Dynamic Stripes: Exploiting the Dynamic Precision Requirements of\n Activation Values in Neural Networks

2017/06/01 by Alberto Delmás, Delmas, Alberto, Patrick Judd +5 · 3 citations
Computer Science · Physics and Astronomy · #Advanced Neural Network Applications #FOS: Computer and information sciences #Machine Learning (cs.LG) #Model Reduction and Neural Networks #Neural Networks and Applications #Neural and Evolutionary Computing (cs.NE) #Parallel Computing and Optimization Techniques

paper · pdf · doi:10.48550/arxiv.1706.00504

openalex publication_date 2017/06/01 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Stripes is a Deep Neural Network (DNN) accelerator that uses bit-serial\ncomputation to offer performance that is proportional to the fixed-point\nprecision of the activation values. The fixed-point precisions are determined a\npriori using profiling and are selected at a per layer granularity. This paper\npresents Dynamic Stripes, an extension to Stripes that detects precision\nvariance at runtime and at a finer granularity. This extra level of precision\nreduction increases performance by 41% over Stripes.\n

Cited by

• PADE: A Predictor-Free Sparse Attention Accelerator via Unified Execution and Stage Fusion
• MCBP: A Memory-Compute Efficient LLM Inference Accelerator Leveraging Bit-Slice-enabled Sparsity and Repetitiveness
• Laconic Deep Learning Computing

Related

• DPRed: Making Typical Activation and Weight Values Matter In Deep Learning Computing
• Precision Gating: Improving Neural Network Efficiency with Dynamic Dual-Precision Activations
• Quantization Error as a Metric for Dynamic Precision Scaling in Neural Net Training
• Bit Fusion: Bit-Level Dynamically Composable Architecture for\n Accelerating Deep Neural Networks
• DASNet: Dynamic Activation Sparsity for Neural Network Efficiency Improvement