2. 2
• 1996 - Today HAVELSAN, Inc.
• 1996 - Development of HVL Firewall (The 1st in Turkey)
• 2001 - Developer in various projects: TuAF IS, MELTEM, etc.
• 2010 - YGO Project/Product manager
• 2014 - Move to HVL Istanbul Office
• 2014 - Systems Engineer
• 2016 - R&D Coordinator
• June 2016 - Organizer at www.dataistanbul.org
BSc, Bilkent University Computer Engineering, 1996
MSc, Bilkent University Computer Engineering, 1999
PhD, Gebze Technical University Computer Engineering
(in progress)
3. • Meetup Community
• Established: March 2016
• Members: 5000+
• Event: 90+
• ML/DL Course
• Introduction to Kaggle Challenges
• Reinforcement Learning Day
• NLP with R Workshop
• …
/data_istanbul/dataistanbul
4. 4
• Introduction to Artificial Intelligence
• Introduction to Data Science
• Limitations of Artificial Intelligence
• Technical Limitations
• Practical Limitations
• Application Limitations
Agenda
7. 7
• A branch of computer science that aims to create
intelligent machines.
• Research Areas:
• Knowledge (understanding)
• Reasoning
• Problem solving
• Perception
• Learning
• Ability to manipulate and move objects
• …
Artificial Intelligence (AI)
8. 8
• Artificial Narrow Intelligence
• Weak AI that focuses on a single task.
• Common “AI” in general.
• Artificial General Intelligence
• Intelligence of a machine that could successfully perform any
intellectual task that a human being can.
• Strong Artificial Intelligence
• A machine with consciousness, sentience and mind.
Artificial Intelligence – continued
9. 9
• A research area of artificial intelligence (AI) in which
*machines* learn and improve from experience
automatically without being explicitly programmed.
• Access and use data to learn/develop computer programs.
Machine Learning (ML)
• Tasks/Problems:
• Regression
• Classification
• Clustering
• Ranking
• Feature Engineering
• Feature Learning
• Anomaly Detection
• …
• Methods
• Linear Regression
• Logistic Regression
• Decision Trees
• Artificial Neural Networks
• Deep Learning
• …
10. 10
• Particular subset of ML methodologies using artificial
neural networks, inspired by the structure of neurons
located in the human brain.
• Deep usually refers to presence of multiple layers of artificial
neural networks, as in the human brain.
Deep Learning (DL)
• Methods
• Convolutional Neural Network
• Recurrent Neural Network
• Generative Adversarial Network
• …
13. 13
• Supervised Learning
• When your data has one or more target variables (expected
results), and you make “predictions” on the data.
Machine Learning Tasks
14. 14
• Unsupervised Learning
• When your data has no target variables, and you detect
“patterns” on the data.
Machine Learning Tasks
15. 15
• Transfer Learning
• When you have data that can be used with an existing model, you
re-train, and predict.
Machine Learning Tasks – continued.
16. 16
• Reinforcement Learning
• When you have no data but you can generate data via trial-and-
error, and you “play.”
Machine Learning Tasks – continued.
18. 18
• Before 1600, empirical science
• Direct observations
• 1600-1950s, theoretical science
• Each discipline has grown a theoretical component. Theoretical models often motivate
experiments and generalize our understanding.
• 1950s-1990s, computational science
• Over the last 50 years, most disciplines have grown a third, computational branch (e.g.
empirical, theoretical, and computational ecology, or physics, or linguistics.)
• Computational Science traditionally meant simulation. It grew out of our inability to find
closed-form solutions for complex mathematical models.
• 1990-now, data science
• The flood of data from new scientific instruments and simulations
• The ability to economically store and manage petabytes of data online
• The Internet and computing Grid that makes all these archives universally accessible
Evolution of Sciences
Jim Gray and Alex Szalay, The World Wide Telescope: An Archetype for Online Science, Comm. ACM, 45(11): 50-
54, Nov. 2002
19. 19
• …the exploration and quantitative analysis of all
available structured and unstructured data to develop
understanding, extract knowledge, and formulate
actionable results.
