ACLI: A CNN Pruning Framework Leveraging Adjacent Convolutional Layer Interdependence and γ-Weakly Submodularity

Keyphrases

• Absolute Error 50%
• Benchmark Network 50%
• Computational Cost 50%
• Convolutional Layer 100%
• Convolutional Neural Network 100%
• Easy-to-implement 50%
• Efficiency Metrics 50%
• High Efficiency 50%
• Importance Function 100%
• Interdependence 100%
• L2 Norm 50%
• Low Quality Data 50%
• Network Parameters 50%
• Neural Network Pruning 100%
• Oblivious Algorithms 50%
• Order of Magnitude 50%
• Pruning Strategy 100%
• Resource Efficiency 50%
• Source-free 50%
• Submodular Function 50%
• Submodularity 100%
• Weak Submodularity 50%

Computer Science

• Computational Cost 50%
• Convolutional Layer 100%
• Convolutional Neural Network 100%
• Interdependence 100%
• Network Parameter 50%
• Oblivious Algorithm 50%
• Theoretical Framework 50%