Download to read offline
Uploaded byAnjani Phuyal
Aws cloud computing conference
This document provides an overview of artificial intelligence, machine learning, and deep learning techniques as well as Amazon Web Services products that can be used to implement these techniques in the cloud. It begins with introductions to key concepts like machine learning, deep learning, and common business use cases. It then describes specific AWS services for machine learning like Amazon SageMaker, Rekognition for computer vision, and services for natural language processing. The document concludes with information on AWS certification in machine learning.
More Related Content
Similar to Aws cloud computing conference
Aws cloud computing conference
- 1. By Anjani Phuyal Global CTO-GeneseSolution CEO- Genese Cloud Academy AWS Cloud Computing For AI/ML © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
- 2. ● MSc Computerand Network Security with Distinction (Middlesex university) ● Bachelor of Engineering in Electronics and communication Introduction Anjani Phuyal
- 3. Some feathers onmy wing
- 4.
- 5. Artificial intelligence, machinelearning, and deep learning © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Artificial intelligence Machine learning Deep learning 5
- 6. Machine learning Machine learningis the scientific study of algorithms and statistical models to perform a task using inference instead of instructions. © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 6 Machine learning flow Prediction Model Data
- 7.
- 8. Deep learning © 2020,Amazon Web Services, Inc. or its Affiliates. All rights reserved. 8
- 9. ML and technologyadvancements © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 9 Traditional computing Modern machine learning Cloud computing and Moore’s law
- 10. Common business usecases © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Spam versus regular email Recommendations Frau d Recommended items 10
- 11.
- 12.
- 13. Job Demand inAI and ML 13
- 14. The AWS MLStack 14
- 15. Amazon Core MLTeam
- 16. Introducing Computer Vision Overviewof computer vision © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
- 17. Computer vision overview Computervision is the automated extraction of information from digital images. © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 17
- 18. Computer vision applications Publicsafety and home security Authentication and enhanced computer-human interaction Content management and analysis Medical imaging Manufacturing process control Autonomous driving © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 18
- 19. Computer vision problems Contentrecognition Image analysis • Object classification Food? Breakfast? Lunch? Dinner? © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 19
- 20. Computer vision problems Contentrecognition Image analysis • Object classification • Object detection Bounding boxes (top,left,width,height) Milk 97.1 Peaches 92.3 Ice Cream 97.1 Salad 69.5 Nuggets 77.5 Bread Roll 94.5 Confidence © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 20
- 21. Computer vision problems Contentrecognition Image analysis • Object classification • Object detection • Object segmentation Milk Peaches Ice Cream Salad Nuggets Bread Roll © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 21
- 22. Computer vision usecases Content recognition Video analysis • Instance tracking Pathing – You can capture the path of people in the scene. For example, you can use the movement of athletes during a game to identify plays for post-game analysis. © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 22
- 23. Computer vision usecases Content recognition Video analysis • Instance tracking • Action recognition Analyze shopper behavior and density in your retail store by studying the path that each person follows © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 23
- 24. Computer vision usecases Content recognition Video analysis • Instance tracking • Action recognition • Motion estimation © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 24
- 25. Introducing Computer Vision AmazonRekognition © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
- 26. Amazon Rekognition • Managedservice for image and video analysis • Types of analysis • Searchable image and video libraries • Face-based user verification • Sentiment and demographic analysis • Unsafe content detection • Text detection • Security and compliance © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 26
- 27. Amazon Rekognition Can addpowerful visual analysis to your application Is highly scalable and continuously learns Integrates with other AWS services JavaScript Python PHP .NET Ruby Java Go Node.js C++ Languages supported by the Amazon Rekognition SDKs: © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 27
- 28. Face detection • Boundingbox • Attributes • Emotions • Facial landmarks • Quality • Pose • Confidence score © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 28
- 29. Image analysis: Objectand scene detection Urban 97.6% City 97.6% Building 97.6% Car 97.9% Person 98.9% Road 93.0% Street 93.0% Downtown 95.6% Architecture 92.2% Pedestrian 90.6% © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 29
- 30. DEMO © 2020, AmazonWeb Services, Inc. or its Affiliates. All rights reserved. 30
- 31.
