using Monte Carlo Methods EMC2 Workshop Gonalo Mordido Hasso - - PowerPoint PPT Presentation

▶
using monte carlo methods
SMART_READER_LITE
LIVE PREVIEW

using Monte Carlo Methods EMC2 Workshop Gonalo Mordido Hasso - - PowerPoint PPT Presentation

May 18, 2023 102 likes •176 views

Instant Quantization of Neural Networks using Monte Carlo Methods EMC2 Workshop Gonalo Mordido Hasso Plattner Institute @ NeurIPS 2019 Matthijs Van Keirsbilck NVIDIA Alexander Keller NVIDIA 1 Motivation and idea neural network

slide-1
SLIDE 1

Instant Quantization of Neural Networks using Monte Carlo Methods

Gonçalo Mordido Matthijs Van Keirsbilck Alexander Keller

1

EMC2 Workshop @ NeurIPS 2019 Hasso Plattner Institute NVIDIA NVIDIA

slide-2
SLIDE 2

Gonçalo Mordido, Matthijs Van keirsbilck, Alexander Keller

  • neural network quantization/sparsity

○ lower cost: compute, memory, power, bandwidth, ...

  • quantization usually requires retraining
  • idea: use importance sampling

○ fast and efficient due to stratified sampling ○ sparsity and bit-width adjustable by the number of samples ○ no additional training

Motivation and idea

2

slide-3
SLIDE 3

Gonçalo Mordido, Matthijs Van keirsbilck, Alexander Keller

Monte Carlo Quantization (MCQ)

3

... ... full precision values PDF

slide-4
SLIDE 4

Gonçalo Mordido, Matthijs Van keirsbilck, Alexander Keller

Monte Carlo Quantization (MCQ)

4

... ... CDF integer values

slide-5
SLIDE 5

Gonçalo Mordido, Matthijs Van keirsbilck, Alexander Keller

Results

5

slide-6
SLIDE 6

Gonçalo Mordido, Matthijs Van keirsbilck, Alexander Keller

Monte Carlo Neural Networks

6

  • simple method to quantize/sparsify models

○ low accuracy loss ○ no retraining

  • general applicability

○ weights and/or activations ○ related to random walks

  • future work

○ quantized gradients ○ integer neural networks