Study on feature learning dynamics in infinite-depth neural networks, focusing on ResNets.
problem Understanding how features evolve during training in deep neural networks, especially in the large-depth limit.
method Conditional Gaussian representations and SDE system with decoupled backward weights.
result Depth-induced suppression of forward-backward coupling in infinite-depth networks, leading to a decoupled forward-backward SDE system.
Researchers quantify the relationship between feature depth and performance in deep neural networks.
problem Understanding how depth affects feature extraction and generalization in deep neural networks.
method Adaptive analysis of feature-depth trade-offs in deep nets, proving optimal generalization performance.
result Optimal generalization performance achieved through empirical risk minimization on deep nets.
New CNN learns depth features from scratch, outperforming transfer learning.
problem Limited depth data for RGB-D scene recognition.
method Bottom-up approach combining local weakly supervised training and global fine-tuning, modified CNN architecture.
result State-of-the-art accuracy on NYU2 and SUN RGB-D datasets.
Deep and wide ReLU networks learn data-dependent features even in the lazy training regime.
problem Understanding the behavior of neural networks with finite depth and width.
method Analyzing the mean and variance of the neural tangent kernel (NTK) in a randomly initialized ReLU network.
result The NTK has a non-trivial evolution during training, with the mean of its first SGD update being exponential in the ratio of depth to width.
Deep CNN architectures improve neonatal seizure detection accuracy.
problem Improving EEG-based neonatal seizure detection accuracy.
method Design and test of deep convolutional networks of varying depths compared to a shallow SVM-based detector.
result A deep 11-layer CNN architecture significantly outperforms shallow architectures, improving AUC90 from 82.6% to 86.8%.
New deep fusion methods improve human action recognition using depth and inertial sensor data.
problem Existing multimodal HAR frameworks lack mid-level feature fusion.
method Proposes three deep multilevel multimodal fusion frameworks, transforming depth and inertial sensor data into images and using convolution with Prewitt filter to create modality within modality.
result Supremacy of proposed fusion frameworks over existing methods on three publicly available datasets.
AdaEnsemble learns adaptive feature interactions for CTR prediction.
problem Learning feature interactions for CTR prediction in recommender systems and Ads ranking.
method AdaEnsemble is a Sparsely-Gated Mixture-of-Experts (SparseMoE) architecture that dynamically selects feature interaction depth.
result AdaEnsemble achieves better prediction accuracy and inference efficiency compared to state-of-the-art models.
Unified spectral framework for μP under joint width-depth scaling.
problem Challenges in stable feature learning and HP transfer for width-depth scaled models.
method Developed a simple and unified spectral framework for μP under joint width-depth scaling.
result Unified and generalized μP formulation for practical architectures with multi-transformation branches.
A new data-level recombination strategy improves RGB-D salient object detection.
problem RGB-D salient object detection struggles with depth information.
method Proposes a novel data-level recombination strategy to fuse RGB and depth data before feature extraction.
result Achieves a new state-of-the-art performance in RGB-D salient object detection.
Random ReLU features are shown to be a universally consistent learning algorithm but struggle with complex functions.
problem Approximating complex functions with random ReLU features.
method Study of random ReLU features through their RKHS and composition of functions.
result Random ReLU features can efficiently approximate complex functions but not as well as multi-layer ReLU networks.
Method combines LD and Fermat Distance for neural network uncertainty.
problem Measuring uncertainty in neural network predictions.
method Statistical Depth (LD) combined with Fermat Distance.
result Effective uncertainty estimation without impacting original model performance.
Deeper networks are better for local labels, but shallower for global labels.
problem Understanding the effect of depth in overparameterized neural networks.
method Introduced local and global labels to investigate the advantage of depth.
result Deeper networks are better for local labels, shallower for global labels.
Improved isolation forest for better outlier detection.
problem Outlier detection in multivariate data.
method Random cuts across feature space, size of feature space, and point assignment information.
result Improved results in many situations without modifying tree structure.
Large neural networks learn low-dimensional representations that balance complexity and regularity.
problem Understanding the tradeoff between low-dimensional representations and complexity in deep neural networks.
method Computed finite depth corrections to reveal a measure of regularity that bounds the pseudo-determinant of the Jacobian.
result Proved the conjectured bottleneck structure in learned features as network depth increases, showing almost all hidden representations are approximately low-dimensional and weight matrices have singular values close to 1.
New approach to abstract neural network representations using renormalization group.
problem Developing truly abstract representations in neural networks.
method Renormalization group approach to expand representations to encompass broader data sets.
result Representations in neural networks become more abstract as data breadth increases and depth increases.
Study examines infinite limits of transformer dynamics, identifying key parameterizations.
problem Understanding the training dynamics of transformer models in the feature learning regime.
method Analysis of infinite scaling limits using dynamical mean field theory.
result Identified parameterizations that admit well-defined infinite width and depth limits.
ResNets and DenseNets converge to NTK with depth and width, offering advantages for kernel regression.
problem Understanding convergence of ResNets and DenseNets to Neural Tangent Kernel (NTK).
method Analysis of finite width and depth corrections for NTK of ResNets and DenseNets.
result ResNets and DenseNets can converge to NTK with depth and width, unlike vanilla networks.
