Paper analyzes approximation and learning of MoEs with (P)ReLU activation.
problem Scaling up deep learning models with MoEs and (P)ReLU activation.
method Approximation and learning-theoretic analysis of MoMLPs with (P)ReLU.
result MoMLPs can uniformly approximate Lipschitz functions with ε accuracy using O(ε−1) parameters. ResNets promote smoother interpolations than MLPs, enhancing generalization.
problem Understanding the difference in smoothness between ResNets and MLPs.
method Neural Tangent Kernel (NTK) analysis during gradient descent training.
result ResNet's NTK results in smoother interpolations than MLPs.
Neural networks extrapolate poorly in simple tasks but succeed in complex ones.
problem Understanding neural networks' extrapolation capabilities and conditions for success.
method Analyzing ReLU MLPs and GNNs, connecting to neural tangent kernel.
result ReLU MLPs learn linear functions but not most nonlinear ones, while GNNs succeed in complex tasks due to task-specific non-linearities.
New neural network class approximates Hölder functions with optimal error and sample complexity.
problem Finding a neural network class that is both expressive and statistically reliable.
method Constructive identification of a ReLU MLP class with optimal approximation properties and near-optimal sample complexity.
result Optimal ReLU MLPs can approximate Hölder functions with uniform error and near-optimal sample complexity.
New method initializes sigmoidal MLPs for interpretable shapes.
problem Creating interpretable decision boundaries in neural networks.
method Introducing a geometry-aware initialization for sigmoidal multi-layer perceptrons (MLPs) using tropical geometry.
result Sigmoidal MLPs can have decision boundaries aligned with prescribed shapes at initialization.
Guarantees sparse recovery for neural networks with iterative hard thresholding.
problem Recovering sparse network weights in neural networks.
method Structural properties of sparse network weights and iterative hard thresholding algorithm.
result Simple iterative hard thresholding algorithm recovers sparse network weights exactly using linear memory.
An appropriate choice of the activation function (like ReLU, sigmoid or swish) plays an important role in the performance of (deep) multilayer perceptrons (MLP) for classification and regression learning. Prototype-based classification learning methods like (generalized) learning vector quantization (GLVQ) are powerful…
A new perspective on self-attention models using MLPs.
problem Improving sequence modeling with self-attention mechanisms.
method Introducing HyperMLP and HyperGLU, which use dynamic two-layer MLPs with reverse-offset layout.
result HyperMLP/HyperGLU consistently outperform softmax-attention baselines.
An interesting approach to analyzing neural networks that has received renewed attention is to examine the equivalent kernel of the neural network. This is based on the fact that a fully connected feedforward network with one hidden layer, a certain weight distribution, an activation function, and an infinite number of…
Study on learning rates in neural networks of varying depth.
problem Dependence of maximal update learning rate on network depth.
method Analysis of random fully connected ReLU networks with mean-field weight initialization.
result Maximal update learning rate scales like L−3/2 with network depth. New Lipschitz bound for ReLU networks resists weight rescaling.
problem Lack of robustness guarantees for ReLU networks under weight perturbations.
method Rescaling-invariant Lipschitz bound based on path-metrics.
result The new bound applies to various ReLU-DAG architectures and resists neuron-wise rescalings.
Deep neural networks can classify any dataset with exact solutions.
problem Classifying any dataset with multilayer perceptrons.
method Constructive analysis of multilayer perceptrons with exact solutions.
result Explicit constructions of MLPs for classification and function approximation.
Transformers exhibit sparse activation maps, reducing computational load and improving robustness.
problem Sparse activation in Transformer models.
method Extensive experiments on various Transformer architectures and tasks.
result Sparsity in Transformers is a prevalent phenomenon, reducing FLOP count and improving model robustness.
Proposes using MLP for predicting optimal penalty in changepoint detection.
problem Predicting optimal penalty for changepoints in sequences.
method Uses a multilayer perceptron (MLP) with ReLU activation function to predict penalty.
result Improves accuracy and F1 score compared to existing models.
Sine activation functions enable two-layer neural networks to learn modular addition more efficiently.
problem Learning modular addition with two-layer neural networks.
method Introduced and analyzed sine activation functions, providing theoretical and empirical evidence.
result Sine activation functions allow for constant-width network realizations of modular addition, whereas ReLU networks require linear width scaling.
Wide hypernetworks don't guarantee convergence under gradient descent.
problem Theoretical guarantees for wide hypernetworks in over-parameterized settings.
method Analyzing infinitely wide hypernetworks and their convergence properties.
result Infinitely wide hypernetworks do not guarantee convergence to a global minimum under gradient descent.
MLP-Mixer achieves better performance through sparsity and wider architecture.
problem Understanding why MLP-Mixer outperforms conventional MLPs.
method Revealed sparseness as a key mechanism, showed effective expression as wider MLP, demonstrated quantitative similarities, and applied a guiding principle.
result MLP-Mixer's performance improvement through sparseness and wider architecture.