- 32. Introducing Natural LanguageProcessing Overview of Natural Language Processing (NLP) © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
- 33. Natural language processing(NLP) “Alexa, what’s it like outside?” © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 33
- 34. What is NLP? NLPdevelops computational algorithms to automatically analyze and represent human language. By evaluating the structure of language, machine learning systems can process large sets of words, phrases, and sentences. Word Phrase Sentence Sentence Phrase Sentence © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 34
- 35. Natural language processinguse cases Search applications Human machine interfaces Market and social research Chatbots © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 35
- 36. Natural language processingmanaged services Amazon Transcribe, Polly, Translate, Comprehend and Lex © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
- 37. Amazon Transcribe • Recognizerecorded voices • Convert streaming audio to text • Customize specialized vocabularies • Integrate with applications by using WebSockets • Build subtitles for multiple languages in real time Amazon Transcribe is a fully managed service that uses advanced machine learning technologies to recognize speech in audio files and transcribe them into text. You can use Amazon Transcribe to convert audio to text and to create applications that incorporate the content of audio files. Amazon Transcribe © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 37
- 38. Amazon Transcribe usecases Medical transcription Streaming content labeling Call center monitoring Subtitles © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 38
- 39. Amazon Polly • Generatevoice from plain text or Speech Synthesis Markup Language (SSML) format • Create output in multiple audio formats • Offers a pay-for-use policy and uses AWS infrastructure to keep costs low Amazon Polly is a managed service that converts text into lifelike speech. Amazon Polly supports multiple languages and includes various lifelike voices. Amazon Polly © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 39
- 40. Amazon Polly usecases News service production Navigation systems Language training Animation production © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 40
- 41. DEMO © 2020, AmazonWeb Services, Inc. or its Affiliates. All rights reserved. 41
- 42. Amazon Translate • Developmultilingual user experiences for your applications • Translate documents to multiple languages • Analyze incoming text in multiple languages Amazon Translate is a fully managed text translation service that uses advanced machine learning technologies to provide high- quality translation on demand. Amazon Translate © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 42
- 43. Amazon Translate usecases International websites Multilingual chatbots Software localization International media © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 43
- 44. Amazon Comprehend • Extractkey entities from a document, such as people or locations • Identify the language that is used in a document • Determine the sentiment—such as positive, negative, neutral, or mixed—that is expressed in a document • Identify the part of speech for individual words in a document Amazon Comprehend uses NLP to extract insights about the content of documents. It develops insights by recognizing the entities, key phrases, language, sentiments, and other common elements in a document. Amazon Comprehend © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 44
- 45. Amazon Comprehend usecases Document analysis Mobile app analysis Fraud detection Content management © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 45
- 46. DEMO © 2020, AmazonWeb Services, Inc. or its Affiliates. All rights reserved. 46
- 47. Amazon Lex • Builda chatbot that can interact with voice and text to ask questions, get answers, or complete tasks • Automatically scale your chatbot with AWS Lambda • Store log files of conversations for analysis Amazon Lex is an AWS service for building conversational interfaces for applications by using voice and text. With Amazon Lex, the same conversational engine that powers Amazon Alexa is now available to any developer. Amazon Lex © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 47
- 48. Amazon Lex usecases Inventory and sales Customer service interfaces Interactive assistants Database queries © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 48
- 49. Amazon SageMaker © 2020,Amazon Web Services, Inc. or its Affiliates. All rights reserved. 49 Ground Truth Set up and manage labeling jobs for highly accurate training datasets by using active learning and human labeling. Notebook Provide AWS and SageMaker SDKs and sample notebooks to create training jobs and deploy models. Trainin g Train and tune models at any scale. Use high- performance AWS algorithms, or bring your own. Inference Create models from training jobs, or import external models for hosting so you can run inferences on new data. AWS Marketplace Find, buy, and deploy ready-to-use model packages, algorithms, and data products in AWS Marketplace.
