Machine Learning

24 December 2020    

Introduction:

Machine Learning: It part of computer science that gives computer to learn without being explicitly programmed. Process is same as how human learn and then use his knowledge in day to day work and inventing new things. Process allow machine to learn form historical data and predict something.

ML Application:

·        E-commerce or entertainment site for predicting / recommending customers. E.g Netflix and Amazon can predict what their customer need and also recommend their customers.

·        Build good decision-making tree or algorithm / application.

·        Healthcare industry: to predict future health condition or develop medicine or treatment for particular health issue.

 

Statistical modeling used for ML

·     Popular arithmetical algorithms: Regression, Classification, and Clustering
·     Recommender Systems: Content-Based and Collaborative Filtering
·      Popular models: Train/Test Split, Gradient Descent, and Mean Squared Error

 

ML techniques:

·      Regression / estimation technique: used to predict continuous value.
·         Classification technique – used for predicting class or category of a cases.
·         Clustering – grouping of similar cases.
·         Association technique -used to find events or items that often co-occur.
·         Anomaly detection – discover abnormal or unusual case or behavior.
·         Sequence mining – used to predict next event.
·         Dimension reduction – used to reduce size of data
·         Recommendation system -recommend customer for particular product or item.
 

 

 

Difference between AI / ML / DL

• AI(Artificial Intelligence)- make computer intelligent to perform function normally done by human. AI includes Computer Vision, Language Processing, Creative and summarization.
• ML(Machine Learning)- is branch of AI that works with statistical part of AI. It helps computer to solve problem.
• DL(Deep Learning)- this is enhancement of ML where computer learn and make decision.
• DS: Data Science

 

Objective of Machine Learning (ML):

In machine learning we create models, that can be used for predication or processing.

 

IBM cognitiveclass.ai/ courses recommends:

·         Python packages, Numpy, SciPy, Matplotlib, Pandas Library and Scikit-learn

 

 

Open Source AI using Scikit-learn: (Source- IBM Cognitive Class)

·         It is a free ML library for the Python programming language.

·         It has built in algorithm or ML techniques.

·         Has good documentation and used with few line of python code.

·         Most task required for ML are implemented already in Sci-kit learn. So easy to implement.

 

 

How to use Scikit-learn:

·         It uses standardized data set, we can take any non-standard data and fix them.

·         You have to split your dataset into train and test sets to train your model. And then test the model’s accuracy separately

·         Normal recommendation

o   70 – 80 % data used as Train Set

o   10 – 20 % validation Set

o   10 – 20 % Test set

·         We setup algorithm and train your model with the train set. Outcome is to get unknow value.

·         We can use different metrics to evaluate your model accuracy.

·         Finally save your model.

 

The Microsoft Cloud-powered AI Platform:

·         Azure AI Services:           

o   Cognitive Services (Pre-built AI)

o   Azure ML services (Custom AI)

o   Bot Services (Conversation AI)

·         Azure Infrastructure:

o   Database, CPU, GPU etc.

·         Tools:

o   Coding & Management Tools:

§  Azure ML for VS code

§  Drag & Drop Designer

§  Automated ML

o   Deep Learning Frameworks

§  Cognitive toolkit

§  Tensor Flow

§  Caffe

Microsoft Pre-Built AI Services.

·         Vision : API for gesture

o   Computer Vision

o   Ink Recognizer

o   Face

o   Video Indexer

o   Custom Vision

o   Form Recognizer

·         Speech

o   Speech Translation

o   Speaker Recognition

o   Speech to Text

o   Text to Speech

·         Language

o   Immersive Reader

o   Translator text

o   Language understanding

o   Text Analytics

·         Search

·         Decision

 

We can download Intelligent Kiosk from Microsoft Store

Azure portal: portal.azure.com/#home

 

 

Type of learning model:

•  Supervised:
• Unsupervised
• Semi-Supervised

Supervised: 

we teach mode or load model with knowledge so that we can have it predict future instance. We teach model by using some data from labeled dataset.

 

Sample: Labeled Data

 

 

Type of supervised learning:

• Classification: it is process of predicting a discrete class label or category

                Decision Trees:

                    Two-Class Classification. Example True / False, Yes / No

                    Multi-Class Classification

• Regression: is the process of predicting a continuous value.

 

Regression Wiki:

In statistical modeling, regression analysis is a set of statistical processes for estimating the relationships between a dependent variable (often called the 'outcome variable') and one or more independent variables (often called 'predictors', 'covariates', or 'features').

 

The most common form of regression analysis is linear regression, in which a researcher finds the line (or a more complex linear combination) that most closely fits the data according to a specific mathematical criterion.

 

Unsupervised Learning:

We don’t provide training to model but allow model to learn or discover of its own. It can work on unlabeled data.  It is using more complicated algorithm then supervised learning.

 

Technique used by unsupervised learning.

·         Dimension reduction
·         Density estimation
·         Market basket analysis
·         Clustering

                     K Means

 

Challenges and Risks Bias can affect result(Source Microsoft AI classroom course)

·         Errors may cause harms
·         Data could be exposed
·         Solution may not work for everyone
·         User must trust a complex system
·         Who’s liable for AI Driven Decisions

Principles of Responsible AI (Source Microsoft AI classroom course)

·         Fairness
·         Reliability and safety
·         Privacy and security
·         Inclusiveness
·         Transparency
·         Accountability