New method uses eigenspectrum to optimize object detector architectures.
problem Understanding the effects of ImageNet pre-training on object detectors.
method Analysis of eigenspectrum dynamics of feature maps in object detectors.
result Object detectors trained from scratch and ImageNet pre-trained models behave differently.
Bayesian evidence framework selects best pre-trained CNN for transfer learning.
problem Selecting the best pre-trained deep representation for transfer learning.
method Formulated on LS-SVM classifier, evidence maximization for regularization parameters, Aitken's delta-squared process for convergence, greedy algorithm for ensemble selection.
result Competitive performance in visual recognition datasets, state-of-the-art performance in accuracy and efficiency.
SimCLR pre-training improves CNN performance with fewer labels.
problem Learning with fewer labeled data.
method SimCLR contrastive learning method combined with supervised fine-tuning.
result SimCLR pre-training with supervised fine-tuning achieves almost optimal test loss with fewer labeled data.
Cost-effective training of deep CNNs with active model adaptation.
problem Challenges in training deep CNNs for specific tasks.
method Active model adaptation of a pre-trained model with selective fine-tuning of examples.
result Significant reduction in labeling cost with improved model performance.
Deep learning identifies smartphone users from motion sensor data.
problem Smartphone user identification using motion sensor signals.
method Transformed motion signals into images, trained CNN for classification, used SVM for few-shot identification.
result CNN achieved 89.75% multi-class user classification and 96.72% few-shot user identification accuracy.
Cosine loss improves CNN performance on small datasets.
problem Training CNNs from scratch on small datasets without pre-training.
method Used cosine loss instead of cross-entropy loss.
result Accuracy on CUB-200-2011 dataset is 30% higher with cosine loss.
CompactNet optimizes CNN models for resource-limited platforms.
problem Challenges in implementing CNN models on resource-limited platforms.
method Guided by a simulator, CompactNet progressively trims a pre-trained network to achieve target speedup while maintaining accuracy.
result Achieves up to 1.8x kernel computation speedup on embedded platforms.
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.
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.
State-of-the-art Twitter sentiment analysis using CNNs and LSTMs.
problem Improving Twitter sentiment classification accuracy.
method Pre-trained word embeddings, distant supervision, fine-tuning, ensemble of CNNs and LSTMs.
result First rank on all five English subtasks of SemEval-2017.
A CNN-based approach tackles malware image classification imbalance.
problem Imbalance in malware families during image classification.
method Proposes a weighted softmax loss to address imbalance.
result Improves classification performance on malware images.
Proposes a new CNN using DTCWT ScatterNet for efficient learning.
problem Training efficiency and feature extraction in CNNs.
method Replaces CNN's first layers with a parametric log based DTCWT ScatterNet.
result Improves training efficiency and learns complex patterns early.
CNNs encode data augmentation transformations, especially in early layers.
problem Whether neural network features encode data augmentation transformations.
method Systematic approach using pre-trained vision models to predict augmentation transformations.
result Neural network features encode data augmentation transformations, especially in early layers.
Paper proposes DeCEF layers to reduce CNN complexity.
problem Reduces complexity of CNNs without pre-trained models.
method Develops Depthwise Convolutional Eigen-Filter (DeCEF) layers.
result Achieves similar or higher accuracy with 2/3 parameters and 2/3 FLOPs.
CNN model predicts fluvial floods quickly and accurately.
problem Real-time flood prediction is computationally demanding.
method Deep Convolutional Neural Network (CNN) trained on 2D hydraulic model outputs.
result CNN model outperforms SVR in predicting flood inundation.
Improved patent classification using fine-tuned BERT model.
problem Classifying large patent datasets efficiently and accurately.
method Fine-tuning a pre-trained BERT model on patent claims.
result Outperforms state-of-the-art methods by 20%.
ConvTimeNet is a pre-trained CNN for time series classification.
problem Training deep neural networks for time series classification requires careful tuning and resources.
method ConvTimeNet is a pre-trained deep convolutional neural network trained on diverse univariate time series datasets. It adapts to new tasks with minimal fine-tuning.
result ConvTimeNet achieves significant gains in classification accuracy and computational efficiency compared to existing methods.
