Machine Learning History & Prbabilistic modelling

The History of Machine Learning & Probabilistic
Modeling
Dr.Venkatasupura Vemulapati
Head ICT & Estudio
Dept. Of CSE(Data Science)
SITE-Tadepalligudem
July 7, 2025
Dr.Venkatasupura VemulapatiHead ICT & EstudioDept. Of CSE(Data Science)SITE-Tadepalligudem
The History of Machine Learning & Probabilistic ModelingJuly 7, 2025 1 / 13

Introduction
Machine Learning is a type of AI that allows software applications to
become more accurate in predicting outcomes without being explicitly
programmed.
It uses historical data as input to predict new output values.
Machine learning algorithms learn from data and improve their
performance over time, unlike traditional programming where tasks are
explicitly defined.

Historical Overview
The history of machine learning is deeply connected with:
Artificial Intelligence (AI)
Early Neural Networks
The AI Winter
Deep Learning Revolution
Modern Machine Learning Landscape

Early Foundations and AI Roots (1950s–1980s)
In the 1950s, computer scientists explored whether machines could be
made to ”think.”
Symbolic AI: Creating intelligence using explicit rules.
Popular in the 1980s through expert systems.
Early chess programs were rule-based, not ML.
Ideas came from logic, probability, economics, and neuroscience.

Early Neural Networks (1950s–1960s)
Inspired by biological neurons.
1957: Frank Rosenblatt introduced the Perceptron.It was the earliest
formulation of an artificial neuron. It was a simple model capable of
classifying binary data, consisting of a single-layer neural network with
a threshold gate at its output
Other models: ADALINE, MADALINE (1959).
Could learn only simple linear patterns.
Computationally expensive and not noise tolerant.

AI Winter and Rise of Traditional ML (1980s–1990s)
Neural networks lost popularity.
Funding and research shifted to traditional ML.
Focused on feature engineering: manually extracting features from
raw data.
Techniques used: Regression, support vector machines (SVMs),
Random Forests.

Return of Neural Networks (1990s–2000s)
As neural networks began to regain respect in the 1990s, partly due
to the development of the SVMs introduced by Vapnik and Cortes.
Kernel methods helped ML handle nonlinear data. SVMs aim to solve
classification problems by finding optimal decision boundaries.
1998: LeNet-5 by LeCun and Bengio,a convolutional neural
networkdemonstrated efficient handwritten character recognition due
to a hierarchical architecture inspired by biological vision, superior
training data, faster processing power, and the backpropagation
algorithm. This marked an early commercial application of deep
learning, used by the United States Postal Service to automate ZIP
code reading.
Used convolutional layers and backpropagation.

Rise of Deep Learning (2010s)
The ”deep” in deep learning refers to the number of successive layers
in the artificial neural networks used to learn from data.
2012: The success of deep learning models like AlexNet is largely
attributed to their ability to automatically extract pertinent features
from raw data, reducing the need for manual feature engineering that
was common in traditional machine learning.AlexNet won ImageNet
competition.
Reduced need for manual feature engineering.
Improved image classification, speech, handwriting, translation.
2015: Machines surpassed human accuracy in image classification.

AI, ML, and DL: The Hierarchy
Artificial Intelligence (AI) – The broadest term.
Machine Learning (ML) – A subset of AI.
Deep Learning (DL) – A subset of ML, based on deep neural
networks.

Modern Machine Learning Landscape
Four Main Branches of ML:
Supervised Learning – Labeled data, classification, regression.
Unsupervised Learning – Clustering, dimensionality reduction.
Reinforcement Learning – Sequential decision making with reward
signals.
Deep Learning – Applied to all above types, uses neural networks.

Summary
ML has evolved from symbolic AI to neural networks and deep
learning.
Deep learning has transformed modern AI.
While DL is powerful, traditional ML remains widely used.
ML continues to expand into various domains with growing
capabilities.

Probabilistic Modelling
Probabilistic modeling is a method that employs statistics to comprehend
data. This approach has a long history and remains widely utilized today.
It is also identified as a component within the history of machine learning.
One of the most widely recognized probabilistic models is Naive
Bayes.
Naive Bayes is described as a simple yet potent algorithm, including
spam filtering, text classification, and medical diagnosis.
Its operational principle relies on the assumption that each feature
within a dataset is independent of others. This implies that the
likelihood of a specific outcome is influenced solely by the probability
of that outcome for the individual feature.

Cont...
The algorithm utilizes Bayes’ theorem for data classification. Bayes’
theorem computes the probability of an event based on prior
knowledge. In the context of Naive Bayes, this prior knowledge is
derived from the distribution of features in the training data.
Naive Bayes can be highly effective for classification tasks and is
valued for its speed and scalability, making it suitable for large
datasets.
Logistic regression is a closely related model, often referred to as the
”hello world” of modern machine learning.
More broadly, probabilistic modeling is connected to the fundamental
ideas of quantifying uncertainty and reasoning with probability within
artificial intelligence. It also plays a role in economics, particularly in
decision theory, which combines probability theory with utility theory
to handle uncertain outcomes.

Machine Learning History & Prbabilistic modelling

More Related Content

Similar to Machine Learning History & Prbabilistic modelling (20)

Recently uploaded (20)

Machine Learning History & Prbabilistic modelling