ML is an application of AI that provides system the ability to automatically
learn and improve without being explicitly programmed. And AI aims at making machine smarter by giving those capabilities to think on their own feet to make decisions or mimic human activities and one can think of ML as a subset of AI.
It focuses on the development of computer programs that can access data and use it to learn themselves. The learning process begins with observations of data, experience or instructions, this process helps to look for a pattern in data and make better decisions in the future.
APPLICATIONS : 1. Personal Digital Assistants that can be further integrated with a variety of platforms.
2. Training a computer to work on the process of video surveillance, tracking the inappropriate behavior of people like, motionless for a long time, stumbling, e.t.c
3. Analyzing and rectifying E-mail Spam.
There are many more applications, will cover it up next time.
Techniques of ML :
SUPERVISED LEARNING: Training dataset contains input data and the value we want to predict.
The model uses the training data to establish a link between the input and the output. Training data can be generalized and that the model can be used on new data with some accuracy.
Algorithms under supervised learning are Naive Bayes, Gradient Boosting, Neural Networks.
It is often used for image recognition, speech recognition and sometimes in financial analysis.
UNSUPERVISED LEARNING: It does not use output data, and can be split into different categories.
- Algorithms that can be used to reduce the dimensions such as PCA, LCA, Autoencoder.
- To detect the observations that do not follow the data set patterns.
- Clustering algorithms like, K-means, mixture models. They try to separate the observations in different groups.
This learning is mostly used to pre-process the data, or pre-train supervised learning algorithms.
REINFORCEMENT LEARNING: These algorithms can be seen as the best possible way for earning the greatest reward. Rewards can be anything from earning money, winning a game or beating opponents.
This learning method follows various steps like the model (agent) will choose the action to maximize the reward based on the state of the environment. These actions will change the state of the model and environment. They may be interpreted to reward the model. By performing this loop, the behavior of the model will be improved and the accuracy of our model will surely show some improvement.
It performs well on a small dynamic system and is definitely to follow for the years to come.