Develops masks to explain neural network predictions.
problem Improving neural network interpretability for various applications.
method Creates explanation masks for pre-trained networks using a secondary network.
result Demonstrates the effectiveness of the method across different types of networks.
Deep 3D CNN predicts Alzheimer's disease from brain images.
problem Early diagnosis of Alzheimer's disease using brain images.
method 3D convolutional neural network (3D-CNN) trained on autoencoder.
result 3D-CNN outperforms conventional classifiers in AD prediction.
The paper proposes new ways to enforce orthogonality in deep CNNs training.
problem Enforcing orthogonality in deep CNNs training.
method Developed novel orthogonality regularizations using analytical tools like mutual coherence and restricted isometry property.
result Consistent performance gains in final accuracies and convergence speed after applying regularizations.
Improved vehicle classification using ResNets and spatial pooling.
problem Fine-grained vehicle classification using ResNet architectures.
method Training ResNet-18, -34, and -50 on Comprehensive Cars dataset. Adding Spatially Weighted Pooling and localisation.
result Combining Spatially Weighted Pooling and localisation increases top-1 accuracy to 96.351%.
Simpler CNN model with spatial attention and temporal pooling outperforms complex models.
problem Emotion recognition from videos with small face deformations and identity variations.
method Spatial attention mechanism and temporal softmax pooling applied to a pre-trained CNN.
result The approach achieves higher accuracy than state-of-the-art methods on the EmotiW dataset.
Deep 3D CNN predicts Alzheimer's disease with high accuracy.
problem Early diagnosis of Alzheimer's disease using brain image features.
method Adaptable 3D convolutional neural network (3D-CNN) trained on autoencoder.
result 3D-CNN outperforms conventional classifiers in AD prediction.
Reduces annotation costs in medical imaging by 50%.
problem Challenges in creating large annotated datasets for medical imaging.
method Integrates active learning and transfer learning into a single framework.
result Reduces annotation efforts by at least half.
Improved 3D MRI classification using contrastive learning with continuous proxy metadata.
problem Insufficient labelled data for 3D medical image classification.
method Proposed a new loss function (y-Aware InfoNCE) to leverage continuous proxy metadata in contrastive learning.
result 3D CNN model pre-trained on 10^4 multi-site healthy brain MRI scans outperforms fully-supervised methods.
Paper presents a CNN-RNN method for multi-dimensional emotion recognition in-the-wild.
problem Dimensional emotion recognition in real-world scenarios.
method Pre-training with Aff-Wild and Aff-Wild2, extracting low-, mid-, and high-level features, using RNN subnets in a multi-task framework, and fusion of networks.
result Our approach outperformed state-of-the-art methods using only visual information.
CNNs help diagnose diabetic retinopathy by localizing lesions.
problem Diabetic retinopathy diagnosis requires identifying lesions in fundus images.
method Post-attention technique (Grad-CAM) on deep learning models' penultimate layer.
result InceptionV3 model achieves best performance and localizes lesions better.
Two KG-based methods explain transfer learning in CNN and ZSL.
problem Uninterpretable transfer learning for non-ML experts.
method Knowledge Graph-based explanation for transferability and model justification.
result Rich, human-understandable explanations for transfer learning.
CNN-PCA method uses deep learning to parameterize complex geological models.
problem Representing complex geological models in a low-dimensional space.
method CNN-PCA method combines PCA and CNN to honor geological features.
result CNN-PCA provides high-quality realizations and history matching results.
Images have become one of the most popular types of media through which users convey their emotions within online social networks. Although vast amount of research is devoted to sentiment analysis of textual data, there has been very limited work that focuses on analyzing sentiment of image data. In this work, we propo…
Simplified trust region method reduces representation change during fine-tuning.
problem Stability and representational collapse in fine-tuning pre-trained models.
method Replaces adversarial objectives with parametric noise in trust region theory.
result Matches or exceeds previous trust region methods in performance and speed.
Study shows variability in CNN predictions for medical images, suggesting ensemble averaging to reduce it.
problem Variability in CNN predictions for medical images due to data ordering during training.
method Reproduced CheXNet results with random seeds to identify variability in predictions.
result Averaging predictions from multiple models reduces variability by nearly 70%.
Collage inference uses redundancy to reduce cloud image classification latency variance.
problem Reducing latency variance in cloud image classification.
method Integrates collage-cnn for low-cost redundancy in multi-image classification.
result Significant reduction in 99th percentile tail latency and inference latency variation.
