13. GTC (SAN JOSE)
Fast and Easy Hyper-Parameter Grid Search for Deep Learning
13
Dr. Mark Whitney
14. NNHyper-Parameter Optimization
input
conv
conv
pool
fullyconn
softmax
● Large set of candidate architectures
● Search space with many GPUs, find most accurate
input
conv
conv
pool
fullyconn
softmax
conv
pool
input
conv
conv
pool
fullyconn
softmax
fullyconn
input
conv
conv
pool
fullyconn
softmax
input
conv
conv
pool
fullyconn
softmax
fullyconn
GPUaccelerated clusters
Image Classification
Labeled trainingimages
Train model on
GPU accelerated
cluster
Trained
Network
CAT
input
conv
conv
pool
fully conn
softmax
Neural Network Library
model . add( Convol ut i on2D( 128, 3, 3)
model . add( Dr opout ( 0. 4) )
. . .
Model definition
DNNOptimization on Rescale
https://platform.rescale.com
mark@rescale.com
DNNの最適化に、Rescaleのパラメータスタディー機能を利用
基本的なモデルに比べて10%程度精度が改善
25
Fast and Easy Hyper-Parameter Grid Search for Deep Learning
Rescale Confidential
28. Dr. Mark Whitney
• Dr. Mark Whitney leads Rescale's Deep
Learning Cloud
– Integrating machine learning workflows with state-of-
the-art GPU clusters.
– Prior to this, he started the Energy Demand
Forecasting Group at EnergyHub, providing smart-
grid analytics to electric utilities.
– Mark received his Ph.D. from UC Berkeley in
quantum computing.
28
長尾はMarkに会った
こと無いですがメール
の雰囲気をみるととて
も親切でフレンドリー
です!