Deep Learning Through Examples: Arno Candel

Deep Learning
through Examples
Arno Candel
!
0xdata, H2O.ai
Scalable In-Memory Machine Learning
!
Silicon Valley Big Data Science Meetup,

Vendavo, Mountain View, 9/11/14
!
Who am I?
@ArnoCandel
PhD in Computational Physics, 2005

from ETH Zurich Switzerland
!
6 years at SLAC - Accelerator Physics Modeling

2 years at Skytree, Inc - Machine Learning
9 months at 0xdata/H2O - Machine Learning
!
15 years in HPC/Supercomputing/Modeling
!
Named 2014 Big Data All-Star by Fortune Magazine

!
H2O Deep Learning, @ArnoCandel
H2O DeepLearning:
Kaggle #1 rank (out of 413) - 40d left
#1
#17
Achieved with H2O Deep

Learning from R!
!
@matlabulous (Jo-fai Chow, Blend it like a Bayesian!) says:
I am 99.99999999999% sure that I can still go further with H2O.
Outline
Intro & Live Demo (10 mins)
Methods & Implementation (20 mins)
Results & Live Demos (25 mins)
Higgs boson detection
MNIST handwritten digits
text classification
Q & A (5 mins)
About H20 (aka 0xdata)

Java, Apache v2 Open Source
Join the www.h2o.ai/community!
#1 Java Machine Learning in Github
Customer Demands for

Practical Machine Learning
Requirements
Value
In-Memory
Fast (Interactive)
Distributed
Big Data (No Sampling)
Open Source
Ownership of Methods
API / SDK
Extensibility
H2O was developed by 0xdata from

scratch to meet these requirements
H2O Integration
Java
H2O
JSON
H2O
Scala
Python
H2O
YARN
Hadoop MR
HDFS
HDFS
HDFS
Standalone
Over YARN
On MRv1
H2O Architecture
Prediction Engine
R Engine
Distributed
In-Memory K-V store
Col. compression
Memory manager
MapReduce
Nano fast
Scoring Engine
Machine
Learning
Algorithms
e.g. Deep Learning
H2O - The Killer App on Spark
http://databricks.com/blog/2014/06/30/
sparkling-water-h20-spark.html
H2O DeepLearning on Spark

// Test if we can correctly learn A, B where Y = logistic(A + B*X)
test("deep learning log regression") {
val nPoints = 10000
val A = 2.0
val B = -1.5
!
!
// Generate testing data

val trainData = DeepLearningSuite.generateLogisticInput(A, B, nPoints, 42)
// Create RDD from testing data
val trainRDD = sc.parallelize(trainData, 2)
trainRDD.cache()
import H2OContext._
// Create H2O data frame (will be implicit in the future)
val trainH2ORDD = toDataFrame(sc, trainRDD)
// Create a H2O DeepLearning model
val dlParams = new DeepLearningParameters()
dlParams.source = trainH2ORDD
dlParams.response = trainH2ORDD.lastVec()
dlParams.classification = true
val dl = new DeepLearning(dlParams)
val dlModel = dl.train().get()
// Score validation data
val validationData = DeepLearningSuite.generateLogisticInput(A, B, nPoints, 17)
val validationRDD = sc.parallelize(validationData, 2)
val validationH2ORDD = toDataFrame(sc, validationRDD)
val predictionH2OFrame = new DataFrame(dlModel.score(validationH2ORDD))('predict)
val predictionRDD = toRDD[DoubleHolder](sc, predictionH2OFrame) // will be implicit in the future
// Validate prediction
validatePrediction( predictionRDD.collect().map (_.predict.getOrElse(Double.NaN)), validationData)
Brand-Sparkling-New Sneak Preview!
10
11
H2O R CRAN package
John Chambers (creator of the S language, R-core member)

names H2O R API in top three promising R projects
H2O + R = Happy Data Scientist

Machine Learning on Big Data with R:
Data resides on the H2O cluster!
12
13
Higgs Particle Discovery

Large Hadron Collider: Largest experiment of mankind!
$13+ billion, 16.8 miles long, 120 MegaWatts, -456F, 1PB/day, etc.
Higgs boson discovery (July 12) led to 2013 Nobel prize!
Higgs
vs
Background
Images courtesy CERN / LHC

http://arxiv.org/pdf/1402.4735v2.pdf
Machine Learning Meets Physics
Or rather: Back to the roots

