Artificial, Machine and Deep Learning ( An Introduction )
AI/ML/DL is as good as the training data collection and model selection. Its uses range from algo/HFT trading, predicting the stock/asset price, volatility in the markets, FX Exchange rates fluctuation, Option pricing to credit events like probability of default of consumer loans or events like bankruptcy of large corporates and ratings changes.
The amount of data generated today is humungous, from each click of the mouse we make, to each site we visit, or each tiktok/Youtube video we watch, tech companies collect a vast amount of data on our behalf.
Machine Learning is a broad term and is categorized as per the techniques used as below:
- Supervised Learning
- Unsupervised Learning
- Reinforcement Learning
Supervised Learning:
As the term suggests, in this type of technique the algorithmic model is trained with lots of data (both input and output) which is labelled. E.g. if you feed an labelled image of a dog and a cat in the algo as an input ( training phase) and then feed new images of cats and dogs, the algo will be able to identify or label the images as cat or a dog.
Uses of Supervised Learning in real world are:
Target Marketing, Predictive Analysis, Face/Image Recognition, Predicting probability of default etc.
Different types of algos used in supervised learning are as below.
- Random Forest
- Decision Trees
- Logistic or Linear Regression
- Gaussian Naive Bayes
- K- Nearest Neighbor (KNN)
- Support Vector Machine (SVM)
Unsupervised Learning:
Contrary to supervised learning, the data fed in the algo is not labelled, instead the algo needs to identify the patterns or trends hidden in the data and then provide the output.
Uses of Supervised Learning in real world are: Suggestions of News feeds on social media platforms or People you may know or Recommendation Engines, Anamoly/Fraud detection used in AML/CFT Compliance etc
Different types of algos used in supervised learning are as below.
- Hierarchical clustering
- Probabilistic Clustering
- K-means
- Principal Component Analysis
- Anomaly Detection
Reinforcement Learning:
This is the most difficult form of Machine learning technique wherein machine/algo needs to learn on its own i.e. its not trained with any sample data like in Supervised learning. Here the algo works in a trial and error format and receives rewards in case of a successful decision and punishment in case of a wrong decision.
Uses of Supervised Learning in real world are: Self Driving cars, Industry automation, Gaming etc
Different types of algos used in reinforcement learning are as below:
- Deep Q Network (DQN)
- State-Action-Reward-State-Action (SARSA)
- Q learning