Data Science is…
Microsoft Data Science and Machine Learning Essentials
22. 22
• When data is available, produce much accurate results than
human.
• Works in hostile environments
• Replace human in repetitive, tedious tasks
• Assistant to human
• Fraud Detection
• Telesurgery
• Work for long hours, no sleep, rest, take break, or get
entertained.
• …
Advantages of Artificial Intelligence
24. 24
• Technical Limitations
• Interpretability
• Explainability
• Practical Limitations
• Data Bias
• Data Availability
• Missing Data
• Application Limitations
• Other Disadvantages
Limitations of AI
25. 25
• Interpretability
• Understanding the model in terms of what it has learnt, and what
it outputs.
• Generally defined as the features contributed to the decision
process for a particular instance.
• Explainability
• Why and how a particular decision is made.
• How trustworthy is the decision.
Technical Limitations
28. 28
• Human Bias:
• Act with ones own belief without considering relevant data.
• Overcome with relevant data and working procedures.
Practical Limitations
29. 29
Example: Survivorship Bias
• Logical error of concentrating on the people or things that
made it past some selection process and overlooking those
that did not, typically because of their lack of visibility.
https://www.wikiwand.com/en/Survivorship_bias
30. 30
• Dataset Construction Bias
• Data Collection Bias
• Data Labeling Bias
• Data Sampling Bias
Practical Limitations – Data Bias
31. 31
• Data Collection
• Dataset Construction Bias
• Data Collection Bias
• Data Labeling Bias
• Data Sampling Bias
• Data Cleaning
• Human Bias
• Data Labeling Bias
• Data Sampling Bias
• Exploratory Data Analysis
• Data Sampling Bias
• Data Modeling
• Data Sampling Bias
Practical Limitations – Data Bias
34. 34
• For some research areas, almost always researchers use
same datasets available on the Internet.
Practical Limitations – Data Availability
35. 35
Example: Coded Gaze by Joy Buolamwini
https://www.media.mit.edu/posts/how-i-m-fighting-bias-in-algorithms/
36. 36
• For other areas, researchers have to dooo something.
• Synthetic Data
• Data Augmentation
• Transfer Learning
• Reinforcement Learning
Practical Limitations – Data Availability
37. 37
Solution – Synthetic Data
https://medium.com/archieai/a-dozen-times-artificial-intelligence-startled-the-world-eae5005153db
38. 38
• For images:
• Scaling
• Translation
• Rotation
• Flipping
• Adding Salt and Pepper noise
• Lighting condition
Solution – Data Augmentation
https://medium.com/ymedialabs-innovation/data-augmentation-techniques-in-cnn-using-tensorflow-371ae43d5be9
39. 39
Solution – Transfer Learning
https://towardsdatascience.com/transfer-learning-using-differential-learning-rates-638455797f00
42. 42
• Some data maybe missing in the dataset.
• Eg: Broken sensors, or communication lines.
• Eg: Data not applicable for that sample.
• …
• Solution depends on the problem:
• Drop missing value samples, or features.
• Data imputation: Replace missing values with mean/median etc.
• Data imputation: Develop linear regression model for features
with missing values.
• …
Practical Limitations – Missing Values
43. 43
• Missing AI Ethics
• Missing AI Strategy
• Executing a successful ML project is costly.
• Narrow AI cannot solve every problem
• AI is not as creative as humans
• AI does not have emotions
• AI does not have common Sense
• People become too dependent to AI
• AI leads to unemployment
• AI leads to the “Destruction of Human Race?”
•
Application Limitations – Disadvantages
44. 44
• When data is available, produce much accurate results than
human.
• Works in hostile environments
• Replace human in repetitive, tedious tasks
• Assistant to human
• Fraud Detection
• Telesurgery
• Work for long hours, no sleep, rest, take break, or get
entertained.
• …
Advantages of Artificial Intelligence