AWS Application Services

Question 1

Amazon Rekognition

Answer 1

Rekognition Image lets you easily build powerful applications to search, verify, and organize millions of images. Rekognition Video lets you:
1) extract motion-based context from stored or live stream videos, and helps you analyze them.
2) index metadata such as objects, activities, scene, celebrities, and faces, making video searches easy. 3) uses deep neural network models to detect and label thousands of objects and scenes in your images.
4) It helps you capture text in an image, a bit like Optical Character Recognition (OCR). A perfect example is a T-shirt with quotes on it. If you were to take a picture of one and ask Amazon Rekognition to extract the text from it, it would be able to tell you what the text says

Question 2

Common uses of Rekognition

Answer 2

Image labeling
Custom Image labeling
Face Detection and Search
People Paths
Text Detection
Celebrity Detection
PPE

Question 3

Rekognition Image and Video Operations

Answer 3

Image operations are synchronous and video ops are asynchronous. Image ops are instantaneous. WHile video ops require for the video to be processed first.

Question 4

How to limit responses for Amazon Rekonition for images

Answer 4

SPecify MaxLabels

Question 5

Key benefit of Rekognition

Answer 5

Ability to work with streaming videos. It can ingest videos from Kenisis streams, process videos, and publish the outputs to Amazon Kinisis Data STreams for stream processing

Question 6

Amazon Textract

Answer 6

Allows you to extract intelligence from documents such as financial reports, medical records, tax forms, and university app forms.

More importantly, it allows you to quickly build automated document processing workflows

Question 7

Common use cases for textract

Answer 7

Creating a search index by storing the outputs of textract document analysis in key value store like dynamodb

Mining text from documents for NLP. It can extract words, lines, and tables

Automating data capture from forms. Textract can extract info fro structured documents such as tax forms or app forms

COst effective: You pay for what you use

Question 8

APIs for Textract

Answer 8

Synchronous - you have option of passing a document to Textract for processing either as a byte array or S3 object. DetectDocumentText or AnalyzeDocument can be used to return JSON of detected or analyzed text.

Detect API - detects text
ANalyzeAPI - recognizes heirarchy in document such as form data, tbles, lines etc.

Question 9

Amazon transcribe features

Answer 9

STream and batch mode
Multiple language support
Multiple language transcription
job queuing
cusotm vocab and filtering
auomatic content redaction: can remove pii
language identification
speaker identification

Question 10

Amazon custom model for transcribe

Answer 10

can build custom data model by providing text as input

Question 11

TRanscribe medical

Answer 11

Transcribe for medical:

asr service that allows you to transcribe medical dictation, patient to physician, conversations, and teleedicine, available in streaing and batch mode(only for primary care).

ALlows you to build custom vocabularies and redact phi

Question 12

Amazon translate

Answer 12

text translation service that provides high quality translations(from different languages) without deep learning experience. Pay as you go.

Question 13

Amazon translate features

Answer 13

Sync and async apis: To asyn process large numbers of documents using a batch job(in 5gb batches). This api is helpful when individual documents in the collection are small, such as social edia postings or user reviews.

For smaller documents, you can run translation in real time.

Custom terinology and parallel data: You can provide custom terminology in a csv file from custom terms in source language and target terminology

Question 14

Amazon polly

Answer 14

Text to speech. User provides text as plain text or in Speech SYnthesis Markup Language(SSML). Polly reads this and generates lifelike syntax and generates lifelike voice using prebuilt voices in different langauges

Question 15

Amazon Lex

Answer 15

allows you to deploy conversational interfaces for their applications

Question 16

COre steps in creating an amazon lex app

Answer 16

1. Create a bot in desired language
2. Text bot
3. Publish the bot as a version so you can roll back to a prior version
4. Deploy the bot to an end to end app

Question 17

Amazon backend actions

Answer 17

Can be performed by lambda functions

Question 18

How kendra works

Answer 18

Index: Beofre you can search for documents, you need to index them
Documents: The items that will be indexed
Data source: You provide kendra with the data source and it automatically indexes the documents

Once your index is ready, you can write queries to the index to return most relevant items.

Question 19

Amazon Personalize Recipes

Answer 19

User Personalization Recipes: These recipes come in three flavors. 1) user personalization uses user item interaction data 2)Popularity count recommends most popular item among all users 3) Legacy recipes that use HRNN models

Ranking based recipes:USes HRNN but also ranks recommendations
Rekated Item recipe:Collaborative filtering algorithm

Question 20

Three forms of personalization data

Answer 20

user data
item data
user item interaction data

Question 21

How personalization models are evaluated

Answer 21

Precision at K: Of the k items recoemmended, how many were actually relevant divided by k

Mean Reciprocal Rank: The mean of the reciprocal rank of the first recommendation out of k, where mean is taken over all queries

Question 22

HRNN is best suited for

Answer 22

datasets greater than 100