Rotation-equivariant CNN reveals common features in V1 neurons.
problem V1 models fail to predict natural stimuli responses accurately.
method Rotation-equivariant convolutional neural network model.
result Rotation-equivariant network outperforms regular CNN and reveals common features.
We propose a new model for digital pathology segmentation, based on the observation that histopathology images are inherently symmetric under rotation and reflection. Utilizing recent findings on rotation equivariant CNNs, the proposed model leverages these symmetries in a principled manner. We present a visual analysi…
DeepSphere improves spherical CNNs by balancing efficiency and rotation equivariance.
problem Designing efficient and rotation-equivariant convolutional layers for spherical data.
method Graph-based approach to represent spherical data, focusing on the number of vertices and neighbors.
result DeepSphere achieves state-of-the-art performance and demonstrates efficiency and flexibility.
Translating or rotating an input image should not affect the results of many computer vision tasks. Convolutional neural networks (CNNs) are already translation equivariant: input image translations produce proportionate feature map translations. This is not the case for rotations. Global rotation equivariance is typic…
Explicit encoding of group actions in deep features makes it possible for convolutional neural networks (CNNs) to handle global deformations of images, which is critical to success in many vision tasks. This paper proposes to decompose the convolutional filters over joint steerable bases across the space and the group …
Performance of neural networks can be significantly improved by encoding known invariance for particular tasks. Many image classification tasks, such as those related to cellular imaging, exhibit invariance to rotation. We present a novel scheme using the magnitude response of the 2D-discrete-Fourier transform (2D-DFT)…
Graph-based CNN for spherical data with equivariance.
problem Efficiently learning from non-uniformly distributed spherical data.
method Discretized sphere as graph, graph convolutions, equivariance using Defferrard's graph neural network.
result Good performance on rotation-invariant learning problems.
We propose a semantic segmentation model that exploits rotation and reflection symmetries. We demonstrate significant gains in sample efficiency due to increased weight sharing, as well as improvements in robustness to symmetry transformations. The group equivariant CNN framework is extended for segmentation by introdu…
Convolutional Neural Networks (CNNs) have become the method of choice for learning problems involving 2D planar images. However, a number of problems of recent interest have created a demand for models that can analyze spherical images. Examples include omnidirectional vision for drones, robots, and autonomous cars, mo…
The paper introduces REQNNs for robust 3D point cloud processing.
problem 3D point cloud processing robustness to rotations.
method Revised neural networks using quaternion features for rotation-equivariance.
result REQNNs exhibit higher rotation robustness compared to original networks.
RIO uses rotation-equivariance to train robust inertial odometry models.
problem Training robust inertial odometry models with limited labeled data.
method Rotation-equivariance as self-supervisor, adaptive Test-Time Training (TTT).
result RIO-trained models achieve on-par performance with full-labeled data models.
B-Spline CNNs on Lie Groups expand G-CNNs to arbitrary groups.
problem Leveraging geometric structure for improved feature learning.
method Lifting feature maps to B-spline expansions on Lie algebra.
result G-CNNs on Lie groups outperform classical 2D CNNs.
RotEqNet preserves rotation symmetry in fluid systems using high-order tensors.
problem Lack of rotational symmetry in machine learning models for fluid systems.
method Introduces RotEqNet, a network that guarantees rotation-equivariance for high-order tensors.
result RotEqNet reduces errors and maintains rotation-equivariance in fluid systems.
Fixed points of mean section operators found in convex bodies.
problem Characterizing fixed points of mean section operators in convex bodies.
method Characterization of rotation equivariant operators using spherical Laplacian mass distribution, and application of Minkowski valuations.
result Euclidean balls are the only fixed points of mean section operators in a C2 neighborhood of the unit ball. RENNs protect input privacy by rotating d-ary features.
problem Protecting input privacy from intermediate-layer features.
method Rotation-equivariant neural networks using d-ary vectors/tensors.
result RENNs effectively hide input information without degrading output accuracy.