- 50. Training models withAmazon SageMaker Amazon SageMaker provides ML algorithms that are optimized for speed, scale, and accuracy. Amazon SageMaker built-in algorithms Amazon SageMaker supported frameworks Amazon SageMaker custom frameworks AWS Marketplace algorithms © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 50
- 51. AWS Academy © 2020,Amazon Web Services, Inc. or its Affiliates. All rights reserved. 51
- 52. DEMO © 2020, AmazonWeb Services, Inc. or its Affiliates. All rights reserved. 52
- 53. AWS Certification AWS CertifiedMachine Learning – Specialty © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
- 54. AWS certification exams ©2020 Amazon Web Services, Inc. or its Affiliates. All rights reserved. 54
- 55. Certification capabilities Certification validatesthe following abilities: • Select and justify the appropriate machine learning approach for a given business problem • Identify appropriate AWS services to implement machine learning solutions • Design and implement scalable, cost-optimized, reliable, and secure machine learning solutions © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 55
- 56. Certification requirements Recommended knowledgeand experience: • 1–2 years of experience developing, architecting, or running ML and deep learning workloads on the AWS Cloud • The ability to express the intuition behind basic ML algorithms • Experience in performing basic hyperparameter optimization • Experience with ML and deep learning frameworks • The ability to follow model-training best practices • The ability to follow deployment and operational best practices © 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved. 56
- 57. Anjani Phuyal Global CTO GeneseSolution Limited aphuyal anjaniphuyal anjani@phuyal.co.uk @anjaniphuyal CONNECT With US For Training Nepal Team Aruna Sharma AI/ML Interestship Inquiry Specialist +977-9801977687 E. Contact@genesecloud.academy
- 58.
Editor's Notes
- #6 This course is an introduction to machine learning (ML), and it shows where machine learning fits in with the larger picture of artificial intelligence (AI). Machine learning is a subset of AI, which is a broad branch of computer science for building machines that can do human tasks. Deep learning itself a subdomain of machine learning. To understand where these ideas fit together, you will learn about each field.
- #7 You will likely find many definitions of machine learning. No standard definition exists, so you will now learn some definitions of machine learning. Machine learning is the scientific study of algorithms and statistical models to perform a task by using inference instead of instructions. To help you better understand this idea, consider the following concrete example. Suppose that you must write an application that determines whether an email message is spam or not. Without machine learning, you write a complex series of decision statements (think if/else statements). Perhaps you use words in the subject or body, the number of links, and the length of the email to determine whether an email message is spam. It would be difficult and laborious to compile such a large set of rules to cover every possibility. However, with machine learning, you can use a list of email messages that are marked spam or not spam to train a machine learning model. The model would learn which patterns of words, lengths, and other indicators are good predictors of spam email messages. When you present the model with an email message that it did not see before, the model would predict whether it was spam or not spam. Tom Mitchell, a pioneer of machine learning, wrote this definition: “A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured by P, improves with experience E.” (Mitchell, Tom. 1997. Machine Learning. McGraw Hill. p. 2.) If you apply this concept to spam, the letters E, T, and P represent: E – The email messages that indicate spam or not T – The task of identifying spam P – The probability that the unseen email message is spam. Don’t be concerned if these ideas are difficult to understand. They are explained more completely later in this module.
- #9 Deep learning represents a significant leap forward in the capabilities for AI and ML. The theory behind deep learning was created from how the human brain works. An artificial neural network (ANN) is inspired from the biological neurons in the brain, although the implementation is different. Artificial neurons have one or more inputs and a single output. These neurons fire (or activate their outputs), which are based on a transformation of the inputs. A neural network is composed of layers of these artificial neurons, with connections between the layers. Typically, a network has input, output, and hidden layers. The output of a single neuron connects to the inputs of all the neurons in the next layer. The network is then asked to solve a problem. The input layer is populated from the training data. The neurons activate throughout the layers until an answer is presented in the output layer. The accuracy of the output is then measured. If the output hasn’t met your threshold, the training is repeated, but with slight changes to the weights of the connections between the neurons. It continues to repeat this process. Each time, it strengthens the connections that lead to success, and diminishes the connections that lead to failure. As you will see in this course, machine learning practitioners spend much time optimizing ML models. They select the best data features to train with, and they select the models with the best results. Deep learning practitioners spend almost no time on those tasks. Instead, they spend their time on modeling data with different ANN architectures. While the theory for deep learning goes back decades, the hardware necessary to run deep learning problems wasn’t generally accessible until recently. Now that it’s available, you can use deep learning to address problems that are more complex than problems you handled before.
