This is the slide deck I used when speaking at Praxis school of Business on the topic "Neural Networks -it’s usage in Corporate" as part of their Distinguished Speaker Series.
2. WHY DOES
CORPORATE NEED
AI
• Draw software requirements to solve a problem
after analyzing data – Product Managers, SMEs,
Customer interaction teams etc.
• Data cannot be too complex.
• Limited human capability
• Data complexity explosion
• Use historical data and results to identify the
requirements – Artificial Intelligence
3. MACHINE LEARNING VS DEEP LEARNING
Artificial Intelligence
Machine Learning
Deep Learning
• Techniques to enable a
computer to mimic human
intelligence
• Using Algorithms to learn from
data without explicit coding
• Emulate the learning approach
of human beings to gain
certain types of knowledge
4. HOW IT WORKS
• Imagine a tabular data
• Can we predict the value of y
using x1,x2,x3,x4,…,xn?
• Y = f(x1,x2,x3,x4,…,xn)
• Function f is computed using
statistics - typically, Machine
Learning
• Function f is computed using
Linear Algebra – Deep Learning
Insurance
Premium
(y)
Age
(x1)
Education
Level (x2)
Mileage
(x3)
Vehicle
Brand
(x4)
5. PERCEPTRON
Fundamental building block of
learning
Contains weights, bias and an
activation function to bring in
linearity
Perceptron can emulate a NAND
Gate.
Consequently, is capable of universal
approximation
6. DEEP NEURAL NETWORKS
• A stacking of tens to millions of neurons in various combinations
• Second generation of neural networks
• Used extensively in all areas of AI
• Often misunderstood to be the same as artificial intelligence.
• Data hungry
• Resource hungry
• Black box.
7. NEURAL
NETWORKS IN
TODAY’S AI
• Hard to think of an area where they are not being
used.
• Almost all domains of Artificial Intelligence use
neural networks
• Dense networks, Sparse networks, Convolutional
Networks, Recurring Networks, Long Short Term
Memories, Encoders and Decoders and a lot more.
8. EXAMPLES
Convolutional Neural Networks
• Resnet, Inception, SSD
• Used for image processing
• Automatic vehicle damage
detection, Manufacturing
quality inspection, Diabetic
Retinopathy
Recurrent Neural Networks
• Apple’s Siri, Google voice
search
• Useful where memory is to be
maintained
• Stock price predictions
(algorithmic trading), Machine
Translation, Voice
transcriptions
9. EXAMPLES
Long Short Term Memories
• Special case of RNNs
• Video classification,
Classifying ECG signals, Text
generation
Encoders and Decoders
• A major component of
Generative AI
• GPT, Dall-E, Google’s Gemini,
Llama, Mistral and a lot
more.
• A lot of functionalities in
Adobe Photoshop, MS Office,