Machine Learning in Finance: Moving forward!

© 2018, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
© 2018, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Amazon Confidential and Trademark
Adrian Hornsby
Sr. Technical Evangelist | Amazon Web Services
Machine Learning in Finance:
Moving forward!
@adhorn

View Slide

Myth: ML is dark magic

View Slide

What are we talking about?
AI
RL
DL
Machine
Learning

View Slide

One of the ”Founding Father" of Artificial Intelligence
John McCarthy
1955

View Slide

Photo from the 1956 Dartmouth
Conference with Marvin Minsky,
Ray Solomonoff, Claude Shannon,
John McCarthy, Trenchard More,
Oliver Selfridge and Nathaniel
Rawchester

View Slide

Frank Rosenblatt, 1957
Perceptron

View Slide

Robert Schlaifer, 1959
Bayesian Decision Theory

View Slide

First known deep network
Alexey Grigorevich Ivakhnenko, 1965
Image of Prof. Alexey Ivakhnenko courtesy of Wikipedia.

View Slide

1969 First financial application

View Slide

Paul Werbos, 1975
Backpropagation

View Slide

Protrader expert system by K.C Chen and Ting-peng Lian
Chen and Lian were able to predict
the 87 point drop in Dow Jones
Industrial Average in 1986

View Slide

LeCun, 1989
First application of
backpropagation
https://www.youtube.com/watch?v=FwFduRA_L6Q

View Slide

Expert System Development – 90’s
• PlanPower & Client Profiling System created by Applied Expert
Systems
• providing tailored financial plans
• Chase Lincoln First Bank and Arthur D. Little Inc.
• investment & debt planning
• retirement & life-insurance planning
• budget recommendations,
• income tax planning and savings achievement
• U.S Department of Treasury created FinCEN Artificial Intelligence
system.
• money laundering.

View Slide

The curse of dimensionality

View Slide

The Advent of ML – 2000’s
Digital
Data
GPUs
& Acceleration
Cloud
Computing
Algorithms

View Slide

Explosion in AI and ML Use Cases
Image recognition and tagging for photo organization
Object detection, tracking and navigation for Autonomous Vehicles
Speech recognition & synthesis in Intelligent Voice Assistants
Algorithmic trading strategy performance improvement
Sentiment analysis for targeted advertisements

View Slide

The Amazon ML stack
A I S E R V I C E S
M L S E R V I C E S
M L F R A M E W O R K S &
I N F R A S T R U C T U R E
A m a z o n
S a g e Ma k e r G r o u n d T r u t h A l g o r i t h m s
N o t e b o o k s
M a r k e t p l a c e
U n s u p e r v i s e d
L e a r n i n g
S u p e r v i s e d
L e a r n i n g
R e i n f o r c e m e n t
L e a r n i n g
O p t i m i z a t i o n
( N e o )
T r a i n i n g
H o s t i n g
D e p l o y m e n t
Frameworks Interfaces Infrastructure
A m a z o n
R e k o g n i t i o n
I m a g e
A m a z o n
P o l l y
A m a z o n
T r a n s c r i b e
A m a z o n
T r a n s l a t e
A m a z o n
C o m p r e h e n d
A m a z o n
L e x
A m a z o n
V i d e o
Vision Speech Language Chatbots
A m a z o n
F o r e c a s t
Forecasting
A m a z o n
T e x t r a c t
A m a z o n
p e r s o n a l i z e
Recommendations
A m a z o n
E C 2 P 3
& P 3 D N
A m a z o n
E C 2 C 5
F P G A s A W S G r e e n g r a s s A m a z o n
E l a s t i c
I n f e r e n c e
A m a z o n
I n f e r e n t i a

View Slide

More machine learning happens on AWS

View Slide

We all want machine learning everywhere

View Slide

Supervised learning
Run an algorithm on a labelled data set, i.e. a data set containing
samples and answers. Gradually, the model learns how to correctly
predict the right answer. Regression and classification are examples of
supervised learning.
Unsupervised learning
Run an algorithm on an unlabelled data set, i.e. a data set containing
samples only. Here, the model progressively learns patterns in data
and organizes samples accordingly. Clustering and topic modeling are
examples of unsupervised learning.
Types of Machine Learning

View Slide

Supervised learning
SOPHISTICATION OF ML MODELS
AMOUNT OF TRAINING DATA REQUIRED

View Slide

Supervised learning

View Slide

Reinforcement learning
(RL)
Supervised learning

View Slide

Remember when you first learned this?