- #10 Mainstream machine learning is a fairly recent occurrence. The mid-2000s marked the beginning of rapid advancements in machine learning and deep learning, partly because of Moore’s law and the rise of cloud computing. The result is easier access to larger, faster, and cheaper compute and storage capabilities. You can now rent computing power for a few hours for pennies, compared to huge investments that were needed to buy and operate large-scale compute clusters. In 2012, the use of neural networks began in the ImageNet Large Scale Visual Recognition Challenge, a machine learning competition for image recognition. The accuracy rate jumped to about 82 percent, and has been steadily climbing ever since. It exceeded human performance in 2015.
- #11 Machine learning is used throughout a person’s digital life. Here are some examples: Spam – Your spam filter is the result of an ML program that was trained with examples of spam and regular email messages. Recommendations – Based on books that you read or products that you buy, ML programs predict other books or products that you might want. Again, the ML program was trained with data from other readers’ habits and purchases. Credit card fraud – Similarly, the ML program was trained on examples of transactions that turned out to be fraudulent, along with transactions that were legitimate. Many more examples exist, including facial detection in social media applications to group your photos, detecting brain tumors in brain scans, or finding anomalies in X-rays.
- #17 Introducing Section 1: Introduction to computer vision.
- #18 Computer vision is an exciting space in machine learning. The advances in computing power and algorithms over the last 10 years have led to an increase in capabilities and easier access to computer vision technologies. You can think of computer vision as the automated extraction of information from digital images. Computer vision enables machines to identify people, places, and things in images with accuracy at or above human levels, with greater speed and efficiency. Often built with deep learning models, computer vision automates the extraction, analysis, classification, and understanding of useful information from a single image or a sequence of images. The image data can take many forms, such as single images, video sequences, views from multiple cameras, or three-dimensional data.
- #19 Some of the primary use cases for computer vision include these examples. Public safety and home security Computer vision with image and facial recognition can help to quickly identify unlawful entries or persons of interest. This process can result in safer communities and a more effective way of deterring crimes. Authentication and enhanced computer-human interaction Enhanced human-computer interaction can improve customer satisfaction. Examples include products that are based on customer sentiment analysis in retail outlets or faster banking services with quick authentication that is based on customer identity and preferences. Content management and analysis Millions of images are added every day to media and social channels. The use of computer vision technologies—such as metadata extraction and image classification—can improve efficiency and revenue opportunities. Autonomous driving By using computer-vision technologies, auto manufacturers can provide improved and safer self-driving car navigation, which can help realize autonomous driving and make it a reliable transportation option. Medical imaging Medical image analysis with computer vision can improve the accuracy and speed of a patient's medical diagnosis, which can result in better treatment outcomes and life expectancy. Manufacturing process control Well-trained computer vision that is incorporated into robotics can improve quality assurance and operational efficiencies in manufacturing applications. This process can result in more reliable and cost-effective products.
- #20 Classification in machine learning is used to decide which category or categories that a picture or object belongs to. This process is no different than any other classification problem for machine learning. In the picture here, what is represented? Is it breakfast, lunch, or dinner? Would the classification only be food? The answer depends on the model that you use to perform the classification. Models must be trained, and the training data provides the algorithm with the data to learn from. Say that you have a model that was trained with pictures of different types of food. You might expect a result that the image outputs categories of Milk, Cookie, Orange, Hamburger (or is it beef burger, or a vegetarian burger?), Salad, and French Fries. If you trained the model with different images, you might have classified the objects in the image as Tray, Cutlery, and Napkin. When you have multiple classes, this is known as a multi-class classification problem. When you have only two classes, this is known as a binary classification problem.
