arxiv_cv 90% Match Research Paper ML engineers,Researchers in efficient deep learning,Mobile developers 3 days ago

FastBoost: Progressive Attention with Dynamic Scaling for Efficient Deep Learning

computer-vision › scene-understanding

📄 Abstract

Abstract: We present FastBoost, a parameter-efficient neural architecture that achieves state-of-the-art performance on CIFAR benchmarks through a novel Dynamically Scaled Progressive Attention (DSPA) mechanism. Our design establishes new efficiency frontiers with: CIFAR-10: 95.57% accuracy (0.85M parameters) and 93.80% (0.37M parameters) CIFAR-100: 81.37% accuracy (0.92M parameters) and 74.85% (0.44M parameters) The breakthrough stems from three fundamental innovations in DSPA: (1) Adaptive Fusion: Learnt channel-spatial attention blending with dynamic weights. (2) Phase Scaling: Training-stage-aware intensity modulation (from 0.5 to 1.0). (3) Residual Adaptation: Self-optimized skip connections (gamma from 0.5 to 0.72). By integrating DSPA with enhanced MBConv blocks, FastBoost achieves a 2.1 times parameter reduction over MobileNetV3 while improving accuracy by +3.2 percentage points on CIFAR-10. The architecture features dual attention pathways with real-time weight adjustment, cascaded refinement layers (increasing gradient flow by 12.7%), and a hardware-friendly design (0.28G FLOPs). This co-optimization of dynamic attention and efficient convolution operations demonstrates unprecedented parameter-accuracy trade-offs, enabling deployment in resource-constrained edge devices without accuracy degradation.

Authors (1)

JunXi Yuan

Submitted

November 2, 2025

arXiv Category

cs.CV

arXiv PDF

Key Contributions

Introduces FastBoost, a parameter-efficient neural architecture featuring a novel Dynamically Scaled Progressive Attention (DSPA) mechanism. DSPA incorporates adaptive fusion, phase scaling, and residual adaptation to achieve state-of-the-art accuracy on CIFAR benchmarks with significantly reduced parameter counts compared to existing models like MobileNetV3.

Business Value

Enables the deployment of high-performance computer vision models on resource-constrained devices (e.g., mobile phones, edge devices), leading to more efficient and powerful AI applications.

Paper Metadata

Innovation Type

Model Architecture & Attention Mechanism

Deployment Feasibility

High (designed for efficiency)

Limitations Addressed

High parameter count and computational cost of deep learning models while maintaining high accuracy.

Performance Gains

2.1 times parameter reduction over MobileNetV3 while improving accuracy by +3.2 percentage points on CIFAR-10. Achieves state-of-the-art accuracy with significantly fewer parameters.

Technical Tags

parameter-efficientattention mechanismdynamic scalingprogressive attentionCIFAR benchmarksMBConv blockschannel-spatial attentionphase scalingresidual adaptation

Research Topics

Deep Learning ArchitecturesModel EfficiencyComputer VisionAttention MechanismsNeural Network Design

Methods & Architectures

Dynamically Scaled Progressive Attention (DSPA)Adaptive FusionPhase ScalingResidual AdaptationEnhanced MBConv blocks FastBoostMBConv blocks

Applications & Tasks

Image Classification Computer Vision Mobile AI Model EfficiencyPerformance OptimizationParameter Reduction Image classificationAchieving high accuracy with fewer parameters

Datasets & Benchmarks

Datasets

CIFAR-10, CIFAR-100

Benchmarks

CIFAR-10: 95.57% accuracy (0.85M parameters) • CIFAR-10: 93.80% accuracy (0.37M parameters) • CIFAR-100: 81.37% accuracy (0.92M parameters) • CIFAR-100: 74.85% accuracy (0.44M parameters)

AccuracyNumber of parameters

Related Fields

Deep LearningMobile AIComputer VisionNeural Network Optimization

Keywords

efficient deep learningparameter efficiencyattentionneural network architecturecomputer visionCIFARmobile AIMBConvFastBoostdynamic scaling

Academic Context

#Deep Learning Architectures#Model Efficiency#Computer Vision#Attention Mechanisms#Neural Network Design

Commercial Potential

Potential Products

Efficient image classification models for mobile devicesLightweight computer vision backbones for various applications

Target Industries

TechnologyMobileConsumer ElectronicsAutomotive

Use Case Examples

Real-time image recognition on smartphonesEfficient object detection for embedded systems

Competitive Edge

Achieves superior accuracy-parameter trade-offs compared to existing efficient architectures like MobileNetV3.

Market Opportunity

Large market for efficient deep learning models in mobile and edge AI.

Revenue Models

Licensing of efficient model architectures

Resource Requirements

Compute Needs

Low to moderate (designed for efficiency)

Data Requirements

Image datasets (e.g., CIFAR)

Deployment Constraints

Limited by the computational capabilities of the target edge device.

Scalability

Designed for scalability on resource-constrained devices.

Production Readiness

Maturity Level

Research

Time to Market

1-2 years (for integration into products)

Patent Potential

Moderate (novel attention mechanism and architecture)

View Full Paper Back to Papers