View Slide

Let’s start with Deep Learning?

View Slide

Predicting the price of a house with humans
Price
City
ZipCode Life Quality
Parking
Size
# Room
Accessibility
Family Friendly

View Slide

Predicting the price of a house with neural net
Price
City
ZipCode Life Quality
Parking
Size
# Room
Accessibility
Family Friendly
Input Output
Discovered by the neural network

View Slide

Deep Learning Training
Dataset The MNIST database of handwritten digits, available from this
page, has a training set of 60,000 examples, and a test set of 10,000
examples.
Human-in-the-loop

View Slide

https://ml4a.github.io/ml4a/neural_networks/

View Slide

https://ml4a.github.io/ml4a/neural_networks/
Learning

View Slide

Learning = Minimizing the loss (error) function
Backpropagation

View Slide

Early stopping
Training accuracy
Loss function
Accuracy
100%
Epochs
Validation accuracy
Loss
Best epoch
OVERFITTING

View Slide

Deep Learning is a big deal
It’s able to do better than other ML and Humans

View Slide

Convolutional Neural Networks (CNN)
Conv 1 Conv 2 Conv n
…
…
Feature Maps
Labrador
Dog
Beach
Outdoors
Softmax
Probability
Fully
Connected
Layer

View Slide

https://github.com/precedenceguo/mx-rcnn https://github.com/zhreshold/mxnet-yolo
CNN: Object Detection

View Slide

CNN: Object Detection

View Slide

CNN: Object Segmentation

View Slide

CNN: Text Detection and Recognition
https://github.com/Bartzi/stn-ocr

View Slide

https://github.com/tornadomeet/mxnet-face
CNN: Face Detection

View Slide

FDA-approved
medical imaging
https://www.periscope.tv/AWSstartups/1vAGRgevBXRJl
https://www.youtube.com/watch?v=WE81dncwnIc
CNN: Object Segmentation

View Slide

CNN: Real-Time Pose Estimation
https://github.com/dragonfly90/mxnet_Realtime_Multi-Person_Pose_Estimation

View Slide

CapsNet: Capsule Networks
Spatial Memory
https://medium.com/ai%C2%B3-theory-practice-business/understanding-hintons-capsule-networks-part-i-intuition-b4b559d1159b

View Slide

Long Short Term Memory Networks (LSTM)
Oh I remember!
https://github.com/awslabs/sockeye

View Slide

PredNet: Prediction Networks
What comes next
https://coxlab.github.io/prednet/

View Slide

Generative Adversarial Networks (GAN)
The future at work (already) today
Generating new ”celebrity” faces
https://github.com/tkarras/progressive_growing_of_gans

View Slide

Generative adversarial networks (GAN)
The future at work (already) today
Semantic labels → Cityscapes street views
https://tcwang0509.github.io/pix2pixHD/

View Slide

Wait! There’s more!
Models can also generate text from text, text from images, text from video,
images from text, sound from video, 3D models from 2D images, etc.

View Slide

How do people ”build” Neural Nets?

View Slide

Data Visualization &
Analysis
Business Problem –
ML problem framing Data Collection
Data Integration
Data Preparation &
Cleaning
Feature Engineering
Model Training &
Parameter Tuning
Model Evaluation
Are Business
Goals met?
Model Deployment
Monitoring &
Debugging
Yes
No
Data Augmentation
Feature
Augmentation
The ML Process
Re-training
Predictions

View Slide

AI/ML Workflow
Performance monitoring
& adaptation
8
Data Acquisition & Storage
1
Model & Framework selection
3
Model Training
4
Hyper parameter tuning
5
Model testing and simulation
6
Model Deployment (inference)
7
Data Labelling
2

View Slide

Model Zoos & Transfer Learning
• Full implementations of many state-of-the-art models reported in
the academic literature.
• Complete models, with scripts, pre-trained weights and
instructions on how to build and fine tune these models.