Graphs avoid oversmoothing with properly initialized weights.
problem Oversmoothing in graph convolutional networks (GCNs) limits their depth.
method Used Gaussian process equivalence to analyze GCNs, identifying a non-oversmoothing phase.
result Properly initialized GCNs can be deep and expressive.
Improved graph neural networks by separating feature aggregation and depth.
problem Understanding feature importance in graph neural networks without prior information.
method Decoupling feature aggregation and depth, using softmax as a regularizer, and introducing 'Soft-Selector' and 'Hop-Normalization'.
result FSGNN model achieves up to 64% accuracy improvements in node classification tasks.
Deep networks with orthogonal weights show stable fluctuations, improving generalization and training speed.
problem Fluctuations in deep networks with Gaussian weights can impair training, especially in networks with depth comparable to width.
method Analytical and numerical studies of fully-connected networks with orthogonal weight initialization and tanh activations.
result Rectangular networks with orthogonal weights have stable fluctuations independent of network depth, leading to better generalization and training speed.
Poly-GNNs achieve similar performance regardless of depth, highlighting graph noise's dominance.
problem Performance of poly-GNNs in semi-supervised node classification.
method Analysis of poly-GNNs under a contextual stochastic block model (CSBM).
result For a sufficiently large graph, depth k > 1 k > 1 k > 1 poly-GNNs exhibit the same rate of separation as depth k = 1 k=1 k = 1 counterparts. e-values rank features for model performance.
problem Feature selection for parametric models.
method Data depths and resampling-based algorithm.
result e-values distinguish essential features.
This paper simplifies deep learning networks by mapping them to a linear function of a feature map.
problem Understanding how weights in deep networks coordinate across layers and generalize.
method Reparameterizes DNNs as a linear function of a feature map, transforming depth-dependencies into tensor products.
result Develops sample compression representation of neural networks in terms of support vectors, showing sample complexity of O(ns/epsilon).
The paper improves bounds on skein tree depth and delta-crossing numbers for knots and links.
problem Improving bounds on skein tree depth and delta-crossing numbers for knots and links.
method Theoretical and computational analysis of skein trees and knot invariants.
result New upper and lower bounds on skein tree depth and delta-crossing numbers are derived.
Unified theorem for deep and shallow joint-equivariant machines.
problem Universal approximation of joint-equivariant machines.
method Constructive universal approximation theorem based on ridgelet transform.
result Unified approximation of deep and shallow networks.
Wide neural networks learn features under μ μ μ P, identifying weights and decomposing support.
problem Feature learning in wide neural networks under μ μ μ P. method Proving mean-field limit, characterizing identifiability, sparse-dictionary decomposition, and feature-learning-error decomposition.
result The triple ( w ∗ , D o r b ∗ , S ∗ ) (w^*, D^*_{\mathrm{orb}}, S^*) ( w ∗ , D orb ∗ , S ∗ ) identifies the natural learning cell of the architecture-data pair ( σ , ρ ) (σ, ρ) ( σ , ρ ) . SCARLET-NAS improves neural architecture search by stabilizing feature inconsistencies.
problem Difficulty in ranking subnetworks due to feature inconsistency in skip connections.
method Introduced an equivariant learnable stabilizer to homogenize feature disparities.
result SCARLET-A achieves 76.9% top-1 accuracy on ImageNet.
Enhanced ODT with Feature Concatenation boosts learning efficiency.
problem Insufficient learning efficiency of ODT due to linear projections not being transmitted to child nodes.
method Feature Concatenation ( exttt{FC-ODT}) to transmit linear projections along decision paths.
result Experiments show exttt{FC-ODT} outperforms state-of-the-art decision trees with a limited tree depth.
Study of deep linear neural networks with proportional width and depth.
problem Lack of descriptive power in Gaussian limit of deep linear neural networks.
method Proportional infinite-width infinite-depth limit for deep linear neural networks.
result Characterization of limiting distribution as a nontrivial mixture of Gaussians.
Study liquidity provision with exogenous competition using a reference market maker.
problem Liquidity provision in the presence of exogenous competition.
method Model a reference market maker who optimizes posted depths, competing market makers using a rule of thumb.
result Model admits approximate closed-form solution for a linear-quadratic goal functional.
Revisits Gaussian process model with spherical harmonics for scalable deep learning.
problem Scaling Gaussian process models to large input dimensions with high frequency learning.
method Introduces new kernels related to deep models, variational learning of spherical harmonic phases, and sparseness in eigenbasis.
result Enables scaling to larger input dimensions and learning of high frequency variations.
We present a model for the joint estimation of disparity and motion. The model is based on learning about the interrelations between images from multiple cameras, multiple frames in a video, or the combination of both. We show that learning depth and motion cues, as well as their combinations, from data is possible wit…
This paper reduces the complexity of deep CNNs by optimizing their topology.
problem Reduces computational complexity and storage in deep CNNs for resource-constrained platforms.
method Analyzes the impact of CNN topology (depth and width) on feature extraction, focusing on scattering networks.
result Designs networks of fixed depth to retain a significant portion of input signal energy in the feature vector.
