Professional Documents
Culture Documents
net/publication/351412945
CITATIONS READS
0 71
1 author:
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Israël Tankam Chedjou on 08 May 2021.
Use cases of AI
02
03 AI Careers
What is (not) AI ?
Artificial Intelligence (AI)
What is Artificial Intelligence (AI)?
Statistics
Cognitive Computer
sciences sciences
Machine
Learning
Supervised learning
The process
Input: Output:
Data Artificial Action
Sensors Intelligence Movement
Images… Text…
Machine Learning
Algorithm
Labelled Data
Training
Prediction
ACP
Performance
or
New Feature Just Salary
Unsupervised learning
Reinforcement learning
1. The input and output
system
2. The reward
3. The AI environment
gradient descent
In practice
1- Collect and prepare the data
2- Built your first model quick and train it on
the data
3- Create a single measure of accuracy
4- Test the accuracy of your model on
ground truth data
5- Modify the model to have better accuracy
• Solve variance or bias problems
• Have more data for training
• Train longer
• Regularize
• Dropout
• Deeper or shallower models
• Etc.
6- Train the new model
Summary: the machine learning pipeline
Data
Data collection Data
visualization
and integration preparation
and analysis
Model Feature
evaluation Model training selection and
(error analysis) engineering
Prediction
The process (in details)
Re-training
ML problem framing Data collection
Predictions
Model evaluation
Data visualization &
analysis
Model deployment
augmentation
Feature
• Environmental
Robot displacement recognition
• Reinforcement learning • Natural language
processing
• Facial expression
processing
Bank and finance
10 dimensional input
Composite Index
vector of volatility
Volatility Forecast
forecast for 10 indices
Consumer
Corporate
Wholesale
Credit risk
Credit card
Principal risk Concentration
Counterparty
Financial risk Collateral
Interest rate
Market risk Foreign exchange
Equity
Commodity
Liquidity risk Credit spread
Risk
• Profiling
• Sentiment analysis
• Context analysis
• Decision making
• Tactics
⚠ Social biases
Health Care
Google DeepBlind
• identify the early signs of eye disease
• Audience targeting
• Speech analysis
• Individualized AI Tutors
• Personalized Games
Agricultural drones
• Monitor the growth and
production of crops
• Id. weeds and damaged
plants.
• Potential of a given terrain
• Autonomous tractors
• Music generator
• Art generator
• Faces generator
• Architecture
• Object/animals/scenes generator
GAN and NLP
Image Generation – From Descriptions
GAN and NLP
Image Generation – Handwriting
Computer vision and NLP
LipNet - Sentence-level Lipreading
NLP for audio processing
Sound generation
Natural Language Processing
Syntax Parsing
Natural Language Processing
Generating text
Natural Language Processing
Translation
Natural Language Processing
Chatbot
Email Spam Filter
Example • System detects patterns in Probability of spam
Email about likely markers
“Earn cash” of spam Contains Contains
• Can use such detected “Free” “Earn cash”
patterns to make 70% spam 90% spam
automated decisions
about
future emails
1. Algorithm improves in
“Earn cash” performance in auto- From
appears in 10% identifying spam Nigeria
Of spam emails, 2. Algorithm is “learning” 85% spam
0% of non-spam over time from additional
It’s a (generalizable) pattern examples
Recommender systems
Content-based filtering Collaborative filtering
A content A content
genre genre
Similar users
Similar content
graphs
Query
database
Others
@AiAfrican
Chat ? @israeltankam
View publication stats