- #21 When you work with images, you might want to know both what kind of objects are in the image and the locations of those objects. Object detection provides the categories of the image and the location of the objects in the image. The location is provided by a set of coordinates for a box that surrounds the image, which is known as the bounding box. Bounding boxes for object detection typically provide top, left, width, and height coordinates that surround the images. You can use these coordinates in your applications. When objects are detected in an image, a confidence number is usually associated with that object. This percentage indicates how probable it is that the object belongs to a specific class. This confidence level is important when you want to determine an action that is based on object detection, especially in applications that use facial detection.
- #22 At the time of writing, object segmentation (also known as semantic segmentation) is a key problem in the field of computer vision. Object segmentation is similar to object detection, but you go into more detail to get fine boundaries for each detected object. It is a fine-grained inference for predicting each pixel in the image. Applications that require object segmentation include autonomous vehicles and advanced computer-human interactions. Object segmentation is not covered in this course.
- #23 With Amazon Rekognition Video, you can capture the position of each person in a video. The TrackPersons operation detects people and how they move, even when the camera is in motion. It can also attribute motion to the same person when their face is blocked or if they move in and out of the frame. The TrackPersons operation returns time segments and confidence scores.
- #24 You can analyze shopper behavior and density in your retail store by studying the path that each person follows. By using face analysis, you can also understand the average age ranges, gender distribution, and emotions expressed by the shoppers without identifying them.
- #25 Amazon Rekognition Video enables you to automatically identify thousands of objects. Example objects might include vehicles or pets; scenes like a city, beach, or wedding; and activities like delivering a package or dancing. Amazon Rekognition Video relies on motion in the video to accurately identify complex activities, such as blowing out a candle or extinguishing fire. Some examples from the image might include: Capturing the batter’s accuracy, the pitcher’s pitching style, the type of pitch (slow ball, slider, and others), the inning, and the batter’s performance versus the specific pitcher. All that data could (and does!) get used by managers to coach players on how to improve their performance. Coaches can also use the data during the game to make game-time decisions. Initiating various actions based on the speed of the ball leaving the bat and its trajectory. A hit that is calculated by an ML model could lead to: An audio or visual warning about a possible foul ball into the crowd. A preemptive alarm that a hit has a high probability of being a home run. This would enable events that follow a homerun to be well-timed and automated (such as music or fireworks when the homerun is hit by the home team).
- #26 Introducing Section 2: Analyzing images and videos.
- #27 Amazon Rekognition is a computer vision service based on deep learning. You can use it to add image and video analysis to your applications. Amazon Rekognition enables you to perform the following types of analysis: Searchable image and video libraries – Amazon Rekognition makes images and stored videos searchable so that you can discover the objects and scenes that appear in them. Face-based user verification – Amazon Rekognition enables your applications to confirm user identities by comparing their live image with a reference image. Sentiment and demographic analysis – Amazon Rekognition interprets emotional expressions, such as happy, sad, or surprise. It can also interpret demographic information from facial images, such as gender. Unsafe content detection – Amazon Rekognition can detect inappropriate content in images and in stored videos Text detection – Amazon Rekognition Text in Image enables you to recognize and extract text content from images. You need to check if the applications you build using Amazon Rekognition would fall under any regulatory restrictions as defined within your field or country. Security and compliance for Amazon Rekognition is a shared responsibility between AWS and the customer. For more information, see the AWS Compliance page (https://aws.amazon.com/compliance/)
- #28 Amazon Rekognition is an AWS managed service that enables you to integrate image and video analysis into your applications. Because it’s a managed service, Amazon Rekognition hosts the machine learning models, maintains an API, and scales out to meet demand for you. You also benefit from a set of models that constantly learn and improve. It integrates with other AWS services, such as Amazon S3 for storage and AWS Identity and Access Management (IAM) for authentication and authorization. You only need to focus on building applications that use the API and, optionally, training the service to understand your unique business requirements. Amazon Rekognition provides APIs, SDKs, and AWS Command Line Interface (AWS CLI) commands. You can use these resources to access and interact with Amazon Rekognition. The languages that the SDKs support include JavaScript, Python, PHP, .NET, Ruby, Java, GO, Node.js, and C++.