(WWW was invented at CERN in 89)
14
Higgs: Binary Classification Problem

Current methods of choice for physicists:
- Boosted Decision Trees
- Neural networks with 1 hidden layer
BUT: Must first add derived high-level features (physics formulae)
HIGGS UCI Dataset:
21 low-level features
AND
7 high-level derived features

Train: 10M rows, Test: 500k rows
Metric: AUC = Area under the ROC curve (range: 0.51, higher is better)
Algorithm
low-level H2O AUC
all features H2O AUC
Generalized Linear Model
0.596
0.684
Random Forest
0.764
Gradient Boosted Trees
0.753
Neural Net 1 hidden layer
0.760
add
derived
features
0.840
0.839
0.830
Higgs: Can Deep Learning Do Better?

Algorithm
low-level H2O AUC
all features H2O AUC
0.596
0.684
Random Forest
0.764
0.840
0.753
0.839
Neural Net 1 hidden layer
0.760
0.830
Deep Learning
<Your guess goes here>

reference paper results: baseline 0.733
Lets build a H2O Deep Learning model and

find out! (That was my last weekend)
15
What is Deep Learning?

Wikipedia:
Deep learning is a set of algorithms in
machine learning that attempt to model
high-level abstractions in data by using
architectures composed of multiple
non-linear transformations.
Example:
Prediction
Input data
(who is it?)
(image)
Facebook's DeepFace (Yann LeCun)
recognises faces as well as humans
16
What is NOT Deep

Linear models are not deep
(by definition)
Neural nets with 1 hidden layer are not deep

(only 1 layer - no feature hierarchy)
SVMs and Kernel methods are not deep

(2 layers: kernel + linear)
Classification trees are not deep

(operate on original input space, no new features generated)
17
18
Deep Learning is Trending

Google trends
2009
2011
2013
Businesses are using

Deep Learning techniques!
Google Brain (Andrew Ng, Jeff Dean & Geoffrey Hinton)
!
FBI FACE: $1 billion face recognition project
!
Chinese Search Giant Baidu Hires Man Behind the Google Brain (Andrew Ng)
Deep Learning History

slides by Yan LeCun (now Facebook)
Deep Learning wins competitions

AND
makes humans, businesses and
machines (cyborgs!?) smarter
19
Deep Learning in H2O

1970s multi-layer feed-forward Neural Network
(supervised learning with stochastic gradient descent using back-propagation)
+ distributed processing for big data

(H2O in-memory MapReduce paradigm on distributed data)
+ multi-threaded speedup
(H2O Fork/Join worker threads update the model asynchronously)
+ smart algorithms for accuracy

(weight initialization, adaptive learning rate, momentum, dropout regularization,
l1/L2 regularization, grid search, checkpointing, auto-tuning, model averaging)
= Top-notch prediction engine!
20
Example Neural Network

fully connected directed graph of neurons
input/output neuron
hidden neuron
information flow
age
married
income
single
employment
Input layer
#neurons
#connections 3x4
Hidden
layer 1
Hidden
layer 2
3
4x3
Output layer
2
3x2
21
22
Prediction: Forward Propagation

neurons activate each other via weighted sums
pl is a non-linear function of xi:
can approximate ANY function
with enough layers!
age
income
zk
wkl
pl
married
per-class probabilities
sum(pl) = 1
xi
uij
yj
vjk
single
employment
yj = tanh(sumi(xi*uij)+bj)
zk = tanh(sumj(yj*vjk)+ck)
activation function: tanh
alternative:
x -> max(0,x) rectifier
bj, ck, dl: bias values

(indep. of inputs)
pl = softmax(sumk(zk*wkl)+dl)
softmax(xk) = exp(xk) / sumk(exp(xk))
23
Data preparation & Initialization

Neural Networks are sensitive to numerical noise,
operate best in the linear regime (not saturated)
age
income
wkl
married
xi
single
employment
Automatic standardization of data
Automatic initialization of weights
xi: mean = 0, stddev = 1
Poor mans initialization: random weights wkl
horizontalize categorical variables, e.g.
{full-time, part-time, none, self-employed}
Default (better): Uniform distribution in
->
+/- sqrt(6/(#units + #units_previous_layer))
{0,1,0} = part-time, {0,0,0} = self-employed
Training: Update Weights & Biases