PAC-Bayesian bounds for MLPs with cross entropy loss validated.
problem Generalization bounds for MLPs with cross entropy loss.
method Introduced probabilistic explanations and proved PAC-Bayesian bounds using ELBO.
result MLPs with cross entropy loss inherently guarantee PAC-Bayesian generalization bounds.
Dropout schedules can be optimized to significantly reduce model test loss.
problem Improving model performance in neural networks.
method Developed a mean-field theory of dropout at the edge of chaos, proposing front-loaded dropout schedules.
result Front-loaded dropout schedules reduce test loss by 18-35% over constant dropout.
Interpolated-MLPs control inductive bias for better performance in low-compute tasks.
problem Low-compute performance gap between MLPs and CNNs.
method Introduced Interpolated MLP (I-MLP) approach to control inductive bias incrementally.
result Continuous logarithmic relationship between inductive bias and performance in low-compute tasks.
ES-MLP combines Graph-MLP with edge splitting for node classification on both homophilic and heterophilic graphs.
problem Node classification on graphs with mixed homophilic and heterophilic properties.
method Combines Graph-MLP with edge splitting mechanism from ES-GNN to learn two adjacency matrices based on relevant and irrelevant feature pairs.
result ES-MLP achieves performance comparable to homophilic and heterophilic models without using edges during inference.
A new GNM model outperforms MLP for tabular data.
problem Learning with tabular data.
method Proposes Graph Neural Machine (GNM) replacing MLP's graph representation with a nearly complete graph and using synchronous message passing.
result GNM outperforms MLP in classification and regression tasks.
Novel KAN-based autoencoder improves asset pricing models' accuracy and interpretability.
problem Improving asset pricing models' accuracy and interpretability.
method Kolmogorov-Arnold Networks (KANs) inspired autoencoder for latent factor exposures.
result Outperforms Multilayer Perceptrons in both accuracy and interpretability.
This study compares GNNs and GA-MLPs, finding GA-MLPs can distinguish graphs but not count walks.
problem Comparing expressive power and graph isomorphism testing capabilities of GNNs and GA-MLPs.
method GA-MLPs augment node features with multi-hop operators and apply MLPs node-wise; GNNs are compared as a baseline.
result GA-MLPs can distinguish almost all non-isomorphic graphs but cannot count attributed walks, unlike GNNs.
Bilinear MLPs offer a new way to interpret deep learning models without complex nonlinearities.
problem Lack of mechanistic understanding in how MLPs compute.
method Introduced bilinear MLPs without element-wise nonlinearities, analyzed their weights using tensor and eigendecomposition.
result Bilinear MLPs provide interpretable weight structures and enable adversarial attacks and overfitting analysis.
We study the concentration of NTK for MLPs at EOC, proving finite-width approximation of gradient independence.
problem Understanding the concentration of Neural Tangent Kernel (NTK) for MLPs at the Edge of Chaos (EOC).
method Proved approximate gradient independence holds at finite width, using maximal inequalities to show NTK matrix concentrates around its infinitely wide limit.
result The NTK matrix of MLPs at EOC concentrates around its infinitely wide limit, requiring hidden layer widths to grow quadratically.
New findings connect shaped and unshaped neural networks using differential equations.
problem Understanding the behavior of neural networks with different activation scaling methods.
method Deriving differential equation-based asymptotic characterizations for shaped and unshaped neural networks.
result Two types of unshaped networks converge to the same infinite-depth-and-width limit at initialization.
Single neuron with ADA learns XOR and outperforms other functions.
problem Classifying linearly non-separable data.
method Proposed a new artificial neuron with apical dendrite activation.
result ADA function achieves 100% accuracy on XOR and superior performance on benchmark datasets.
This paper presents a new family of backpropagation-free neural architectures, Gated Linear Networks (GLNs). What distinguishes GLNs from contemporary neural networks is the distributed and local nature of their credit assignment mechanism; each neuron directly predicts the target, forgoing the ability to learn feature…
KANs replace fixed MLP weights with learnable edge functions, improving accuracy and interpretability.
problem Lack of interpretability and scalability in MLPs.
method KANs use learnable activation functions on edges instead of fixed weights, replacing weights with spline functions.
result KANs outperform MLPs in accuracy and interpretability with smaller models.
Artificial Neural Networks(ANN) has been phenomenally successful on various pattern recognition tasks. However, the design of neural networks rely heavily on the experience and intuitions of individual developers. In this article, the author introduces a mathematical structure called MLP algebra on the set of all Multi…
Revisits neural collaborative filtering vs. matrix factorization, showing dot product superiority.
problem Comparing neural collaborative filtering to matrix factorization in recommendation systems.
method Revisited experiments using MLPs as similarity functions, comparing dot product to MLP outputs.
result Simple dot product outperforms MLP-based learned similarities in practical settings.
