AI Use Case - Exec Briefing

Please note this page is being prepared and refined....However you will get the high-level usage of AI applications below

Computer Vision:

• Classical

  • best for detecting common objects

• Deep Learning

  • Convolutional Neural networks

Application Example: Tomato Sorting

Feed Computer vision with lot of different types of labelled data

Costs are minimised,

Errors are minimised

Speed is maximised

Humans can focus on more creative tasks

Applications of CV:

• Image Classification

• Image Segmentation

• Object Detection

• Object Tracking

• Image Generation

• Edge Detection

• Face Detection

• Facial Recognition

• OCR

• Pattern Detection

• Feature Matching

Computer Vision - Real Use case:

• Skanska - Use to optimise the building requirements

• Tesla - cars uses scanners to sense the surroundings

• Harvest Croo Robotics - Robots use CV to pick right strawberries

• Ebay - Image search capability

• AI cure - Computer Vision for medication , helps taking medicines at right time , right amount

• Osprey - Automatic Oil Monitoring

• Roaf - Computer Vision in waste sorting, increases recycling

• Swiss Technology - CV to identify stroke

• Cortexa - workplace safety monitoring

• Tomra - automatic Ore sorting

Deep Learning

No Predefined framework , rather have neural network that learns by experience.

Example : Distintiguish between cats and dogs, by comparing labels. But needs lot of data

Building Neural Networks:

Neural Networks may have multiple hidden layers between Input and Output layers

Finding the right neural network architecture is the key

How NN works? Every neuron's have some weights

Machine Learning encapsulates RL, CV and NLP but Deep Learning not necessarily require all three of them , which is an important aspect of Deep Learning .

Deep Learning Use cases:

Google AI - Deep Learning for Cancel Detection, 89% accuracy

Spotify - Deep Learning in recommending songs

Gold Spot Discovery - finding valuable resources , gold deposits

Digital Domain - Deep Learning for VFX

Ayasdi - Anti Money Laundering

Deep Instinct - In Cyber Security

Doxel - For productivity monitoring

Amazon Rekognition - Facial Recognisition

Estimate - estimate Real Estate

Zest Finance - Loan Approval

Reinforcement Learning

Machine learning has 3 types -

Unsupervised Learning

Supervised - provide label data upfront

Reinforcement Learning - an Imaginary agent with a problem ,

Reward System :reward + 1 for positive , reward -1 for punishment

Advantages of RL:

• RL is the future of ML, doesn't require large data sets

• RL can come up with new solutions

• BIAS resistence

• Real time learning

• RL is adaptable

RL in Marketing:

Creating personalised recommendations

Optimising advertising budget

Selecting best content for advertisment

Increasing Customer lifetime value

Predicting customer responses to price changes

Use cases with Results:

Google - AlphaGo,

Google - Energy Management

Electa - Energy management

Fanuc/Tesla - Robots

Cambridge university - Healthcare

Inventory management

Tesla/Google - Self Driving cars

Decision service (Marketing)

Trendyol (Marketing)

Alibaba (Marketing) Displaying advertisements

NLP

Types of Data:

Structured Data

Unstructured Data

Has Two Parts:

Natural Language Understanding

Natural Language Generation

Speech Recognition and generation comes when its comes to audio data

Applications of NLP

• Sentiment Analysis

• Speech Recognition

• Chatbots

• Machine Translation

• Auto completing text

• Spell Check

• Keyword Search

• Advertisement Matching

• Information Extraction

• Spam Detection

• Text Generation

• Automatic Summarization

• Question Answering

• Image captioning

• Video Captioning

ChatBots: Deep Dive

Simple ChatBot for basic questions

Advanced ChatBots - Relie heavily on NLP, NLU

Use Cases with Results:

Rotterdam Airport

Autodesk Chatbot

RPA

Robotics Process Automation

Easily programmable software for highly repetetive tasks

Key Risks:

Business Leaders are best to control RPA

Targeting the wrong process

Change Management

Machine Learning

How to make predictions based on data.

Learning + Teaching and Prediction

Innovation, Marketing and Cost

Regression : Part of ML

Independent variable /Dependant variable

Linear Regression

Polynominal regression

Multiple Regression

Classification: Part of ML

Classify data based on

Clustering :

Unsupervised data

Associated Rule Learning : Diapers and Beer example, Crackers and Dip.

Gen AI

LLM

Layers of Neural Networks,

Pre Training - Next word predictions

Fine Tuning - Trained on Specific data sets

Foundational Architecture - Transformer - ability to understand the context

Applications of GenAI/ LLM

Customer Service Automation

Content creation and curation

Language translation and localization

Automated Software development