TransCODE: Co-Design of Transformers and Accelerators for Efficient Training and Inference

Computer Science

• Participatory Design 100%
• Energy Consumption 66%
• Machine Learning 33%
• Design Technique 33%
• Lower Energy Consumption 33%
• Speed-up 33%
• Hardware Platform 33%
• Memory Consumption 33%
• Art Performance 33%
• Memory Resource 33%
• Memory Requirement 33%
• Training Process 33%
• Inference Model 33%
• Model Architecture 33%
• Compute Resource 33%
• Bidirectional Encoder Representations From Transformers 33%

Keyphrases

• Parallelizability 33%
• Dynamic Training 33%
• Machine Learning Evaluation 33%
• Energy chips 33%
• Resource Constrained Hardware 33%