Is deep learning (DL) for me

Hari Nair

It has come a long way from the first use of the word "robot" in English during 1923 to emotion and facial expression recognition in recent times. Now MIT “Policy Congress” examines the complex terrain of artificial intelligence for the law. It is fascinating and sometimes challenging, given the potential impact of AI. I really encourage you to be open and curious about AI; its impact and its possibilities.

What is probably missing is the clear understanding of, do we really require computers or not for a particular task; since we know that tasks done by computers are not performing as intended due to lack of understanding on how best to program it. This is a never ending story! Software works the way it is programmed. The programmer writes the codes and the software performs the planned task. Machine Learning uses complex algorithms and set of predefined rules, to read the patterns in data and then based on the analysis, generates the relevant result or performs the intended task .

Coming from ‘expert systems’ to Machine Learning (ML) is not easy and require lot more understanding of varies technologies. Let me elaborate my understanding of expert systems – The traditional way of writing code using human logic and databases, we are doing this for for a long time now. So what is so much different now?. Deep learning is all about methods based on learning data representations, as opposed to task-specific algorithms. Learning can be supervised, semi-supervised or unsupervised. Hence this technique teaches computers to do what comes naturally to humans! Deep learning is part of machine learning that is inspired by the structure and function of the brain.

Let us look at some of the limitations of rule based systems:

Lots of Data
Complex rules
Inexplicable
Rules unknown

Prerequisite to learn Machine Learning

Few of my colleagues asked me a fundamental question.

What are prerequisite to learn Machine Learning (ML).

And my answer was:

a)A laptop or a mobile phone with a larger display
b)Gmail account!
c)A good understanding of Vectors, Algebra, Calculus, Statistics and a programming language like C++ or python.

From your browser type:

https://colab.research.google.com/

and try the Code

# my 1st python program using matplotlib library

import matplotlib.pyplot as plt

import numpy as np

x = np.random.random((10, 1))

print(x)

plt.plot(x, '*-')

plt.show()

Neural networks are good at one narrow task, but they fail at handling multiple tasks. Human brain is capable of handling multiple tasks at amazing speed and accuracy. Modern neuroscience say that the information in our brain is shared and communicated across different parts.

Linear threshold gate or MP Neuron

McCulloch, a Neuroscientist, and Walter Pitts, a logician, published "A logical calculus of the ideas immanent in nervous activity" in the Bulletin of Mathematical Biophysics . They gave a highly simplified model of a neuron. The McCulloch and Pitts model of a neuron (MP Neuron) which model key features of biological neurons.

Artificial Neural Network (ANN) is an information processing paradigm inspired by the way biological nervous systems. The key element is the structure of the information processing system. It is composed of a large number of highly interconnected neuron's working to solve specific problems. ANN s learn by example. An ANN is configured for a specific application, such as pattern recognition or data classification, through a learning process.

Learning in biological systems involves adjustments to the synaptic connections that exist between the neurons. Training of deep neural networks(Deep learning) is currently highly complex and computationally intensive.

Perceptron is a single layer neural network and a multi-layer perceptron is called Neural Networks.

The perceptron consists of 4 parts .

Input values or One input layer
Weights and Bias
Net sum
Activation Function

Perceptron is usually used to classify the data into two parts. Therefore, it is also known as a Linear Binary Classifier.

Sigmoid Neuron - Logistic regression

The sigmoid function is a special case of the more general logistic function, and it essentially squashes input to be between zero and one. Its derivative has advantageous properties, which partially explains its widespread use as an activation function in neural networks.

Basic components of AI and Machine learning
Let us look at basic components

Data, Input
Task, Classification
Model, Perceptron, MP neuron
Loss, Loss functions
Learning, Algorithms, gradient descent
Evaluation, Accuracy

Artificial intelligence (AI) is the most disruptive technologies fueled by endless amounts of data, and advances in deep learning. The rise of deep learning (DL) has been fueled by three recent trends namely:

The explosion in the amount of Training data
The use of graphics processing units (GPUs)
Advancement in the Training algorithms and Neural Network Architectures.

Linear Regression using Gradient Descent Algorithm is the first step towards Machine Learning. Gradient Descent can fit a line to DATA by finding the optimal values for the intercept and the slope.

1	Predicting a continuous numeric value	Gas price in international market)?	Supervised Learning - Regression
2	Predicting a class, category, or label	Good piece or defective?	Supervised Learning - Classification
3	Creating groupings of similar data to understand the groups profiles	Customer segments)?	Unsupervised Learning - Clustering
4	Identifying highly unusual or dangerous outliers	Network security, fraud, worker without helmets	Unsupervised/Supervised Learning – Anomaly Detection
5	Personalizing your product, services, features or content for customers?	Ranking content, offers, messaging, interactions, layout	Unsupervised/Supervised Learning – Recommendations (e.g., media, products, services) Ranking/ scoring Personalization (e.g., content, offers, messaging, interactions, layout)
6	Detect, label, and identify specific spatial, temporal patterns, say in audio or video?	Detection: motion, gesture, expression, sentiment) Natural language processing	Unsupervised/Supervised Learning – Recognition (image, audio, speech, video, handwriting, text) Computer vision (e.g., detection: motion, gesture, expression, sentiment) Natural language processing (NLP)
7	AI to self-learn in order to optimize a process over time	Game playing, automation, intelligent guessing)?	Reinforcement Learning
8	Allow people and processes to find highly relevant information quickly and easily	Articles, images, videos, documents, URL, application link)?	Classical/Symbolic AI Search (text, speech, visual) Use Supervised Learning Ranking/scoring
9	Creating a personal or virtual assistant, chatbot, or language-driven agent	Text, audio, image, video, speech)?	Generative AI (including natural language generation, NLG)

Search This Blog

AI and Deep learning

Is deep learning (DL) for me