For each training row, we make a prediction and compare
with the actual label (supervised learning):
predicted actual
0.8
married
1
0.2
single
0
Objective: minimize prediction error (MSE or cross-entropy)
Mean Square Error = (0.22 + 0.22)/2 penalize differences per-class
!
Cross-entropy = -log(0.8)
strongly penalize non-1-ness

1
Stochastic Gradient Descent: Update weights and biases via

gradient of the error (via back-propagation):
E
w < w - rate * E/w

rate
24
25
Backward Propagation
How to compute E/wi for wi < wi - rate * E/wi ?
Naive: For every i, evaluate E twice at (w1,,wi,,wN) Slow!
Backprop: Compute E/wi via chain rule going backwards
net = sumi(wi*xi) + b
xi
y = activation(net)
wi
E = error(y)
E/wi = E/y * y/net * net/wi

= (error(y))/y * (activation(net))/net * xi
H2O Deep Learning Architecture

initial model: weights and biases w
H2O atomic
in-memory
K-V store
K-V
w
nodes/JVMs: sync
threads: async
each node trains a

copy of the weights
HTTPD
i
map:
(some* or all of) its

w
and biases with
local data with

asynchronous F/J
threads
communication
w1
w3
w2
w4
2
reduce:
model averaging:
w1+w3
Query & display

the model via
JSON, WWW
K-V
w2+w4
1
w*
= (w1+w2+w3+w4)/4
1
average weights and

biases from all nodes,
speedup is at least
#nodes/log(#rows)
arxiv:1209.4129v3
HTTPD
updated model: w*
Keep iterating over the data (epochs), score from time to time
*auto-tuned
(default) or user-specified number of points per MapReduce iteration
26
Secret Sauce to Higher Accuracy

Adaptive learning rate - ADADELTA (Google)
Automatically set learning rate for each neuron
based on its training history
Regularization
L1: penalizes non-zero weights
L2: penalizes large weights
Dropout: randomly ignore certain inputs
Grid Search and Checkpointing
Run a grid search to scan many hyperparameters, then continue training the most
promising model(s)
27
28
Detail: Adaptive Learning Rate

!
Compute moving average of wi2 at time t for window length rho:
!
E[wi2]t = rho * E[wi2]t-1 + (1-rho) * wi2
!
Compute RMS of wi at time t with smoothing epsilon:
!
RMS[wi]t = sqrt( E[wi2]t + epsilon )
Do the same for E/wi, then
obtain per-weight learning rate:
rate(wi, t) =
RMS[wi]t-1
RMS[E/wi]t
Adaptive acceleration / momentum:

accumulate previous weight updates,
but over a window of time
Adaptive annealing / progress:
Gradient-dependent learning rate,
moving window prevents freezing
cf. ADADELTA paper
(unlike ADAGRAD: no window)
Detail: Dropout Regularization

Training:
For each hidden neuron, for each training sample, for each iteration,
ignore (zero out) a different random fraction p of input activations.
age
income
married
single
employment
Testing:
Use all activations, but reduce them by a factor p
(to simulate the missing activations during training).
cf. Geoff Hinton's paper
29
MNIST: digits classification

MNIST = Digitized handwritten
digits database (Yann LeCun)
Yann LeCun: Yet another advice: don't get fooled
by people who claim to have a solution to
Artificial General Intelligence. Ask them what
error rate they get on MNIST or ImageNet.
Data: 28x28=784 pixels with

(gray-scale) values in 0255
Train: 60,000 rows
Test:
10,000 rows
784 integer columns

784 integer columns
10 classes
10 classes
Standing world record:

Without distortions or convolutions,
the best-ever published error rate on
test set: 0.83% (Microsoft)
Lets see how H2O does on the MNIST dataset!
30
H2O Deep Learning on MNIST:

0.87% test set error (so far)
test set error: 1.5% after 10 mins
1.0% after 1.5 hours
0.87% after 4 hours
World-class
results!
No pre-training
No distortions
No convolutions
No unsupervised
training
Frequent errors: confuse 2/7 and 4/9
Running on 4
nodes with 16
cores each
31
H2O Deep Learning, A. Candel
Weather Dataset
Predict RainTomorrow from Temperature,
Humidity, Wind, Pressure, etc.
32
H2O Deep Learning, A. Candel
Live Demo: Weather Prediction

