2019
DOI: 10.1007/978-3-030-14880-5_1
|View full text |Cite
|
Sign up to set email alerts
|

Sparsity in Deep Neural Networks - An Empirical Investigation with TensorQuant

Abstract: Deep learning is finding its way into the embedded world with applications such as autonomous driving, smart sensors and augmented reality. However, the computation of deep neural networks is demanding in energy, compute power and memory. Various approaches have been investigated to reduce the necessary resources, one of which is to leverage the sparsity occurring in deep neural networks due to the high levels of redundancy in the network parameters. It has been shown that sparsity can be promoted specifically… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
2
1
1

Citation Types

0
6
0

Year Published

2020
2020
2023
2023

Publication Types

Select...
2
2
1

Relationship

0
5

Authors

Journals

citations
Cited by 5 publications
(6 citation statements)
references
References 19 publications
0
6
0
Order By: Relevance
“…Furthermore, source level instrumentation might interact with the compiler's ability to optimize the code and therefore the result of the emulation might by inaccurate in the context of evaluating hardware design. TensorQuant [10], proposes two source level approaches, intrinsic (fine-grain) and extrinsic (coarse-grain), to emulate low precision using Tensorflow. The extrinsic approach is an approximation where the rounding process is done just on high level operators like convolutions.…”
Section: Introductionmentioning
confidence: 99%
“…Furthermore, source level instrumentation might interact with the compiler's ability to optimize the code and therefore the result of the emulation might by inaccurate in the context of evaluating hardware design. TensorQuant [10], proposes two source level approaches, intrinsic (fine-grain) and extrinsic (coarse-grain), to emulate low precision using Tensorflow. The extrinsic approach is an approximation where the rounding process is done just on high level operators like convolutions.…”
Section: Introductionmentioning
confidence: 99%
“…Apart from the property of saturation of the activation functions, another property that is important in the context of training deep networks is sparsity [13] . Sparse networks are important because they have less number of parameters and hence are easier to train and less prone to the problem of overfitting thus giving better generalization performance.…”
Section: Introductionmentioning
confidence: 99%
“…Firstly, we apply quantization for each output matrix element to represent the numbers using BF16 and RNE rounding (coarse-grain quantization label) over the reference result. This is akin to the coarsegrain methods used by QPyTorch [104] and TensorQuant [59]. Secondly, we attach FASE to the benchmark binary, which instruments the code from the dynamically linked Intel MKL library.…”
Section: Emulation Accuracy Methodologymentioning
confidence: 99%
“…TensorQuant [59], proposes two source-level approaches, intrinsic and extrinsic, to emulate low precision using Tensorflow. The extrinsic approach is an approximation where the rounding process is done just on high-level operators like convolutions.…”
Section: Hardware Setupmentioning
confidence: 99%
See 1 more Smart Citation