View Slide

https://mxnet.apache.org/model_zoo/index.html

View Slide

Transfer Learning (hidden gem)
• Initialize parameter with pre-trained model
• Use pre-trained model as fixed feature extractor and build model
based on feature
• Why?
It takes a long time and a lot of resources to train a neural network from
scratch.

View Slide

https://www.youtube.com/watch?v=qGotULKg8e0
• Over 10 million images from 300,000 hotels
• Fine-tuned a pre-trained CNN using 100,000 images
• Hotel descriptions now automatically feature the
best available images
Ranking Hotel Images Using
Transfer Learning
https://news.developer.nvidia.com/expedia-ranking-hotel-images-with-deep-learning/

View Slide

Let’s talk about
Reinforcement Learning (RL)

View Slide

Defining Reinforcement Learning
Source: Wikipedia

View Slide

The players
Simulation
environment
RL
agent
SageMaker
Training
R O L L O U T
Initial
model

View Slide

At first, it can’t even stand up

View Slide

Actions and observations
Simulation
environment
Actions
Observation
RL
agent
Initial
model
SageMaker
Training
R O L L O U T

View Slide

The model learns through actions and observations

View Slide

Interactions in the environment generate
training data
Simulation
environment
Actions
Observation
RL
agent
Training Data
Observation, action, reward
SageMaker
Training

View Slide

Training results in model updates
Simulation
environment
Actions
Observation
RL
agent
Model updates
Training data
Observation, action, reward
SageMaker
Training

View Slide

Harry learns to stand and step

View Slide

Making progress

View Slide

RL agents try to maximize rewards

View Slide

Eventually, the model learns how to walk and run
You can continue training Harry to jump obstacles, play games, dance, and more

View Slide

This makes RL applicable in many domains
and not just gaming

View Slide

Financial portfolio management
Objective Maximize value of a financial portfolio
S TAT E Current stock portfolio, price history
A C T I O N Buy, sell stocks
R E W A R D Positive when return is positive
Negative when return is negative

View Slide

Vehicle routing
Objective Fulfill customer orders
S TAT E Current location, distance from homes …
A C T I O N Accept, pick up, and deliver order
R E WA R D Positive when we deliver on time
Negative when we fail to deliver on time

View Slide

Autoscaling
Objective Adapt instance capacity to load profile
S TAT E Current load, failed jobs, active machines
A C T I O N Remove or add machines
R E W A R D Positive for successful transactions
High penalty for losing transactions

View Slide

The Amazon ML stack
A I S E R V I C E S
M L S E R V I C E S
A m a z o n
N o t e b o o k s
L e a r n i n g
S u p e r v i s e d
L e a r n i n g
L e a r n i n g
( N e o )
T r a i n i n g
H o s t i n g
D e p l o y m e n t
A m a z o n
I m a g e
A m a z o n
P o l l y
A m a z o n
T r a n s c r i b e
A m a z o n
T r a n s l a t e
A m a z o n
C o m p r e h e n d
A m a z o n
L e x
A m a z o n
V i d e o
A m a z o n
F o r e c a s t
Forecasting
A m a z o n
T e x t r a c t
A m a z o n
Recommendations
A m a z o n
E C 2 P 3
& P 3 D N
A m a z o n
E C 2 C 5
E l a s t i c
I n f e r e n c e
A m a z o n
I n f e r e n t i a

View Slide

Where do you start?
The Low Hanging Fruits

View Slide

The Amazon ML stack
A I S E R V I C E S
M L S E R V I C E S
A m a z o n
N o t e b o o k s
L e a r n i n g
S u p e r v i s e d
L e a r n i n g
L e a r n i n g
( N e o )
T r a i n i n g
H o s t i n g
D e p l o y m e n t
A m a z o n
I m a g e
A m a z o n
P o l l y
A m a z o n
T r a n s c r i b e
A m a z o n
T r a n s l a t e
A m a z o n
C o m p r e h e n d
A m a z o n
L e x
A m a z o n
V i d e o
A m a z o n
F o r e c a s t
Forecasting
A m a z o n
T e x t r a c t
A m a z o n
Recommendations
A m a z o n
E C 2 P 3
& P 3 D N
A m a z o n
E C 2 C 5
E l a s t i c
I n f e r e n c e
A m a z o n
I n f e r e n t i a