3D Convolutional Neural Networks are sensitive to transformations applied to their input. This is a problem because a voxelized version of a 3D object, and its rotated clone, will look unrelated to each other after passing through to the last layer of a network. Instead, an idealized model would preserve a meaningful r…
The projection body operator Π, which associates with every convex body in Euclidean space Rn its projection body, is a continuous valuation, it is invariant under translations and equivariant under rotations. It is also well known that Π maps the set of polytopes in Rn into itself. We show that Π is the only non-trivi…
First time projective elliptic genera constructed for oriented manifolds.
problem Defining and proving properties of projective elliptic genera.
method Topological construction and analytic interpretation via fractional index theorem.
result Modularity properties of projective elliptic genera proven.
Overview of methods for rotating 2D and 3D data.
problem Processing data with equivariance/invariance under rotations.
method An overview of methods for 2D and 3D rotations.
result Identification of commonalities and links between methods.
New approximative kernels improve PDE-G-CNNs for geometric deep learning.
problem Inaccurate approximations of exact kernels in PDE-G-CNNs.
method Developed new approximative kernels that work regardless of spatial anisotropy.
result New kernels provide better error estimates and maintain reflectional symmetries.
CNNs improve medical image classification with few samples.
problem Classifying medical images with limited training data.
method Transfer learning using CNNs, representation extraction, and a novel metric for performance prediction.
result CNN-based transfer learning outperforms feature-based methods with high correlation to test set performance.
In recent years, deep learning poses a deep technical revolution in almost every field and attracts great attentions from industry and academia. Especially, the convolutional neural network (CNN), one representative model of deep learning, achieves great successes in computer vision and natural language processing. How…
AT-CNNs show improved shape recognition over texture recognition.
problem Understanding adversarial training's impact on CNNs' feature learning.
method Systematic qualitative and quantitative approaches to interpret AT-CNNs.
result Adversarial training reduces texture bias and improves shape recognition.
Paper learns an explainer to interpret CNN features without annotations.
problem Interpreting complex features in CNNs without labeled data.
method Unsupervised learning of an explainer to decompose and reconstruct feature maps.
result Explainer learns to reconstruct CNN features without losing information.
Paper uses CNNs to classify heart sounds from short segments.
problem Classifying heart sounds from short segments of individual beats.
method Developed a 1D-CNN and 2D-CNN ensemble for feature learning and score-level fusion.
result ECNN ensemble achieved 89.22% accuracy and 89.94% sensitivity on the PhysioNet CinC 2016 database.
Simplified Butterfly-Net2 improves CNN efficiency in solving PDEs and signal processing tasks.
problem Improving CNN efficiency in solving PDEs and signal processing tasks.
method Introducing BNet2, a simplified Butterfly-Net, and Fourier transform initialization.
result BNet2 achieves similar accuracy as CNN but with fewer parameters and improves accuracy over randomly initialized CNN.
Automates resource-efficient CNN design for IoT.
problem Designing custom CNNs for IoT applications is impractical and resource-intensive.
method Automated synthesis of resource scalable CNNs from an existing optimized baseline CNN.
result Synthesized CNNs are resource-efficient and competitive in accuracy.
2D CNNs approximate Korobov functions with near-optimal rates.
problem Approximating Korobov functions using 2D CNNs.
method Constructive approach for 2D CNNs with ReLU activations and fully connected layers.
result 2D CNNs achieve near-optimal approximation rates for Korobov functions.
Proposes interpretable filters in CNNs for object classification.
problem Learning interpretable filters in deep CNNs without additional annotations.
method Assigns each filter in a CNN to an object part during training.
result Interpretable filters are more semantically meaningful than traditional filters.
New metrics differentiate effective OOD sets for training calibrated CNNs.
problem Vanilla CNNs struggle with out-of-distribution (OOD) samples.
method Developed metrics based on generalization errors of Augmented-CNN.
result Most protective OOD sets lead to better A-CNN performance.
In image classification, visual separability between different object categories is highly uneven, and some categories are more difficult to distinguish than others. Such difficult categories demand more dedicated classifiers. However, existing deep convolutional neural networks (CNN) are trained as flat N-way classifi…
Proposes a fixed smooth convolutional layer to reduce checkerboard artifacts in CNNs.
problem Checkerboard artifacts in CNNs during upsampling and strided convolution.
method Fixed convolutional layer with adjustable smoothness, applied to four CNNs and GANs.
result Significantly improves classification performance and image generation quality.
