Skip to main content

Command Palette

Search for a command to run...

AWS AI practitioner

Published
3 min read
M

I am a developer from Mumbai with 2.5 years of experience in C,Python and shell scripting. Also worked as GreenPlum database administrator for more than a year at National Stock Exchange of India. 2x aws certified in solution architect

Lect 01 : Beta exam overview


  • Aug 13th 2024 - Registration opens

  • 85 Questions , 120 min ,75$ cost

  • The exam includes these topics:

    1. Week 1

      Fundamental concepts and terminologies of AI, ML and generative AI

      Use cases of AI, ML and generative AI

    2. week 2

      Model Training and fine tuning

      Design considerations for foundation models

    3. week3

      Foundation model evaluation criteria

      Prompt engineering

    4. week4

      Responsible AI

      Security and compliance for AI systems

complete skill-builder course

Difference between AI,ML and Gen-AI

any way computer can mimic the human intelligence

AIany way computer can mimic the human intelligence
MLsubset of AI. or finding pattern among data
Deep learningsubset of ML. or neural network - works like human brain
Gen AIsubset of deep learning - used to generate stuff using LLM etc

Traditional programming : Input + Rules (code) = Output

Machine learning : Input + Rules (algorithms) = Outputs (ML model: Prediction based on input)

USE ML when...

  • when you cant code it

    eg: recognizing speech/images

  • when you cant scale it

    eg: recommendations, spam, fraud detection, machine translation

  • when you have to adapt/personalize

    eg: recommendation and personalizing. That's how amazon started as book store which can recommends

  • when you cant track it

    eg: automated driving

ML types

  1. Supervised learning

    \=> like How human learns

    \=> here we have data and a lable on it (data with lable)

    \=> Human intervention and validation is required. eg: photo classification and tagging

    \=> I have a dataset (image with label) , I used some mathematics to teach machine that for this type of image this should be the label. based on this next time machine will predict the result based on given dataset.

    SUPERVISED

    RegressionClassification

    NumericBinary

    Multiclass

  2. Unsupervised learning

    input => ML algorithm => Prediction

    eg: hey! you bought this product, similarly people who brought this also purchased this product.

    Grouping/Clustering

    Labels unknown

    Find patterns in data

  3. Reinforcement learning

    \=> Dog does not know anything , but after some training he knows what to do. and you give some sort of treat to your dog after successful execution of your command, that is reinforcement learning

    Correct action rewarded

Generative AI models are self-supervised type learning

Unlabeled training data = The quick brown fox jumped over the lazy dog.

Remove word = The quick brown fox jumped over the [ ] dog.

Model predicts word = The quick brown fox jumped over the [excited] dog.

Compare to pseudo model >> excited != lazy => update model parameter and try again.

Terminology and concepts

Statistical DefinitionEveryday definition
Label/TargetDependent variablewhat u r trying to predict
FeatureIndependent variabledata that helps u make predictions (individual column name in spreadsheets)
Feature engineeringData transformationprocess of reshaping data to get more value out of it
Feature selectionVariable/subset selectionprocess of using the most valuable data. (helps to label)

at -44:30 => note down ML process overview diagram

Business Problem

ML problem framing

Data phase Training phase Deployment phase

Amazon SageMaker => one stop solution for ML

prepare => build => train & tune => deploy & manage

Common Use Cases

  1. Financial Fraud Detection

  2. Healthcare Diagnostics

  3. Autonomous Vehicles

  4. Manufacturing Quality Control

  5. Credit risk assessment

  6. Cybersecurity threat detection

  7. Predictive maintenance

Gen-AI models excel at tasks involving language processing, open-ended processing, and generation of human-like text or other modalities

Traditional ML and AI approaches are preferred in domains where interpretability, transparency, robustness, and strict compliance with regulations are critical requirements (a -32:30 Amazon ML stack)

Amazon SageMaker

Amazon Rekognition

Amazon Textract

Amazon Comprehend

Amazon Personalize

Amazon Kendra

Amazon Forecast

Amazon Fraud-detector

AWS AI practitioner