- #29 Facial detection uses a model that was tuned to perform predictions specifically for detecting faces and facial features, including: Bounding box – The coordinates of the bounding box that surrounds the face. Confidence – The level of confidence that the bounding box contains a face. Facial landmarks – An array of facial landmarks. For each landmark (such as the left eye, right eye, and mouth), the response provides the x and y coordinates. Facial attributes – A set of facial attributes, such as whether the face has a beard. For each attribute, the response provides a value. The value can be of different types, such as a Boolean type (whether a person is wearing sunglasses) or a string (whether the person is male or female). For most attributes, the response also provides a confidence score in the detected value for the attribute. Quality – Describes the brightness and the sharpness of the face. Pose – Describes the rotation of the face inside the image. Emotions – A set of emotions with confidence in the analysis. Again, confidence is a feature here, and it is provided for each feature that was detected. The feature prediction is based only on visual observation.
- #30 When Amazon Rekognition performs predictions, it often returns multiple labels. Each label has a confidence level. This confidence level indicates how likely the label was found in the image. Like this example shows, labels can also have hierarchies.
- #31 When Amazon Rekognition performs predictions, it often returns multiple labels. Each label has a confidence level. This confidence level indicates how likely the label was found in the image. Like this example shows, labels can also have hierarchies.
- #33 Introducing Section 1: Overview of natural language processing. In this section, you will review the meaning of natural language processing.
- #34 Before you see an explanation of natural language processing (NLP), look at an NLP example with Amazon Alexa. An Amazon device, such as an Echo, records your words. The recording of your speech is sent to Amazon servers to be analyzed more efficiently. Amazon breaks down your phrase into individual sounds. Then, it connects to a database that contains the pronunciations of various words to find which words most closely correspond to the combination of individual sounds. It identifies important words to make sense of the tasks and to carry out corresponding functions. For instance, if Alexa notices words like outside or temperature, it opens the weather app. Amazon servers send the information back to your device, and Alexa might speak.
- #35 NLP is a broad term for a general set of business or computational problems that you can solve with machine learning (ML). NLP systems predate ML. Two examples are speech-to-text on your old cell phone and screen readers. Many NLP systems now use some form of machine learning. NLP considers the hierarchical structure of language. Words are at the lowest layer of the hierarchy. A group of words make a phrase. The next level up consists of phrases, which make a sentence, and ultimately, sentences convey ideas. NLP systems face several significant challenges, which you will learn about next.
- #36 You can apply NLP to a wide range of problems. Some of the more common applications include: Search applications (such as Google and Bing) Human machine interfaces (such as Alexa) Sentiment analysis for marketing or political campaigns Social research that is based on media analysis Chatbots to mimic human speech in applications
- #37 Introducing Section 2: Natural language processing managed services. In this section, you will review five managed Amazon ML services that you can use for various NLP use cases. These services simplify the process of creating a NLP application.
- #38 Amazon Transcribe is the first managed machine learning service that you will learn about. You can use Amazon Transcribe to recognize speech in audio files and produce a transcription. Amazon Transcribe can recognize specific voices in an audio file, and you can create customized vocabulary for terms that are specialized for a particular domain. You can also add transcription service to your applications by integrating with WebSockets. WebSockets provide an internet-facing interface that enables two-way communication between an application and Amazon Transcribe.
- #39 Some of the more common use cases for Amazon Transcribe include: Medical transcription – Medical professionals can record their notes, and Amazon Transcribe can capture their spoken notes as text. Video subtitles – Video production organizations can generate subtitles automatically from video. It can also be done in real time for a live feed to add closed captioning (CC). Streaming content labeling – Media companies can capture and label content, and then feed the content into Amazon Comprehend for further analysis. Customer call center monitoring – Companies can record customer service or sales calls, and then analyze the results for training or strategic opportunities.
