An Analysis of Deep Neural Network Models for Practical Applications
arXiv:1605.07678
Abstract
Since the emergence of Deep Neural Networks (DNNs) as a prominent technique in the field of computer vision, the ImageNet classification challenge has played a major role in advancing the state-of-the-art. While accuracy figures have steadily increased, the resource utilisation of winning models has not been properly taken into account. In this work, we present a comprehensive analysis of important metrics in practical applications: accuracy, memory footprint, parameters, operations count, inference time and power consumption. Key findings are: (1) power consumption is independent of batch size and architecture; (2) accuracy and inference time are in a hyperbolic relationship; (3) energy constraint is an upper bound on the maximum achievable accuracy and model complexity; (4) the number of operations is a reliable estimate of the inference time. We believe our analysis provides a compelling set of information that helps design and engineer efficient DNNs.
7 pages, 10 figures, legend for Figure 2 got lost :/
References in corpus (4)
Cited by in corpus (140)
- Deep learning in agriculture: A survey
- Improved Regularization of Convolutional Neural Networks with Cutout
- Reconciling modern machine learning practice and the bias-variance trade-off
- Benchmark Analysis of Representative Deep Neural Network Architectures
- The History Began from AlexNet: A Comprehensive Survey on Deep Learning Approaches
- Computational bioacoustics with deep learning: a review and roadmap
- Transfer learning in hybrid classical-quantum neural networks
- Systematic evaluation of CNN advances on the ImageNet
- Sensitivity and Generalization in Neural Networks: an Empirical Study
- MCUNet: Tiny Deep Learning on IoT Devices
- Real-time Hand Tracking under Occlusion from an Egocentric RGB-D Sensor
- Compute and Energy Consumption Trends in Deep Learning Inference
- Tiny Machine Learning: Progress and Futures
- Dynamic Convolutions: Exploiting Spatial Sparsity for Faster Inference
- Classification of Urban Morphology with Deep Learning: Application on Urban Vitality
- Towards the Systematic Reporting of the Energy and Carbon Footprints of Machine Learning
- Survey of Machine Learning Accelerators
- Mobile Augmented Reality: User Interfaces, Frameworks, and Intelligence
- Speed/accuracy trade-offs for modern convolutional object detectors
- Apprentice: Using Knowledge Distillation Techniques To Improve Low-Precision Network Accuracy
- Generative Poisoning Attack Method Against Neural Networks
- CNN-based Methods for Object Recognition with High-Resolution Tactile Sensors
- Machine learning and AI-based approaches for bioactive ligand discovery and GPCR-ligand recognition
- Super-Resolution via Deep Learning
- Deep Meta-Learning: Learning to Learn in the Concept Space
- To understand deep learning we need to understand kernel learning
- Data augmentation instead of explicit regularization
- Silicon Photonic Architecture for Training Deep Neural Networks with Direct Feedback Alignment
- AI Accelerator Survey and Trends
- Machine and Deep Learning for IoT Security and Privacy: Applications, Challenges, and Future Directions
- What Do Compressed Deep Neural Networks Forget?
