PruneGNN: Algorithm-Architecture Pruning Framework for Graph Neural Network Acceleration

Keyphrases

• Algorithm Architecture 100%
• Algorithmic Level 10%
• Architectural Level 10%
• Architecture Framework 10%
• Architecture Pruning 100%
• Computational Challenges 10%
• Dimensionality Reduction 10%
• Dynamic Graph 10%
• Floating-point Operations 10%
• Graph Complexes 10%
• Graph Neural Network 100%
• High Sparsity 10%
• Inference Performance 10%
• Irregular Model 10%
• Latency Constraint 10%
• Matrix Operator 10%
• Model Pruning 10%
• Model Sparsity 10%
• Neural Network Acceleration 100%
• Neural Network Model 30%
• Neural Network Performance 10%
• Neural Network Pruning 10%
• NN Model 10%
• NVIDIA A100 10%
• Optimization Algorithm 10%
• Parallel Architecture 10%
• Parallelization 30%
• Performance Gain 10%
• Performance Implications 10%
• Performance Improvement 10%
• Performance Scalability 10%
• Pruned Models 10%
• Pruning 10%
• Pruning Strategy 20%
• Representative Graphs 10%
• Sparse Input 10%
• Sparse Matrix multiplication 10%
• Sparse Training 10%
• Sparsity 10%
• Static Graph 10%
• Structured Graph 10%
• Structured Model 10%
• Structured Pruning 30%
• Tight 10%
• Training Methods 10%
• Training Performance 10%
• Vectorized 20%

Computer Science

• Architecture Framework 10%
• Architecture Level 10%
• Computer Hardware 20%
• Deep Neural Network 10%
• Experimental Result 10%
• Floating-Point Operation 10%
• Graph Neural Network 100%
• Graphics Processing Unit 20%
• Level Parallelism 10%
• Matrix Multiplication 10%
• Matrix Operator 10%
• Multiplication Kernel 10%
• Network Performance 10%
• Neural Network Model 30%
• Parallel Architectures 10%
• Parallelism 10%
• Performance Gain 10%
• Performance Implication 10%
• Potential Parallelism 10%
• SIMD 10%
• Sparsity 30%
• Structured Model 10%