Artificial Intelligence is a breakthrough technology that is powering all kinds of interesting applications. Some examples include Predictive Analytics, Image Processing, Natural Language Processing, Driverless Cars, Gaming, Financial Predictions, Supply Chain Analytics, Robotics, Weather forecasting, Ecommerce, Recommender Systems etc.
AI is based on training models with data where input and output has been provided. E.g. training a house price prediction model based on historical house price datasets, where the input factors are mapped to house prices. Or image classification, where images are annotated with object classes (e.g. Image 1 contains a cat, vs Image 2 contains a dog).
Machine Learning
Machine Learning is a branch of AI, where we use data science algorithms to train AI models based on datasets provided. The model essentially learns, or discovers, the relation between input x features and output y.
y = f(X)
where X represents a set of input or independent variables and y is the dependent variable. In case of house price prediction, input could be area of house, age of house, no of bedrooms etc and y is simply the house price.
Given many such X-Y pairs from historical data, we can train an AI model (the equation itself is the AI model) which learns to predict Ynew for any new input Xnew.
Machine Learning can be of two major types Supervised and Unsupervised. Supervised machine learning means Y label is provided for X input. Unsupervised machine learning means that Y is NOT provided for X input.
The major kind of Supervised machine learning are :-
- Regression (Linear Regression, Polynomial Regression etc.)
- Classification (Logistic Regression, KNN, Decision Trees etc.)
Regression is about predicting contiguous value Y label (e.g. house prices, how much bank loan to provide, how much it will rain etc.). Classification is about discrete class label prediction (e.g. issue bank loan Yes or No, will it rain Yes or No, what object is contained in the image etc.)
When one says one is training a supervised ML model, what one is really doing is learning the ideal value of weights for various input variables.
y = w1 * x1 + w2 * x2
Here we are learning weights w1 and w2 during the ML training process. Let’s say the ideal weights learnt are w1 = 5 and w2 = 8.2, so our “learnt” or “training” AI model then becomes :-
y = 5 * x1 + 8.2 * x2
This equation can now be used to predict Ynew for any new X sample (x1new, x2new).
The popular kinds of Unsupervised machine learning are :-
- Clustering
- Anomaly Detection
Clustering is able to identify groups among samples, such as grouping hospital patients into high-risk vs low-risk clusters, without having the Y label available in training data.
Deep Learning
Deep Learning is a specialized type of Machine Learning, and is used for more complex information processing applications, such as detecting objects in images or understanding text or speech.
Deep Learning uses Artificial Neural Networks (ANNs), or networks of neurons which work in a “divide-and-conquer” approach to gradually discover abstract features in complex datasets in a forward propagation scenario.
Each neuron is a computation of the kind “y = f(X)”, same as what we saw in classic ML earlier.
By combining subsequent layers of neurons, a Neural Network progressively discovers more higher level features. The final layer output is then used to predict either class label (e.g. dog or cat in image) or regressive output (e.g. how much bank loan to provide) for some problem statement.
ANNs are of various kinds :
- Dense ANN or regular ANN — fully connected neural network (as shown in the image above)
- Convolutional Neural Networks or CNN — sparsely connected neural network. Useful for data which is sensitive to neighbourhood intelligence (e.g. in an image, nearby pixels work together to constitute features such as eye, nose, head, full face in a facial image)
- Recurrent Neural Networks or RNN — a neural network which captures sequential intelligence. Useful for predictions on sequential data such as text sentences, speech signal, time-series of stock market prices over several days etc.
CellStrat AI ML Crash Course Bootcamp (starts tomorrow 22nd Feb ‘22)
CellStrat AI Lab is pleased to present an AI ML Crash Course Bootcamp starting tomorrow (22–27 Feb) via online hands-on workshops. The course curriculum includes
- Basics of Python
- Machine Learning
- Deep Learning
For further details, check the CellStrat Meetup page !
Bootcamp Fee — Rs 200
Topic — CellStrat AI-ML-DL Bootcamp
Dates — 22–27 Feb, 7–9 PM IST / 8:30–10:30 AM US EST
Presenter — CellStrat AI Mentors
REGISTER HERE
CellStrat AI ML Bootcamp is conducted on CellStrat Hub — the world’s most popular AI learning platform.
BONUS — Top 10% performers in bootcamp to be offered free 3 mths. product internship with CellStrat Inc.
Join the AI ML Bootcamp ! Accelerate your career with cutting-edge AI skills !
USEFUL LINKS :-
- CellStrat Hub
- CellStrat Hub AI Project Packs
- CellStrat Hub AI Apps and API Marketplace
- CellStrat YouTube Channel
- CellStrat product demo : Vision Apps and NLP Apps
- CellStrat AI Lab community (Disrupt4.0)
- CellStrat research : https://bit.ly/CS-research