- On the Origin of Deep Learning
- Memorizing without overfitting: Bias, variance, and interpolation in over-parameterized models
- FingerVision Tactile Sensor Design and Slip Detection Using Convolutional LSTM Network
- Image-Based Size Analysis of Agglomerated and Partially Sintered Particles via Convolutional Neural Networks
- Modeling the Resource Requirements of Convolutional Neural Networks on Mobile Devices
- A Modern Take on the Bias-Variance Tradeoff in Neural Networks
- Great Power, Great Responsibility: Recommendations for Reducing Energy for Training Language Models
- Compact recurrent neural networks for acoustic event detection on low-energy low-complexity platforms
- FermiNets: Learning generative machines to generate efficient neural networks via generative synthesis
- Efficient Convolutional Neural Networks for Depth-Based Multi-Person Pose Estimation
- Restructuring Batch Normalization to Accelerate CNN Training
- On-Device Machine Learning: An Algorithms and Learning Theory Perspective
- Analyzing Green View Index and Green View Index best path using Google Street View and deep learning
- Accelerating SGD with momentum for over-parameterized learning
- TensorOpt: Exploring the Tradeoffs in Distributed DNN Training with Auto-Parallelism
- Disaster Monitoring using Unmanned Aerial Vehicles and Deep Learning
- VWA: Hardware Efficient Vectorwise Accelerator for Convolutional Neural Network
- Finite Versus Infinite Neural Networks: an Empirical Study
- An Automated Pipeline for Character and Relationship Extraction from Readers' Literary Book Reviews on Goodreads.com
- Gender Classification and Bias Mitigation in Facial Images
- Deep Learning for Spacecraft Pose Estimation from Photorealistic Rendering
- Impacts of DEM Type and Resolution on Deep Learning-Based Flood Inundation Mapping
- Checkmate: Breaking the Memory Wall with Optimal Tensor Rematerialization
- Deep learning in radiology: an overview of the concepts and a survey of the state of the art
- Real-time Convolutional Networks for Depth-based Human Pose Estimation
- Pilot Comparative Study of Different Deep Features for Palmprint Identification in Low-Quality Images
- Musical Chair: Efficient Real-Time Recognition Using Collaborative IoT Devices
- Energy-efficient neural network inference with microcavity exciton-polaritons
- Towards Compact and Robust Deep Neural Networks
- Towards On-Device AI and Blockchain for 6G enabled Agricultural Supply-chain Management
- Recent advances in deep learning theory
- A convolutional neural network approach for reconstructing polarization information of photoelectric X-ray polarimeters
- A survey of Object Classification and Detection based on 2D/3D data
- SegBlocks: Block-Based Dynamic Resolution Networks for Real-Time Segmentation
- Machine learning and AI research for Patient Benefit: 20 Critical Questions on Transparency, Replicability, Ethics and Effectiveness
- Advances in Deep Space Exploration via Simulators & Deep Learning
- An interpretable classifier for high-resolution breast cancer screening images utilizing weakly supervised localization
- Model Distillation with Knowledge Transfer from Face Classification to Alignment and Verification
- How Much Can We Really Trust You? Towards Simple, Interpretable Trust Quantification Metrics for Deep Neural Networks
- DarKnight: A Data Privacy Scheme for Training and Inference of Deep Neural Networks
- Minnorm training: an algorithm for training over-parameterized deep neural networks
- Unbiasing Semantic Segmentation For Robot Perception using Synthetic Data Feature Transfer
- Time for a Background Check! Uncovering the impact of Background Features on Deep Neural Networks
- Road Segmentation Using CNN with GRU
- CortexNet: a Generic Network Family for Robust Visual Temporal Representations
- Transferability of Adversarial Examples to Attack Cloud-based Image Classifier Service
- An Efficient Light-weight LSB steganography with Deep learning Steganalysis
- Improving the HardNet Descriptor
- Training Compact Neural Networks with Binary Weights and Low Precision Activations
- Leveraging Outdoor Webcams for Local Descriptor Learning
- Breaking the Activation Function Bottleneck through Adaptive Parameterization
- Surface Defect Classification in Real-Time Using Convolutional Neural Networks
- To Compress, or Not to Compress: Characterizing Deep Learning Model Compression for Embedded Inference
