Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Neural Networks
Search
Cheesecake Labs
March 11, 2019
Technology
0
38
Neural Networks
Fred
Cheesecake Labs
March 11, 2019
Tweet
Share
More Decks by Cheesecake Labs
See All by Cheesecake Labs
Cats' wellness & care
cheesecakelabs
0
38
How do we create the first impressions?
cheesecakelabs
0
42
Menstrual cup: suit and freedom
cheesecakelabs
0
43
Life is a cycle, better with a bicycle
cheesecakelabs
0
37
Interview Process: how to get the best of people
cheesecakelabs
1
64
My capsule wardrobe experience
cheesecakelabs
3
48
Stonewall Rebellion and its impact on LGBTQIA+ history
cheesecakelabs
1
27
Pregnancy, childbirth and breastfeeding: What do I have to do with it?
cheesecakelabs
0
38
MBTI - Psychological types described by Jung
cheesecakelabs
0
88
Other Decks in Technology
See All in Technology
Delivering Millions of Messages within seconds @ Duolingo
pelelgrino
0
350
一生覚えておきたい「システム開発=コミュニケーション」〜初めての実務案件振り返りLT〜
maimyyym
1
170
Kernel MemoryでAzure OpenAI Serviceとお手軽データソース連携
mitsuzono
1
260
LangSmith入門―トレース/評価/プロンプト管理などを担うLLMアプリ開発プラットフォーム
os1ma
3
380
Cypress or Playwright?
rainerhahnekamp
0
110
FrontDoorとWebAppsを組み合わせた際のリダイレクト処理の注意点
kenichirokimura
1
530
どうするコスト最適化のトレードオフ
tetsuyaooooo
1
530
地理空間データ可視化・解析・活用ソリューション Pacific Spatial Solutions (PSS)
pacificspatialsolutions
0
300
競技としてのKaggle、役に立つKaggle
yu4u
4
1.9k
LayerXにおけるLLMプロダクト開発の今までとこれから
layerx
PRO
1
400
20240418_Google ColabにLLMが搭載されたようなのでPython x データ分析の勉強方法を考えてみる
doradora09
0
140
[新卒向け研修資料] テスト文字列に「うんこ」と入れるな(2024年版)
infiniteloop_inc
4
16k
Featured
See All Featured
It's Worth the Effort
3n
180
27k
Docker and Python
trallard
34
2.7k
Six Lessons from altMBA
skipperchong
21
3k
Building Your Own Lightsaber
phodgson
99
5.7k
Design by the Numbers
sachag
274
18k
Designing for humans not robots
tammielis
248
25k
Writing Fast Ruby
sferik
621
60k
Building Flexible Design Systems
yeseniaperezcruz
319
37k
The Illustrated Children's Guide to Kubernetes
chrisshort
31
46k
Why You Should Never Use an ORM
jnunemaker
PRO
51
8.6k
JavaScript: Past, Present, and Future - NDC Porto 2020
reverentgeek
40
4.4k
Producing Creativity
orderedlist
PRO
337
39k
Transcript
Neural Networks Brief introduction and sample case
"Artificial Intelligence"
AI in Popular Culture
AI Effect "AI is whatever hasn't been done yet." Douglas
Hofstadter "Every time we figure out a piece of it, it stops being magical; we say, 'Oh, that's just a computation.'" Rodney Brooks
Types of AI • Machine learning ◦ Neural networks ▪
Perceptron ▪ Recurrent neural network ▪ Convoluted neural network ◦ Support Vector Machines (SVM) • Fuzzy systems • Evolutionary algorithms ◦ Genetic algorithm ◦ Differential evolution • Swarm Intelligence • Probabilistic methods
Neural Networks
What is this ?
Perceptrons
Perceptrons
Perceptrons • x 1 – Is it raining? • x
2 – Does your girlfriend/boyfriend want to go? • x 3 – Is it near public transportation?
Perceptrons
Perceptrons
Perceptrons
Perceptrons
Perceptrons
Perceptrons
Neurons
Perceptrons
Neurons
Why is this relevant?
Learning!
Learning
Learning
Learning
Real world problem
Recognizing Handwritten Digits
Database The MNIST (Modified National Institute of Standards and Technology)
database Contains 60,000 training images and 10,000 testing images.
Database
Neural Network Architecture
Neural Network Architecture
Neural Network Architecture
Neural Network Architecture
Neural Network Architecture
Measuring Outcome
Neural Network Architecture
Cost Function
Learning
Gradient Descent
Gradient Descent
Gradient Descent
Gradient Descent
Gradient Descent
Gradient Descent
Backpropagation
Backpropagation
Let's CODE!
Repository
__init__()
feedforward()
sigmoid()
sigmoid()
Example use
Training Our Network
Learning
SGD()
update_mini_batch()
backprop()
cost_derivative()
Performance
95%
Current Record The current (2013) record is classifying 9,979 of
10,000 images correctly (99,79%).
How to improve?
Improving our Neural Network • Improved Cost Function ◦ Faster
Training ◦ Reduced Overfitting • Regularization • Dropout • Artificially expanding training data • Convolutional Networks
Thank you!
Acknowledgement Michael Nielsen NeuralNetworksAndDeepLearning.com
Do we have time?
Convolutional Neural Networks
Regular Networks
Convolutional Neural Networks
Convolutional Neural Networks
Convolutional Neural Networks
Convolutional Neural Networks
Feature Maps (Kernels)
Pooling Layers
Max-Pooling
Architecture