Universal MLPs with a single hidden layer can learn any function.
problem Learning on various data structures like sequences, images, sets, and graphs.
method Using group theory, the paper proves the universality of a broad class of equivariant MLPs with a single hidden layer.
result Having a hidden layer on which the group acts regularly is sufficient for universal equivariance (invariance).
Study uses MLP models to predict large-cap US stocks, finding 2-3 hidden layers more flexible.
problem Predicting asset prices for large-cap US stocks.
method Applied MLP models with dynamic structure to factor models, focusing on firm characteristics.
result MLP models with 2-3 hidden layers more flexible in modeling factors, better for downside risk control.
SpArX creates faithful explanations of neural networks' decision-making.
problem Challenges in explaining neural networks' decisions.
method Sparsifies MLPs while maintaining structure, then translates into QAFs for argumentative explanations.
result SpArX provides more faithful explanations than existing methods.
This paper proposes a new approach to Transformers by integrating hierarchical associative memory with MetaFormers.
problem Theoretical framework for Transformers and MLP-Mixers remains underdeveloped.
method Integrating hierarchical associative memory with MetaFormers to create a parallelized MLP-Mixer.
result Symmetry-breaking effects improve the performance of the MLP-Mixer in image recognition tasks.
Designs an MLP from LDA for multi-Gaussian class classification.
problem Classifying inputs with multiple Gaussian distributions.
method Interprets MLP as generalized LDA, using LDAs for half-space partitioning, neurons for subspace isolation, and merging for class-wise representation.
result Automatic feedforward design for MLP architecture and weights.
m-arcsinh improves SVM and MLP reliability and speed in scikit-learn.
problem Improving SVM and MLP reliability and speed in scikit-learn.
method Modified arcsinh function for kernel and activation in SVM and MLP.
result Competitive classification performance and reliability of SVM and MLP with m-arcsinh.
This study compares MLPs and KANs in low-data regimes, finding MLPs with personalized activation functions outperform KANs.
problem Comparing MLPs and KANs in low-data regimes.
method Introduced an effective technique for designing MLPs with unique, parameterized activation functions for each neuron.
result MLPs with personalized activation functions achieve significantly higher predictive accuracy with only a modest increase in parameters, especially in low-data regimes.
Wav-KAN improves neural network interpretability and performance.
problem Challenges in interpretability, training speed, robustness, and performance of traditional neural networks.
method Integrates wavelet functions into the Kolmogorov-Arnold network structure for efficient data representation.
result Enhanced accuracy, faster training speeds, and increased robustness compared to existing methods.
New approach finds minimum width for deep, narrow MLPs.
problem Finding the minimum width for deep, narrow MLPs to approximate continuous functions.
method Proposes a framework to simplify finding minimum width into determining a geometrical function w(dx,dy) based on input and output dimensions. result Proves that w(dx,dy) equals the optimal minimum width for deep, narrow MLPs to achieve universality. Investigates transfer learning in spatial statistics.
problem Applying transfer learning to spatial statistics.
method Simple MLP models for spatial data.
result Potential of transfer learning in spatial statistics.
Hybrid model improves wind speed prediction accuracy using MLP and WOA.
problem Improving wind speed prediction accuracy for renewable energy control.
method Combining MLP with Whale Optimization Algorithm (WOA) for data preprocessing and model optimization.
result The hybrid MLP-WOA model outperformed standalone MLP model in wind speed prediction accuracy.
S-GAI initializes MLPs using spectral geometry from data, improving performance.
problem Lack of guidance on initial weights encoding data geometry.
method S-GAI uses SVD to estimate spectral class geometry, initializing MLPs from training data.
result S-GAI-initialized MLPs start from a more informative hidden state and achieve comparable accuracy.
Closed-form polynomial approximations replace MLPs in transformers, enabling new interpretability methods.
problem Replacing MLPs with polynomial approximations for transformer models.
method Theoretical derivation of closed-form least-squares approximations of MLPs and GLUs using polynomial functions.
result Polynomial approximations explain over 95% of MLP and GLU outputs' variance, enabling interpretability.
Topo-MLP learns network representations without message passing.
problem Lack of efficient higher-order network modeling methods.
method Proposes Topo-MLP, a simplicial neural network algorithm using MLP and HONC loss.
result Demonstrates improved robustness and efficiency in representation learning.
MLPs can approximate any function in context, challenging the importance of in-context universality.
problem Understanding why transformers are more effective than classical models.
method Proved MLPs with trainable activation functions are universal in context.
result Transformer success is likely due to factors other than in-context universality.
Transformers show better in-context learning resilience under distribution shifts than simple MLPs.
problem Understanding in-context learning under varying distribution shifts.
method Comparing transformers and set-based MLPs on linear regression tasks.
result Transformers better emulate OLS performance and exhibit better resilience to mild distribution shifts.