Corpus ID: 203593200

AdaptivFloat: A Floating-point based Data Type for Resilient Deep Learning Inference

@article{Tambe2019AdaptivFloatAF,
  title={AdaptivFloat: A Floating-point based Data Type for Resilient Deep Learning Inference},
  author={Thierry Tambe and En-Yu Yang and Zishen Wan and Y. Deng and V. Reddi and Alexander M. Rush and D. Brooks and Gu-Yeon Wei},
  journal={ArXiv},
  year={2019},
  volume={abs/1909.13271}
}
Conventional hardware-friendly quantization methods, such as fixed-point or integer, tend to perform poorly at very low word sizes as their shrinking dynamic ranges cannot adequately capture the wide data distributions commonly seen in sequence transduction models. We present AdaptivFloat, a floating-point inspired number representation format for deep learning that dynamically maximizes and optimally clips its available dynamic range, at a layer granularity, in order to create faithful… Expand
Adaptive Posit: Parameter aware numerical format for deep learning inference on the edge
Quantized Neural Network Inference with Precision Batching
EdgeBERT: Optimizing On-Chip Inference for Multi-Task NLP
Value-Aware Error Detection and Correction for SRAM Buffers in Low-Bitwidth, Floating-Point CNN Accelerators
Pushing the Limits of Narrow Precision Inferencing at Cloud Scale with Microsoft Floating Point
A Survey of FPGA-Based Robotic Computing
Position-based Scaled Gradient for Model Quantization and Pruning

References

SHOWING 1-10 OF 49 REFERENCES
Rethinking floating point for deep learning
Flexpoint: An Adaptive Numerical Format for Efficient Training of Deep Neural Networks
Fixed Point Quantization of Deep Convolutional Networks
Accurate and Efficient 2-bit Quantized Neural Networks
Quantization and Training of Neural Networks for Efficient Integer-Arithmetic-Only Inference
Deep Positron: A Deep Neural Network Using the Posit Number System
Beating Floating Point at its Own Game: Posit Arithmetic
...
1
2
3
4
5
...