In this paper, we address the issue of how to enhance the generalization performance of convolutional neural networks (CNN) in the early learning stage for image classification. This is motivated by real-time applications that require the generalization performance of CNN to be satisfactory within limited training time…
TinyCNN accelerates CNN models on embedded FPGA with 15x speedup.
problem Limited memory on embedded FPGAs restricts CNN performance.
method Software and hardware design tool for FPGA resource-aware CNN accelerator.
result 3% accuracy loss with 15.75x speedup on image classification.
Generative adversarial networks fix aliasing issues by making signals continuous.
problem Alias-free generation in GANs to prevent unwanted information leakage.
method Interpreting all signals as continuous, deriving small architectural changes.
result Generative models match FID of StyleGAN2 but have better internal representations.
DoPa detects various physical adversarial attacks on CNNs.
problem Vulnerability of CNNs to physical adversarial attacks.
method Interprets CNN's vulnerability, adds self-verification stage.
result Achieves 90% success rate for image attacks and 92% for audio attacks.
Study improves CNNs for audio scene classification by restricting receptive fields and adding frequency awareness.
problem Improving CNNs for robust acoustic scene classification.
method Investigated different receptive field configurations for various CNN architectures and introduced Frequency Aware CNNs.
result Several well-performing submissions to DCASE 2019 Challenge were achieved.
Simple 1D-CNN network predicts electricity loads 36 hours ahead.
problem Forecasting electricity loads for future time periods.
method Used a one-dimensional CNN with parameter scanning to optimize kernel size, filters, and dense size.
result Good forecast quality achieved with basic CNN architectures.
Morph accelerates 3D CNNs for video recognition, reducing energy consumption and improving performance.
problem Efficiently accelerating 3D CNNs for video recognition is challenging due to their large memory footprint and higher dimensionality.
method Designing a flexible accelerator called Morph that adapts to different spatial and temporal tiling strategies, and codesigning a software infrastructure to control the hardware.
result Morph achieves up to 3.4x reduction in energy consumption and up to 5.1x improvement in performance/watt compared to a baseline 3D CNN accelerator.
Convolutional neural networks (CNN) have achieved state of the art performance on both classification and segmentation tasks. Applying CNNs to microscopy images is challenging due to the lack of datasets labeled at the single cell level. We extend the application of CNNs to microscopy image classification and segmentat…
MBS reduces CNN model size with minimal accuracy loss.
problem Reducing CNN model size while maintaining accuracy.
method Adaptive macroblock scaling based on effective flops.
result Significant model size reduction across various CNN architectures.
Improved robustness of 1D CNNs for heart arrhythmia classification.
problem Improving the robustness of 1D CNNs for classification tasks.
method Parameterization using Cayley transform and controllability Gramian for Lipschitz-bounded CNNs.
result Improved robustness of trained Lipschitz-bounded 1D CNNs for heart arrhythmia classification.
Efficient CNN for VQA achieves similar performance to standard models.
problem Computational intensity of standard VQA models.
method Proposes a sparsely activated CNN architecture.
result Sparsely activated CNN achieves comparable performance.
Computer vision performances have been significantly improved in recent years by Convolutional Neural Networks(CNN). Currently, applications using CNN algorithms are deployed mainly on general purpose hardwares, such as CPUs, GPUs or FPGAs. However, power consumption, speed, accuracy, memory footprint, and die size sho…
This paper reports the performances of shallow word-level convolutional neural networks (CNN), our earlier work (2015), on the eight datasets with relatively large training data that were used for testing the very deep character-level CNN in Conneau et al. (2016). Our findings are as follows. The shallow word-level CNN…
Study on CNNs' learning rates and approximation capacities.
problem Learning rates and approximation capacities of CNNs.
method New approximation bound and covering number analysis for CNNs.
result Derives minimax optimal convergence rates for CNNs in various learning problems.
Graph-CNN for 3D point cloud classification tackles non-regular graph topology.
problem Classifying 3D point cloud data with non-regular graph topology.
method Developed PointGCN combining localized graph convolutions and graph downsampling.
result Achieves competitive performance on 3D object classification benchmark ModelNet.
Paper introduces zero-space memory protection for CNNs without ECC overhead.
problem Ensuring reliability of CNNs in safety-critical applications.
method In-place zero-space ECC with weight distribution-oriented training.
result First known zero-space cost memory protection for CNNs.