View Slide

Amazon Rekognition
Deep learning-based image & video analysis

View Slide

http://timescapes.org/trailers/

View Slide


View Slide

Amazon Polly
“What's the
difference between a
hippo and a zippo?
One is really heavy,
and the other is a
little lighter.”
Amazon Polly: Text In, Life-like Speech Out
The Text-To-Speech technology behind Amazon Polly takes advantage of
bidirectional long short-term memory (LSTM)*
52 voices across 25 languages

View Slide


The price of this book is €45

A Focus On Voice Quality & Pronunciation
Support for Speech Synthesis Markup Language (SSML) Version 1.0
https://www.w3.org/TR/speech-synthesis

View Slide

“With Amazon Polly our users benefit from
the most lifelike Text-to-Speech voices
available on the market.”
Severin Hacker
CTO, Duolingo

View Slide

Amazon Comprehend
Discover insights from text
Entities
Key Phrases
Language
Sentiment
Amazon
Comprehend
Topic Modeling

View Slide


View Slide

« Amazon Comprehend helps us analyze
the key sentiments, objects, and geos in
our 30 million plus reviews & testimonies
[…] so our customers can make the best
decision possible for their travel.”
Matt Fryer, VP and Chief Data Science Officer
Hotels.com
https://aws.amazon.com/solutions/case-studies/expedia/

View Slide

Amazon Forecast
Accurate time-series forecasting service, based on the same technology
used at Amazon.com.
No ML experience required.
NEW

View Slide

Historical data
Sales, inventory, pricing, etc.
Related data
Weather, competitive promotions, etc.
1. Load data
2. Inspect data
3. Identify features
4. Select algorithms
5. Select hyperparameters
6. Train models
7. Optimize models
8. Deploy and host models
Amazon Forecast
Customized
Forecasting
API
Private

View Slide

Sample use cases
Product demand Workforce demand Financial
metrics
Inventory
planning

View Slide

Amazon Personalize
Real-time personalization and recommendation service, based on the
same technology used at Amazon.com. No ML experience required.
NEW

View Slide

Real-time data can be consumed by Amazon
Personalize
Historical user
activity
User
attributes
Item
catalog
Real-time data
Mobile
SDKs
(coming soon)
JavaScript SDK
Amazon S3
bucket
Server-Side SDKs
Offline data

View Slide

Activity stream
Views, signups, conversion, etc.
Inventory
Videos, products, articles, etc.
Customized
Personalization
API
Demographics (optional)
Name, age, location, etc.
1. Load data
2. Inspect data
3. Identify features
4. Select algorithms
5. Select hyperparameters
6. Train models
7. Optimize models
8. Build feature store
9. Deploy and host models
10.Create real-time caches
Amazon Personalize

View Slide

More machine learning happens on AWS

View Slide

The Amazon ML stack
A I S E R V I C E S
M L S E R V I C E S
A m a z o n
N o t e b o o k s
L e a r n i n g
S u p e r v i s e d
L e a r n i n g
L e a r n i n g
( N e o )
T r a i n i n g
H o s t i n g
D e p l o y m e n t
A m a z o n
I m a g e
A m a z o n
P o l l y
A m a z o n
T r a n s c r i b e
A m a z o n
T r a n s l a t e
A m a z o n
C o m p r e h e n d
A m a z o n
L e x
A m a z o n
V i d e o
A m a z o n
F o r e c a s t
Forecasting
A m a z o n
T e x t r a c t
A m a z o n
Recommendations
A m a z o n
E C 2 P 3
& P 3 D N
A m a z o n
E C 2 C 5
E l a s t i c
I n f e r e n c e
A m a z o n
I n f e r e n t i a

View Slide

There’s Never Been
A Better Time
To Build New
Businesses
aws.com/ml

View Slide

Thank You!
Adrian Hornsby
Sr. Technical Evangelist | Amazon Web Services
@adhorn

View Slide

Machine Learning in Finance: Moving forward!

Machine Learning in Finance: Moving forward!

Adrian Hornsby

More Decks by Adrian Hornsby

Other Decks in Technology

Featured

Transcript