PRIMATE: Processing in Memory Acceleration for Dynamic Token-pruning Transformers

Keyphrases

• Primates 100%
• Transformer 100%
• Processing-in-memory 100%
• Low Quality Data 33%
• High Bandwidth Memory 33%
• Transformer Model 33%
• Superior Performance 16%
• Optimization Algorithm 16%
• Energy Efficiency 16%
• Space Efficiency 16%
• Memory Requirements 16%
• Data Reuse 16%
• Pruning 16%
• Underutilization 16%
• Promising Solutions 16%
• Attention Model 16%
• Selection Mechanism 16%
• Bandwidth Utilization 16%
• Hardware-software Co-design 16%
• Maximum Efficiency 16%
• Optimization Framework 16%
• Mapping Scheme 16%
• Data Movement 16%
• In-memory 16%
• Memory Bandwidth 16%
• Massive Parallelism 16%
• Movement Cost 16%
• Processing-In-Memory Architectures 16%
• Model Computation 16%
• Machine Learning Tasks 16%

Computer Science

• In-Memory Processing 100%
• Bandwidth Memory 33%
• Energy Efficiency 16%
• Memory Requirement 16%
• Data Reuse 16%
• Superior Performance 16%
• Parallelism 16%
• Bandwidth Utilization 16%
• Memory Bandwidth 16%
• Mapping Scheme 16%
• Optimization Framework 16%
• Learning System 16%
• Machine Learning 16%
• Computer Hardware 16%
• Hardware Co-Design 16%