- #40 Amazon Polly can convert text into lifelike speech. You can input either plaintext files or a file in Speech Synthesis Markup Language (SSML) format. SSML is a markup language that you can use to provide special instructions for how speech should sound. For example, you might want to introduce a pause in the flow of speech. You can add an SSML tag to instruct Amazon Polly to pause between two words. You can also output speech from Amazon Polly to MP3, Vorbis, and pulse-code modulation (PCM) audio stream formats. Amazon Polly has various applications. Common use cases for Amazon Polly include mobile apps (such as newsreaders), games, eLearning platforms, and accessibility applications for visually impaired people. Amazon Polly is eligible for use with regulated workloads for the U.S. Health Insurance Portability and Accountability Act of 1996 (HIPAA), and the Payment Card Industry Data Security Standard (PCI DSS).
- #41 Some of the more common use cases for Amazon Polly include: News service production – Major news companies use Amazon Polly to generate vocal content directly from their written stories. Language training systems – Language training companies use Amazon Polly to create systems for learning a new language. Navigation systems – Amazon Polly is embedded in mapping application programming interfaces (APIs) so that developers can add voice to their geo-based applications. Animation production – Animators use Amazon Polly to add voices to their characters.
- #42 When Amazon Rekognition performs predictions, it often returns multiple labels. Each label has a confidence level. This confidence level indicates how likely the label was found in the image. Like this example shows, labels can also have hierarchies.
- #43 With Amazon Translate, you can create multilanguage experiences in your applications. You can create systems for reading documents in one language, and then render or storing it in another language. You can also use Amazon Translate as part of a document analysis system. Amazon Translate is fully integrated with other Amazon ML services, such as Amazon Comprehend, Amazon Transcribe, and Amazon Polly. With this integration, you can: Extract named entities, sentiment, and key phrases by integrating with Amazon Comprehend Create multilingual subtitles with Amazon Transcribe Speak translated content with Amazon Polly
- #44 Some of the more common use cases for Amazon Translate include: International websites – You can use Amazon Translate to quickly globalize your websites. Software localization – Localization is a major cost for all software that is aimed at a global audience. Amazon Translate can decrease software development time and significantly reduce costs for localizing software. Multilingual chatbots – Chatbots are used to create a more human-like interface to applications. With Amazon Translate, you can create a chatbot that speaks multiple languages. International media management – Companies that manage media for a global audience use Amazon Translate to reduce their costs for localization.
- #45 Amazon Comprehend implements many of the NLP techniques that you reviewed earlier in this module. You can extract key entities, perform sentiment analysis, and tag words with parts of speech.
- #46 Some of the more common use cases for Amazon Comprehend include: Analysis of legal and medical documents – Legal, insurance, and medical organizations have used Amazon Comprehend to perform many of the NLP functions that you learned about in this module. Financial fraud detection – Banking, financial, and other institutions have used Amazon Comprehend to examine very large datasets of financial transactions to uncover fraud and look for patterns of illegal transactions. Large-scale mobile app analysis – Developers of mobile apps can use Amazon Comprehend to look for patterns in how their apps are used so they can design improvements. Content management – Media and other content companies can use Amazon Comprehend to tag content for analysis and management purposes.
- #47 When Amazon Rekognition performs predictions, it often returns multiple labels. Each label has a confidence level. This confidence level indicates how likely the label was found in the image. Like this example shows, labels can also have hierarchies.
- #48 With Amazon Lex, you can add a human language frontend to your applications. Amazon Lex enables you to use the same conversational engine that powers Amazon Alexa. You can automatically increase capacity by creating AWS Lambda functions to scale on demand. In addition, you can store log files of the conversations for further analysis.
- #49 Some of the more common use cases for Amazon Lex include: Building frontend interfaces for inventory management and sales – Voice interfaces are becoming more common. Companies use Amazon Lex to add chatbots to their inventory and sales applications. Developing interactive assistants – By combining Amazon Lex with other ML services, customers create more sophisticated assistants for many different industries. Creating customer service interfaces – Human-like voice applications are quickly becoming the norm for customer service applications. Amazon Lex can reduce the time and increase the quality of these chatbots. Query databases with a human-like language – Amazon Lex is combined with other AWS database services to create sophisticated data analysis applications with a human-like language interface.
