernels, striving towards a better understanding of deep learning. We show that initial representations generated by common random initializations are sufficiently rich to express all functions in the dual kernel space. Hence, though the training ob…
FI-GNNs learn expressive node representations from sparse features.
problem Sparse and high-dimensional node features limit GNN performance.
method Plug-and-play GNN framework that highlights informative feature interactions.
result FI-GNNs learn highly expressive node representations on feature-sparse graphs.
Simplified NAS for GNN architectures improves efficiency and expressiveness.
problem Efficient and effective discovery of optimal GNN architectures.
method SNAG framework with a novel search space and reinforcement learning.
result SNAG framework outperforms human-designed and existing NAS methods.
The number of linear regions in neural networks is far below exponential, even after training.
problem Understanding the expressivity of neural networks, especially those with ReLU activation.
method Mathematical framework to count linear regions and measure their boundaries, analyzing networks at initialization and after training.
result The average number of regions along any one-dimensional subspace grows linearly in the total number of neurons, not exponentially.
Study shows shallow ReLU networks struggle with high-dimensional Lipschitz functions.
problem Expressing high-dimensional Lipschitz functions with shallow ReLU networks.
method Established lower bounds on shallow network complexity for polynomial approximation.
result Shallow ReLU networks suffer from the curse of dimensionality for Lipschitz functions.
Unified framework explains geometric properties of CNNs.
problem Understanding why encoder-decoder CNNs perform well.
method Unified mathematical framework based on recent neural network theories.
result Encoder-decoder CNNs are related to nonlinear basis representation using combinatorial convolution frames.
A new method to measure neural network expressiveness using tighter upper bounds.
problem Measuring the expressiveness of deep neural networks (DNNs).
method Proposes a new tighter upper bound for the number of linear regions in rectifier networks, using matrix computation.
result The proposed upper bound is tighter than existing ones and explains the performance improvements of skip connections and residual structures.
ISP improves GNN expressivity by stratifying nodes based on graph invariants.
problem Graph Neural Networks struggle with expressivity and structural heterogeneity.
method Invariant-Stratified Propagation (ISP) using ISP-WL and ISPGNN.
result ISP achieves enhanced expressivity beyond 1-WL, with theoretical guarantees and practical improvements.
A new aggregation strategy improves GNN performance and learning dynamics.
problem Improving expressivity and learning dynamics of GNNs.
method Proposes a variance-preserving aggregation function (VPA) for GNNs.
result VPA leads to increased predictive performance and improved learning dynamics.
Neural-guided symbolic regression uses asymptotic constraints to find unknown functions.
problem Finding unknown functions from data points with additional mathematical constraints.
method A neural network generates expressions with desired leading powers, and Monte Carlo Tree Search optimizes the expressions.
result The system effectively finds unknown functions outside the training set compared to existing methods.
Enhanced GNN with expanded attention window and partially random embeddings.
problem Limited expressivity of traditional GNNs in distinguishing non-isomorphic graphs.
method Graph attention network with expanding attention window and partially random initial embeddings. Head dropout for regularization.
result Improved ability to differentiate between non-isomorphic graphs.
Deep neural networks can approximate complex functions through repeated compositions of a fixed-size ReLU network.
problem Understanding the expressive power of deep neural networks through function compositions.
method Demonstrated the surprising expressive power of repeated compositions of a single fixed-size ReLU network.
result Repeated compositions of a single fixed-size ReLU network can approximate 1-Lipschitz continuous functions on [0,1]d with an error O(r−1/d). Finslerian graph neural networks recover nonlinear diffusion geometry
problem Graph neural networks on point clouds
method Estimates of the Finsler Laplacian
result Recovery of Finsler geometry
We propose a new approach to the problem of neural network expressivity, which seeks to characterize how structural properties of a neural network family affect the functions it is able to compute. Our approach is based on an interrelated set of measures of expressivity, unified by the novel notion of trajectory length…
New neural networks with variable time constants for better time-series prediction.
problem Improving neural network performance in time-series prediction.
method Constructing networks of linear dynamical systems modulated by nonlinear gates, using numerical differential equation solvers.
result Liquid Time-Constant Networks (LTCs) yield superior performance on time-series prediction tasks.
