2. Index
Why this Topic?
What is Machine Learning?
Types of ML and their applications
Why now?
When and how to use? Process
Applications at Empirix
3. Why this topic
Machine Learning – subfield of AI
Deep Learning – Neural Networks
Reinforcement Learning
Blockchain – Secure, Immutable, Distributed Ledger
Gene Cell Therapy & CRISPR – Gene Editing
Quantum Computing – Super fast. Qubits
Feynman Technique
4. What is Machine Learning?
Algorithm to predict with probability by recognizing a pattern
Think of it like a baby/kid learning stuff as they grow
Process:
Predict – Hypothesis & Parameters
Run Error function – Cost Function
Learn – Minimize Cost Function
6. Unsupervised Learning
Data is not labeled
Take data and puts in a bin according to some its properties
Segments data by population
Clustering
Examples:
Google News articles
Group people according to Genes/Genome
Astronomical Data Analysis
Market Segmentation
7. Reinforcement Learning
Kind of true AI?!
Semi-supervised learning
Not fully labeled dataset
Rewards/Goal based learning
Algorithm figures out on its own the rules & best strategy to
achieve goal
Deep Reinforcement Learning – Deep Q Networks
Pretty advanced – very promising
Examples:
Self driving Car
AlphaGo, Poker, Chess, Super-Mario, Atari and other games
Natural Language Processing
Robots taking a step
8. Supervised Learning
Training data is labeled
Algorithm is trained
Linear Regression – output a number
What temperature tomorrow?
Price of house
How many units will sell
Logistic Regression
Binary Classification: spam or not, cancer or not, will customer buy or not etc.
Multiclass Classification: genre of movie, product category in Craigslist etc.
Deep Learning/Neural Networks
Emulate the way our brain works
Multiple layers of neurons: thus ‘deep’
9. Applications of Supervised Learning in Industry
Speech Recognition
Natural Language Processing – Siri & Google Assistant
Image Processing
Recognizing Faces, OCR etc.
Health diagnostics – Radiology
Chatbot
Judicial Decisions – circumvent human bias
Online Advertisement
Spam filtering
Sentiment Analysis
And many many more……
11. HPC
Matrix-multiplication operations – need lot of resources
GPU based computing - NVidia
Cloud Computing – AWS, Azure, Google Cloud etc.
GPU-based VM Instances for running training algorithms
Services provided via API
12. Big Data
Amount of data generated recently
Large companies like Facebook, Google, Amazon, Baidu
China
13. When to Use Machine Learning
You cannot code the rules: Many human tasks (such as recognizing
whether an email is spam or not spam) cannot be adequately solved
using a simple (deterministic), rule-based solution. A large number of
factors could influence the answer. When rules depend on too many
factors and many of these rules overlap or need to be tuned very finely,
it soon becomes difficult for a human to accurately code the rules. You
can use ML to effectively solve this problem.
You cannot scale: You might be able to manually recognize a few
hundred emails and decide whether they are spam or not. However,
this task becomes tedious for millions of emails. ML solutions are
effective at handling large-scale problems.
http://docs.aws.amazon.com/machine-learning/latest/dg/when-to-use-machine-learning.html
14. Data Science Process
Define the problem
Collect data for the problem
Clean data
Data Wrangling
Dividing data into training and test sets
Feature Engineering
Run ML Algorithms
Use resulting features
Generate Predictions
Visualize Results – Graphs, Charts
Present Results
15. Math
Linear Algebra
Statistics & Probability
Calculus
Languages Used: Python, R, Matlab/Octave
ML Frameworks: TensorFlow, Theano, Keras, PyTorch
Visualization: D3js, Tableau
Data Wrangling/Storage: All big data techs
Apache Spark, Redis
Apache Hadoop – HDFS, Hive, Pig, HBase
Technical Trends
16. At Empirix
Speech Recognition – instead of Nuance SR
Improve Sales/Marketing – using data in Salesforce
A support chatbot
Pattern recognition from all the customer data – New
opportunities
Tell me more…