- #50 Finally, you can use Amazon SageMaker, which is an AWS service with many capabilities. Amazon SageMaker can deploy machine learning instances that run Jupyter notebooks and JupyterLab. Amazon SageMaker manages the deployment of these compute resources, so you must connect to the Jupyter environment. Amazon SageMaker also provides tools for labeling data, training models, and hosting trained models. AWS Marketplace also provides a selection of ready-to-use model packages and algorithms from machine learning developers.
- #51 Amazon SageMaker provides four different ways that you can train models. Amazon SageMaker built-in algorithms – You can train and deploy these algorithms from Amazon SageMaker console, AWS Command Line Interface (AWS CLI), a Python notebook, or the Amazon SageMaker Python SDK. The available built-in algorithms are itemized and described in the next slide. Containers are used in the background when you use one of the Amazon SageMaker built-in algorithms, but you do not deal with them directly. Amazon SageMaker supported frameworks – Amazon SageMaker provides prebuilt containers to support deep learning frameworks such as Apache MXNet, TensorFlow, PyTorch, and Chainer. It also supports ML libraries, such as scikit-learn and SparkML by providing prebuilt Docker images. If you use the Amazon SageMaker Python SDK, they are deployed by using their respective Amazon SageMaker SDK Estimator class. In this case, you supply the Python code that implements your algorithm, and configure the prebuilt image to access your code as an entry point. You might have functional requirements for an algorithm or model that you developed in a framework, which a prebuilt Amazon SageMaker Docker image doesn't support. If so, you can modify an Amazon SageMaker image to satisfy your needs. Amazon SageMaker custom frameworks – If you have no prebuilt Amazon SageMaker container image to use or modify for an advanced scenario, you can package your own script or algorithm. You can then use this script or algorithm with Amazon SageMaker. You can use any programming language or framework to develop your container. For example, your team might work and build ML models in R. If so, you can build your own containers to train and host an algorithm in R. AWS Marketplace algorithms – A third-party organization might have developed and tuned a model. It’s worth looking in the AWS Marketplace to see what ready-to-use algorithms and models are available.
- #53 When Amazon Rekognition performs predictions, it often returns multiple labels. Each label has a confidence level. This confidence level indicates how likely the label was found in the image. Like this example shows, labels can also have hierarchies.
- #54 Although this course is not designed to prepare you to achieve the AWS Certified Machine Learning – Specialty, you can continue to work towards certification. The next few slides review how you can achieve that goal.
- #55 AWS Certification helps learners build credibility and confidence by validating their cloud expertise with an industry-recognized credential. It also helps organizations identify skilled professionals who can lead cloud initiatives by using AWS. You must earn a passing score by taking a proctored exam to earn an AWS certification. After receiving a passing score, you will receive your certification credentials. AWS Certification does not publish a list of all services or features that are covered in a certification exam. However, the exam guide for each exam lists the current topic areas and objectives that are covered in the exam. You can find exam guides on the Prepare for Your AWS Certification Exam webpage. You will be required to update your certification (or recertify) every 3 years. View the AWS Certification Recertification page for more details. The information on this slide is current as of June 2020. However, exams are frequently updated. Also, the details regarding which exams are available—and what is tested by each exam—are subject to change. For the newest AWS certification exam information, go to AWS Certification.
- #56 The AWS Certified Machine Learning – Specialty means that you can select and justify the appropriate machine learning approach for a given business problem. You can also identify appropriate AWS services to implement machine learning solutions. Finally, you can design and implement scalable, cost-optimized, reliable, and secure machine learning solutions.
- #57 Before you take the AWS Certified Machine Learning – Specialty exam, we recommend that you have the following knowledge and experience. First, you should have 1–2 years of experience in developing, architecting, or running ML and deep learning workloads on the AWS Cloud. Your experience should include performing basic hyperparameter optimization and working with machine learning and deep learning frameworks. You should also be able to express the intuition behind basic ML algorithms. Finally, you should be able to follow model-training best practices along with deployment and operational best practices.