Deep ReLU networks can approximate piecewise smooth functions efficiently.
problem Approximating piecewise smooth functions with neural networks.
method Constructing neural networks with fixed depth and weights to approximate functions from E β ( R d ) \mathcal{E}^β(\mathbb R^d) E β ( R d ) up to L 2 L^2 L 2 error. result Optimal approximation rate requires ReLU networks of a certain depth and number of weights.
Theory explains deep nonlinear networks' plateaus and transitions.
problem Understanding long plateaus and feature acquisition transitions in deep nonlinear networks.
method Derived an exact identity for Frobenius norms, classified activation functions, and reduced matrix flow to a scalar ODE.
result Escape time law τ ⋆ = Θ ( ε − ( r − 2 ) ) τ_\star = Θ(\varepsilon^{-(r-2)}) τ ⋆ = Θ ( ε − ( r − 2 ) ) for deep nonlinear networks, where r r r is the number of bottleneck layers. A study on the depth of graph neural networks on sparse graphs, revealing a dichotomy based on the Kesten-Stigum ratio.
problem Determining the optimal depth of graph neural networks for sparse graphs.
method Analyzing the sparse contextual stochastic block model with a message-passing classifier.
result The value of depth is governed by the Kesten-Stigum ratio, with thresholds dividing performance into geometric and branching processes.
Residual networks with depthwise hyperparameter scaling transfer optimal hyperparameters across width and depth.
problem The challenge of hyperparameter tuning in deep learning, especially for large models.
method Combining μ μ μ P parameterization with residual networks having a residual branch scale of 1 / e x t d e p t h 1/\sqrt{ ext{depth}} 1/ e x t d e pt h . result Optimal hyperparameters transfer across width and depth in residual networks trained with this parameterization.
Rotationally equivariant convolutions improve molecular property prediction.
problem Predicting molecular properties using graph neural networks.
method Ablation study with rotationally equivariant and invariant convolutions on QM9 data set.
result Rotationally equivariant layers decrease test error by an average of 23%.
Study explores how dataset breadth and depth affect Siamese Neural Network performance.
problem Impact of dataset breadth and depth on Siamese Neural Network performance.
method Experiments with three keystroke datasets varying breadth and depth factors.
result Increasing dataset breadth improves model performance, while depth's impact varies by dataset type.
A new statistical concept, lepto-variance, is defined for stock returns using Regression Trees.
problem Understanding the underlying structure of stock returns using statistical methods.
method Defining lepto-variance as the variance that cannot be removed by any regression tree of a specific depth and analyzing stock returns with 1- and 2-bit Regression Trees.
result Lepto-variance quantifies the resolving power of Regression Trees for stock returns, decomposing total variance into lepto-variance and macro-variance.
Our research proves neural collapse in deep ResNets and transformers is globally optimal.
problem Understanding neural collapse in deep learning models.
method Analysis of deep regularized transformers and ResNets trained with cross entropy or mean squared error loss.
result Global optima of deep regularized transformers and ResNets are approximately collapsed, becoming more prominent as depth increases.
RocketStack integrates predictions from multiple base learners using a recursive stacking architecture up to ten levels.
problem Feature redundancy, complexity, and computational burden in deep stacking.
method Level-aware recursive stacking with pruning and compression techniques.
result Increasing accuracy with depth and outperforming standalone ensembles at later levels.
A new pseudo-metric uses data depth to compare probability distributions.
problem Designing a metric between probability distributions for machine learning applications.
method Extension of univariate quantiles to multivariate spaces, using data depth and Hausdorff distance.
result The pseudo-metric is robust, factorizes translations, and has good behavior under transformations.
The study analyzes games and social hierarchies, incorporating luck and depth of competition.
problem Analyzing patterns of wins and losses in games and social hierarchies.
method Generalized probabilistic models incorporating luck and depth of competition.
result Social competition tends to be deeper with many distinct levels, but there is often a chance of upset victories.
This paper introduces depth functions for ranking data, improving statistical summaries.
problem Lack of comprehensive statistical summaries for ranking data.
method Metric-based depth functions on symmetric group to define rankings, depths, and procedures.
result Novel depth functions provide a more informative summary of ranking data.
Flexible deep learning framework controls FDR for feature selection.
problem Controlling Type-I error in feature selection for deep neural networks.
method Approximates FDR control for a wide range of deep architectures using gradient-based feature-importance vectors.
result Theoretical guarantee of FDR control for feature selection in deep learning models.
Paper presents a world model that learns invariant causal features using contrastive unsupervised learning.
problem Learning invariant causal features in unsupervised settings.
method Contrastive unsupervised learning with intervention invariant auxiliary task.
result Significantly outperforms state-of-the-art methods on out-of-distribution point navigation tasks.
Study solves utility maximization in a transient price impact market.
problem Utility maximization in a market with transient price impact.
method Developed a discrete-time model and removed market depth and resilience process restrictions.
result Solved the utility maximization problem without convexity of attainable portfolio values.