Research
On-device research index

arXiv research

A locally-built, LLM-digested index of recent arXiv papers in quant finance, geometry/topology, and statistical ML — keyword search served straight from SQLite on this machine.

169,181 papers · 148 categories

Trend · papers per month

265278104 · Jun 202019922001200920182026
48 results for CNN depth

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 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%.

The paper introduces a method to explain redundancy in deep CNNs using unit impulse response.

problem Redundancy in deep CNNs leads to unnecessary computations and increased cost.
method Empirical demonstration and unit impulse response analysis to identify and quantify redundancy across layers and depth.
result Identifies and quantifies redundancy in deep CNNs, providing better insights into their internal dynamics.

CNNs improve generalization to unseen audio devices with increased width, not depth.

problem CNNs are sensitive to specific audio recording devices in acoustic scene classification.
method Investigated the relationship between over-parameterization and generalization in CNNs for audio classification.
result Increasing width improves generalization to unseen devices without increasing the number of parameters.

CNN improves neutrino event reconstruction in IceCube DeepCore.

problem Difficulties in distinguishing muon neutrinos and reconstructing inelasticity at GeV scale energies.
method 2D Convolutional Neural Network exploiting time and depth translational symmetry.
result CNN model outperforms conventional methods for flavor identification and inelasticity reconstruction.

Unified learning-rate scale for CNNs and ResNets, avoiding depth imbalance.

problem Challenges in choosing an appropriate learning rate for deep networks, especially as depth increases.
method Introduces Arithmetic-Mean μμP (AM-μμP), constraining network-wide average pre-activation second moment to a constant scale, combined with residual-aware He fan-in initialization.
result Demonstrates a 3/2-3/2 scaling law for learning rates across depths, enabling zero-shot learning-rate transfer.

Wide CNNs with shared weights and max pooling have linearly independent features and can achieve zero training error.

problem Understanding the optimization landscape and expressiveness of deep CNNs.
method Analysis of loss landscape and expressiveness of practical deep CNNs with shared weights and max pooling layers.
result Wide CNNs can achieve zero training error and have a well-behaved loss surface with almost no bad local minima.

DNArch learns CNN architectures by backpropagation.

problem Discovering optimal CNN architectures.
method Differentiable Neural Architectures (DNArch) learns CNN architectures by backpropagation, controlling kernel sizes, channels, downsampling positions, and depth.
result DNArch finds performant CNN architectures across various tasks.

Theoretical analysis of CNNs' inductive biases and their efficiency in approximating functions.

problem Understanding and optimizing the inductive biases in deep CNNs.
method Theoretical analysis combining multichanneling, downsampling, weight sharing, and locality.
result Deep CNNs with O(logd)\mathcal{O}(\log d) depth can approximate any continuous function, and require O~(log2d)\widetilde{\mathcal{O}}(\log^2d) samples for sparse functions.

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.

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.

CNNs solve inverse problems during training, proving mutual coherence crucial for convergence.

problem Validation of CNN learning during training.
method Proved CNN elements solve inverse problems, discussed mutual coherence, and set training rules.
result Mutual coherence is necessary for CNNs to converge to optimum solutions.

Gradient descent on deep linear CNNs converges to a penalty-based solution.

problem Understanding gradient descent convergence in deep linear convolutional networks.
method Gradient descent on full-width linear convolutional networks of varying depth.
result Gradient descent converges to a penalty-based solution, not the hard margin SVM solution.

Study of infinitely deep but narrow neural networks using NTK theory.

problem Analyzing the role of depth in deep learning with overparameterized networks.
method Infinite-depth limit analysis of MLP and CNN using Neural Tangent Kernel (NTK) theory.
result Established trainability guarantee for infinitely deep but narrow neural networks.

GP-CNAS uses genetic programming to automatically design CNN architectures.

problem Designing optimal CNN architectures is laborious and error-prone.
method GP-CNAS uses a tree-based representation of CNNs and dynamic crossover operators to search for optimal architectures.
result GP-CNAS finds optimal CNN architectures with balanced depth and width in limited trials.