5-fold cross validation
3 hidden Rectifier
layers, Dropout,
L1-penalty
Interactive ROC curve with

real-time updates
12.7% 5-fold cross-validation error is at

least as good as GBM/RF/GLM models
33
Live Demo: Grid Search

How did I find those parameters? Grid Search!
(works for multiple hyper parameters at once)
Then continue training

the best model
34
Text Classification
Goal: Predict the item from

sellers text description
Vintage 18KT gold Rolex 2 Tone
in great condition
Data: Binary word vector 0,0,1,0,0,0,0,0,1,0,0,0,1,,0

gold
Train: 578,361 rows

Test: 64,263 rows
vintage condition
8,647 cols
8,647 cols
467 classes
143 classes
Lets see how H2O does on the ebay dataset!
35
36
Text Classification
Train: 578,361 rows
Test: 64,263 rows
8,647 cols
8,647 cols
467 classes
143 classes
Out-Of-The-Box: 11.6% test set error after 10 epochs!

Predicts the correct class (out of 143) 88.4% of the time!
Note 1: H2O columnar-compressed in-memory

store only needs 60 MB to store 5 billion
values (dense CSV needs 18 GB)
Note 2: No tuning was done
(results are for illustration only)
Parallel Scalability
(for 64 epochs on MNIST, with 0.87% parameters)
Training Time
in minutes
100
Speedup
40.00
75
30.00
50
20.00
25
10.00
2.7 mins
0.00
0
1
8 16 32 63
H2O Nodes
2 4 8 16 32 63
H2O Nodes
(4 cores per node, 1 epoch per node per MapReduce)
37
Deep Learning Auto-Encoders for

Anomaly Detection
Toy example:
Find anomaly in ECG heart
beat data. First, train a
model on whats normal:
20 time-series samples of
210 data points each
Deep Auto-Encoder:
Learn low-dimensional
non-linear structure of
the data that allows to
reconstruct the orig. data
Also for categorical data!
38
Deep Learning Auto-Encoders for

Anomaly Detection
+
Model of whats normal
Test set with anomaly
=>
Test set prediction is
reconstruction, looks normal
Found anomaly! large

reconstruction error
39
H2O brings Deep Learning to R

All parameters are
available from R
R Vignette with
example R scripts
http://0xdata.com/h2o/algorithms/
40
41
POJO Model Export for

Production Scoring
Plain old Java code is
auto-generated to take
your H2O Deep Learning
models into production!
Higgs Particle Discovery with H2O

How well did H2O
Deep Learning do?
<Your guess goes here>
reference paper results
Any guesses for AUC on low-level features?

AUC=0.76 was the best for RF/GBM/NN (H2O)
Lets see how H2O did in the past 30 minutes!
42
H2O Steam: Scoring Platform

http://server:port/steam/index.html
Higgs Dataset Demo on 10-node cluster

Lets score all our H2O models and compare them!
Live Demo
43
Scoring Higgs Models in H2O Steam
Live Demo on 10-node cluster:

<10 minutes runtime for all algos!
Better than LHC baseline of AUC=0.73!
44
45
Higgs Particle Detection with H2O

HIGGS UCI Dataset:
21 low-level features
7
AND
high-level derived features
Train: 10M rows, Test: 500k rows

*Nature
Algorithm
Papers
paper:
http://arxiv.org/pdf/1402.4735v2.pdf
low-level all features
Parameters (not heavily tuned),
l-l AUC
H2O AUC
H2O AUC
H2O running on 10 nodes
0.596
0.684
default, binomial
Random Forest
0.764
0.840
50 trees, max depth 50
0.73
0.753
0.839
50 trees, max depth 15
Neural Net 1 layer
0.733
0.760
0.830
1x300 Rectifier, 100 epochs
Deep Learning 3 hidden layers
0.836
0.850
3x1000 Rectifier, L2=1e-5, 40 epochs
0.868
0.869
4x500 Rectifier, L1=L2=1e-5, 300 epochs
0.880 running
6x500 Rectifier, L1=L2=1e-5
Deep Learning on low-level features alone beats everything else!