Neural networks are known to be a class of highly expressive functions able to fit even random input-output mappings with 100% accuracy. In this work, we present properties of neural networks that complement this aspect of expressivity. By using tools from Fourier analysis, we show that deep ReLU networks are biased…
New algorithm ensures global convergence in deep neural networks beyond NTK regime.
problem Existing global convergence guarantees do not apply to practical deep networks.
method Proposes an algorithm with global convergence guarantees under the expressivity condition.
result Algorithm ensures global convergence in practical settings beyond NTK regime.
New neural network class reduces VC dimension, leading to better generalization.
problem VC theory struggles with explaining small generalization errors in overparametrized neural networks.
method Developed hyperplane arrangement neural networks (HANNs) and used sample compression analysis.
result HANNs can have significantly smaller VC dimension than the number of weights, yet remain highly expressive.
A new complexity measure for neural networks improves upon classical methods.
problem Lack of a refined complexity measure for comparing different neural network architectures, especially permutation-invariant ones.
method Introduced an equivalence relation among linear functions and counted them relative to this relation.
result The new complexity measure clearly distinguishes between different models and increases exponentially with depth.
New graph neural networks can distinguish graphs better than previous models.
problem Graph isomorphism tests limit the expressive power of GNNs.
method Developed k-order invariant and equivariant graph neural networks, and a reduced 2-order network.
result A reduced 2-order network with a single quadratic operation has 3-WL expressiveness, surpassing message passing models.
GroupSort neural networks can approximate Lipschitz continuous functions.
problem Understanding and improving the expressive power of neural networks with Lipschitz constraints.
method Introduced and studied GroupSort neural networks with constraints on weights, proving their ability to approximate Lipschitz continuous functions.
result GroupSort networks can represent any Lipschitz continuous piecewise linear functions and are well-suited for approximating general Lipschitz continuous functions.
Unified method for deriving ridgelet transforms for various neural network architectures.
problem Deriving closed-form expressions for ridgelet transforms in modern neural network architectures.
method Unified Fourier slice method to derive ridgelet transforms for diverse neural network types.
result Systematic method to derive ridgelet transforms for various neural network architectures.
kth-order invariant graph networks are as powerful as kth-order WL in distinguishing graphs.
problem Measuring the expressive power of graph neural network formalisms.
method Considered kth-order invariant graph networks (k-IGNs) and compared their expressive power to kth-order WL.
result k-IGNs and k-WL are equally powerful in distinguishing graphs.
HDNNs can approximate any continuous function, proving their expressivity.
problem Lack of a comprehensive study on the expressivity of HDNNs.
method Discretization of Hamiltonian Neural Ordinary Differential Equations (HNN-ODEs).
result HDNNs can approximate any continuous function over a compact domain.
Graph Substructure Networks (GSN) improves GNN expressivity by counting subgraph isomorphisms.
problem Limited expressivity of GNNs in detecting and counting graph substructures.
method Topologically-aware message passing scheme based on substructure encoding.
result GSN is strictly more expressive than the Weisfeiler-Leman (WL) test and can disambiguate even hard graph isomorphism instances.
This work explores the relationship between expressivity and generalization in GNNs.
problem Understanding the trade-off between expressivity and generalization in GNNs.
method Introducing a novel framework that connects GNN generalization to the variance in graph structures they can capture.
result Theoretical findings align with empirical results, offering a deeper understanding of how expressivity enhances GNN generalization.
ESAN improves graph neural networks by processing subgraphs.
problem Limitations of MPNNs in graph isomorphism.
method ESAN represents graphs as sets of subgraphs and processes them with equivariant architectures.
result ESAN increases the expressive power of GNNs and more expressive architectures.
Deep learning finds mathematical equations from data.
problem Discovering underlying mathematical expressions from datasets.
method Uses a recurrent neural network to search for mathematical expressions and optimizes using a risk-seeking policy gradient.
result Outperforms existing methods in recovering exact symbolic expressions.