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.

New tighter generalization bounds for deep networks like CNNs and ResNets.

problem Establishing tighter bounds for deep neural networks' generalization error.
method Introducing a new characterization of Lipschitz properties and margin-based data-dependent error bounds.
result Significantly tighter generalization bounds for deep neural networks, including CNNs and ResNets.

Study on memorization vs. generalization in overparameterized networks.

problem Understanding the trade-off between memorization and generalization in neural networks.
method Examined fully-connected and convolutional networks trained to minimize reconstruction error.
result Different architectures exhibit distinct inductive biases, affecting generalization from a single training example.

CNNs generalize well despite learning surface statistical regularities.

problem CNNs' extreme sensitivity to adversarial examples raises doubts about learning high-level abstractions.
method Fourier filtering to construct datasets with same high-level abstractions but different surface statistics.
result CNNs exhibit a tendency to learn surface statistical regularities, leading to a 28% generalization gap.

RIFLE improves deep transfer learning by reinitializing fully-connected layers.

problem Limited improvement in transfer learning accuracy with pre-trained models on small datasets.
method Re-Initializing fully-connected layers with random scratch during fine-tuning.
result Significant improvement in deep transfer learning accuracy across various datasets.

Deep CNNs can be trained without special architectures.

problem Training extremely deep CNNs (10,000 layers) is challenging due to vanishing/exploding gradients.
method Developed a mean field theory for signal propagation and conditions for dynamical isometry. Derived an algorithm for generating random initial orthogonal convolution kernels.
result Vanilla CNNs with ten thousand layers can be efficiently trained using appropriate initialization schemes.

CNNs use a bottleneck structure to focus on a few frequencies, affecting function representation.

problem Understanding how CNNs focus on specific frequencies in their feature learning.
method Defined Convolution Bottleneck (CBN) structure, measured CBN rank, and analyzed parameter norms.
result Parameter norm scales with depth and CBN rank, and networks with optimal parameters exhibit this structure.

Butterfly-Net improves CNN performance with structured connections and initialization.

problem Improving the performance of convolutional neural networks (CNNs).
method Butterfly-Net introduces structured and sparse cross-channel connections, and Butterfly initialization strategy.
result Butterfly-Net approximates Fourier representations with exponentially decaying error as depth increases.

Optimized CNNs for AMC on edge devices reduce complexity without sacrificing accuracy.

problem Developing efficient DL models for AMC on resource-constrained edge devices.
method Pruning, quantization, and knowledge distillation techniques applied to CNNs.
result Optimized models maintain or improve AMC accuracy with reduced complexity.

Improved CNNs detect Alzheimer's with 14% accuracy boost.

problem Early detection of Alzheimer's Disease using MRI scans.
method Optimized 3D CNNs with instance normalization, spatial downsampling, model widening, and age information.
result 14% increase in test accuracy distinguishing AD, MCI, and controls.

Unified model for automated driving tasks improves efficiency and accuracy.

problem Efficiently handle multiple visual perception tasks in automated driving systems.
method Joint multi-task network design sharing convolutional layers, multi-stream learning, auxiliary learning.
result Unified model outperforms single-task models in many cases.

IC-Network improves CNNs by integrating elastic collision units.

problem Designing more effective basic units in neural networks.
method Developed IC layer and IC block units combining the IC structure with convolution operations.
result Significant performance improvements in existing CNNs, reducing top-1 error from 22.85% to 21.49% on imagenet.

New approach predicts generalization of deep neural networks in proportional-width regime.

problem Predicting generalization of deep neural networks in proportional-width regime.
method Equivalent Wishart Ansatz for hierarchical empirical kernels, renormalized NNGP kernel.
result Renormalized NNGP kernel captures dominant stochastic fluctuations in deep neural networks.