- Robustness and Transferability of Universal Attacks on Compressed Models
- Overhead Detection: Beyond 8-bits and RGB
- RL-GAN-Net: A Reinforcement Learning Agent Controlled GAN Network for Real-Time Point Cloud Shape Completion
- ICodeNet -- A Hierarchical Neural Network Approach for Source Code Author Identification
- PRGFlow: Benchmarking SWAP-Aware Unified Deep Visual Inertial Odometry
- Adaptive Selection of Deep Learning Models on Embedded Systems
- Learning Low-rank Deep Neural Networks via Singular Vector Orthogonality Regularization and Singular Value Sparsification
- Deep Learning in the Automotive Industry: Recent Advances and Application Examples
- Adaptive Pruning of Neural Language Models for Mobile Devices
- MGSim + MGMark: A Framework for Multi-GPU System Research
- Free-Space Optical Spiking Neural Network
- Neural Group Testing to Accelerate Deep Learning
- Cascaded Classifier for Pareto-Optimal Accuracy-Cost Trade-Off Using off-the-Shelf ANNs
- Refining the Structure of Neural Networks Using Matrix Conditioning
- Automated age-related macular degeneration area estimation -- first results
- Dubhe: Towards Data Unbiasedness with Homomorphic Encryption in Federated Learning Client Selection
- CADA: Multi-scale Collaborative Adversarial Domain Adaptation for Unsupervised Optic Disc and Cup Segmentation
- Testing the Efficient Network TRaining (ENTR) Hypothesis: initially reducing training image size makes Convolutional Neural Network training for image recognition tasks more efficient
- EnK: Encoding time-information in convolution
- GECKO: Reconciling Privacy, Accuracy and Efficiency in Embedded Deep Learning
- Thanks for Nothing: Predicting Zero-Valued Activations with Lightweight Convolutional Neural Networks
- An Experimental Analysis of the Power Consumption of Convolutional Neural Networks for Keyword Spotting
- Stochastic In-Face Frank-Wolfe Methods for Non-Convex Optimization and Sparse Neural Network Training
- Transfer Learning-Based Crack Detection by Autonomous UAVs
- Benchmarking Physical Performance of Neural Inference Circuits
- Distilling Pixel-Wise Feature Similarities for Semantic Segmentation
- Model Optimization for Deep Space Exploration via Simulators and Deep Learning
- Ship classification from overhead imagery using synthetic data and domain adaptation
- On the Bias-Variance Tradeoff: Textbooks Need an Update
- Evolutionary Cell Aided Design for Neural Network Architectures
- Fine-Grained Energy and Performance Profiling framework for Deep Convolutional Neural Networks
- Distributed Learning on Heterogeneous Resource-Constrained Devices
- ThumbNet: One Thumbnail Image Contains All You Need for Recognition
- Stochastic Model Pruning via Weight Dropping Away and Back
- CRL: Class Representative Learning for Image Classification
- FOSS: Multi-Person Age Estimation with Focusing on Objects and Still Seeing Surroundings
- Understanding and Optimizing Packed Neural Network Training for Hyper-Parameter Tuning
- Survey of Deep Learning Methods for Inverse Problems
- A Visual Domain Transfer Learning Approach for Heartbeat Sound Classification
- Resource Allocation via Model-Free Deep Learning in Free Space Optical Communications
- Memory Slices: A Modular Building Block for Scalable, Intelligent Memory Systems
- DeFungi: Direct Mycological Examination of Microscopic Fungi Images
- A Learning-based Method for Online Adjustment of C-arm Cone-Beam CT Source Trajectories for Artifact Avoidance
- Comparison of Neural Network Architectures for Spectrum Sensing
- A Benchmark Comparison of Visual Place Recognition Techniques for Resource-Constrained Embedded Platforms
- EmbNum: Semantic labeling for numerical values with deep metric learning
- Optimizing High Throughput Inference on Graph Neural Networks at Shared Computing Facilities with the NVIDIA Triton Inference Server
- Can a powerful neural network be a teacher for a weaker neural network?
- Auto Deep Compression by Reinforcement Learning Based Actor-Critic Structure
- Large-scale mammography CAD with Deformable Conv-Nets
- Stochastic Computing for Hardware Implementation of Binarized Neural Networks
- Active fall prevention: robotic vision in AAL
- Cylindrical Transform: 3D Semantic Segmentation of Kidneys With Limited Annotated Images
- Multi-objective Evolutionary Approach for Efficient Kernel Size and Shape for CNN
- Neural network scoring for efficient computing
- Handcrafted vs Deep Learning Classification for Scalable Video QoE Modeling