New CNN architecture detects face spoofing with deep local features.
problem Face recognition systems are vulnerable to face spoofing attacks.
method Two-step CNN architecture: first learns features from facial regions, then fine-tunes on whole images.
result Improves face spoofing detection performance and convergence speed.
Novel channel pruning method accelerates deep CNNs by removing redundant, similar features.
problem Limitations of existing magnitude-based pruning algorithms in similar magnitude cases.
method Hierarchical clustering of channels based on probabilistic similarity metrics, without sparsity training.
result 30% reduction in FLOPs for pruned ResNet-50 on ImageNet, outperforming baseline.
A new method to compress CNNs without changing architectures.
problem Redundancy in CNN filter weights.
method Replacing filters with a fixed set and a learnable 1D layer.
result Significant reduction in parameters and operations with minimal performance loss.
HexaConv improves CNN performance by using hexagonal filters and group convolutions.
problem Improving CNN performance by exploiting more symmetries.
method Implementing planar and group convolutions over hexagonal lattices.
result HexaConv outperforms conventional CNNs on aerial scene classification.
Transfer learning improves TSC models by 71 out of 85 datasets.
problem Improving deep CNNs for TSC using transfer learning.
method Pre-trained deep CNNs fine-tuned on various UCR datasets.
result Transfer learning can improve or degrade TSC model predictions.
A novel pruning method finds relevant units in CNNs for efficient compression.
problem Reduction of computation and storage costs in deep neural networks.
method Pruning by explaining, using relevance scores from explainable AI.
result The method efficiently compresses CNN models without sacrificing performance.
ConViT combines CNN and ViT strengths, improving image classification.
problem Combining the strengths of CNNs and ViTs while avoiding their limitations.
method Introducing GPSA, a form of positional self-attention with a soft convolutional inductive bias.
result ConViT outperforms DeiT on ImageNet while offering improved sample efficiency.
Kernel Quantization improves CNN compression without sacrificing performance.
problem Efficiently compressing CNN models without significant performance loss.
method Quantizes convolution kernels as the unit, learning a codebook for low-bit indexes.
result Significant compression ratio achieved with minimal accuracy loss.
Deep learning automates bacterial image classification.
problem Manual bacterial classification is time-consuming and error-prone.
method ResNet-50 pre-trained CNN architecture with transfer learning.
result Average classification accuracy of 99.2%.
CNNs can be improved by structurally perturbing inputs.
problem Understanding and improving the behavior of deep neural networks.
method Structural perturbations to input images to improve network predictions.
result Guided perturbations can enhance prediction performance of pre-trained CNNs.
Improved VGG networks enhance image classification accuracy.
problem Enhancing image classification accuracy using modified VGG architectures.
method Two improved VGG architectures were created by freezing the first two blocks and applying different dilation rates in the last three blocks.
result Significant out-performance on image classification tasks on CIFAR-10 and CIFAR-100 datasets.
A genetic algorithm-based method prunes deep neural networks for resource-constrained devices.
problem Resource constraints in deploying deep neural networks on embedded devices.
method Genetic algorithm for channel pruning of deep CNNs.
result Significant reduction in model size with minimal accuracy loss.
Proposes a deep learning churn prediction system for telecom using TL and meta-classification.
problem Churn prediction challenges in telecom due to large data, high dimensions, and imbalanced data.
method Transfer Learning (TL) and Ensemble-based Meta-Classification. Two stages: TL on Deep CNNs, then GP-AdaBoost meta-classifier.
result TL-DeepE system achieved 75.4% and 68.2% prediction accuracy on Orange and Cell2cell datasets, respectively.
Survey of deep learning for Hindi text classification.
problem Limited research on morphologically rich, low-resource Hindi text classification.
method Comparison of CNN, LSTM, Transformer, BERT, and LASER for Hindi text classification.
result Multilingual pre-trained sentence embeddings outperform traditional architectures for Hindi text classification.
Hybrid model predicts stock prices with high accuracy.
problem Complex volatility of stock market makes traditional models unsatisfactory.
method Attention-based CNN-LSTM and XGBoost integrated model.
result Hybrid model improves prediction accuracy.