H2O prelim. results compare well with papers results*
(TMVA & Theano)
46
Tips
for
H
O
Deep
Learning
2
!
General:
More layers for more complex functions (exp. more non-linearity).
More neurons per layer to detect finer structure in data (memorizing).
Add some regularization for less overfitting (lower validation set error).
Specifically:
Do a grid search to get a feel for convergence, then continue training.
Try Tanh/Rectifier, try max_w2=1050, L1=1e-5..1e-3 and/or L2=1e-51e-3
Try Dropout (input: up to 20%, hidden: up to 50%) with test/validation
set. Input dropout is recommended for noisy high-dimensional input.
Distributed: More training samples per iteration: faster, but less accuracy?
With ADADELTA: Try epsilon = 1e-4,1e-6,1e-8,1e-10, rho = 0.9,0.95,0.99
Without ADADELTA: Try rate = 1e-41e-2, rate_annealing = 1e-51e-9,
momentum_start = 0.50.9, momentum_stable = 0.99,
momentum_ramp = 1/rate_annealing.
Try balance_classes = true for datasets with large class imbalance.
Enable force_load_balance for small datasets.
Enable replicate_training_data if each node can h0ld all the data.
Extensions for H2O Deep Learning

- Vision: Convolutional & Pooling Layers PUB-644
- Anomaly Detection PUB-806
- Pre-Training: Stacked Auto-Encoders PUB-1014
- Faster Training: GPGPU support PUB-1013
- Language/Sequences: Recurrent Neural Networks
- Benchmark vs other Deep Learning packages
- Investigate other optimization algorithms
Contribute to H2O!
Add your own JIRA tickets!
47
Key Take-Aways
H2O is a distributed in-memory data science
platform. It was designed for high-performance
machine learning applications on big data.
!
H2O Deep Learning is ready to take your advanced

analytics to the next level - Try it on your data!
!
Join our Community and Meetups!

https://github.com/h2oai
http://docs.h2o.ai
www.h2o.ai/community
@h2oai
Thank you!
48

Deep Learning Through Examples: Arno Candel

Hochgeladen von

Dokumentinformationen

Originalbeschreibung:

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Deep Learning Through Examples: Arno Candel

Hochgeladen von

Copyright:

Verfügbare Formate

Deep Learning

Silicon Valley Big Data Science Meetup,

PhD in Computational Physics, 2005

6 years at SLAC - Accelerator Physics Modeling

Named 2014 Big Data All-Star by Fortune Magazine

H2O Deep Learning, @ArnoCandel

Achieved with H2O Deep

H2O Deep Learning, @ArnoCandel

H2O Deep Learning, @ArnoCandel

About H20 (aka 0xdata)

H2O Deep Learning, @ArnoCandel

Customer Demands for

Big Data (No Sampling)

H2O was developed by 0xdata from

H2O Deep Learning, @ArnoCandel

H2O Deep Learning, @ArnoCandel

H2O Deep Learning, @ArnoCandel

H2O - The Killer App on Spark

H2O Deep Learning, @ArnoCandel

H2O DeepLearning on Spark

// Generate testing data

Brand-Sparkling-New Sneak Preview!

H2O Deep Learning, @ArnoCandel

H2O R CRAN package

John Chambers (creator of the S language, R-core member)

H2O Deep Learning, @ArnoCandel

H2O + R = Happy Data Scientist

H2O Deep Learning, @ArnoCandel

Higgs Particle Discovery

Images courtesy CERN / LHC

Machine Learning Meets Physics

Or rather: Back to the roots

H2O Deep Learning, @ArnoCandel

Higgs: Binary Classification Problem

7 high-level derived features

low-level H2O AUC

all features H2O AUC

Generalized Linear Model

Gradient Boosted Trees

Neural Net 1 hidden layer

H2O Deep Learning, @ArnoCandel

Higgs: Can Deep Learning Do Better?

low-level H2O AUC

all features H2O AUC

Generalized Linear Model

Gradient Boosted Trees

Neural Net 1 hidden layer

<Your guess goes here>

Lets build a H2O Deep Learning model and

H2O Deep Learning, @ArnoCandel

What is Deep Learning?

H2O Deep Learning, @ArnoCandel

What is NOT Deep

Neural nets with 1 hidden layer are not deep

SVMs and Kernel methods are not deep

Classification trees are not deep

H2O Deep Learning, @ArnoCandel

Deep Learning is Trending

Businesses are using

H2O Deep Learning, @ArnoCandel

Deep Learning History

Deep Learning wins competitions