Welcome to Curiosity Point – ML/AI in AWS, your resource for a comprehensive understanding of Machine Learning (ML) and Amazon Web Services (AWS). Over this series of articles, this blog will delve into the multi-faceted roles these technologies play in the contemporary business landscape.
Why Machine Learning
Machine Learning is far more than a mere buzzword; it serves as a pivotal engine for technological advancements across a multitude of industries. Areas of applicability range from healthcare applications like predictive diagnostics to financial uses such as fraud detection; passing through IT Services for automated service requests resolution and practically all human activities.
What is Machine Learning
From a words definition perspective, machine learning are the techniques and technologies used to create systems that learn. Pushing that definition a bit further we can think of ML as the way for a system to, based on a previous set of data (model) become capable of identifying patterns and infering results.
Understanding ML involves familiarizing oneself with its main target objectives (predictive analysis, data clustering and intricate decision making) based on their corresponding machine learning technique categories (supervised learning, unsupervised learning, reinforcement learning)
Machine Learning in general can be thought as a blackbox process that has two inputs: training data and new data; and has one output: inferences. The following diagram illustrates with more detail a generic idea of a machine learning process. Everything begins with a big data source containing the data the system will learn from. This data is first preprocessed to clean up invalid entries and then it becomes the training dataset which will be used to actually train the ML system; a sub-set of the training dataset is separated (validation dataset)for validation of the overall process. The training data is passed through a process called training engine which, based on one or more machine learning algorithms generates a trained model. Among other data, this model is a dataset that contains the weights given to input characteristics (features) used later to make inferences through a so-called inference engine. At this point the system is able to take input data previously unknown by the system and output inferences about it. This inference process is then validated by comparing results from the validation dataset.
Predictive Analysis via Supervised Learning
Systems using ML can predict results of processes based on previous results data. This is done through a ML process named Supervised Learning; which is feeding the system with data (model) that includes previous results to enable the system to infer new results.
An example of this is “finding cat pictures” The system is fed with a lot of pictures with cats and without cats; including labeling the pictures as having cats or not. Then the system becomes capable of telling if a new picture has cats or not.
Another example of Supervised Learning is Amazon’s recommendation system. Here, algorithms are trained using labeled data to generate accurate and effective product suggestions for customers.
Data Clustering via Unsupervised Learning
Unsupervised Learning is the process used to feed an ML system or model with data that is not previously labeled, and using statistical methods to identify patterns or groups (clusters) within the data. For instance we can feed the system with data about student’s height; and then the system will be capable of identifying clusters in the data as short, mid-height and tall students. Applications of these techniques range from market segmentation to anomaly detection in network security.
Complex Decision Making via Reinforcement Learning
Reinforcement learning consists of providing systems with feedback loops so they can learn from its own results and become better, enabling these systems to take more accurate decisions over complex sets of data and conditions. For instance, autonomous vehicles utilize this ML techniques to enhance their navigation capabilities through real-time data collection and immediate feedback mechanisms.
The Growing Relevance of Machine Learning
The significance of ML is increasingly prominent, affecting various aspects of our daily lives. Whether it is personalized content on streaming platforms or real-time stock market analytics, Machine Learning’s impact is ubiquitous.
As ML technologies and science field evolves over time, new solutions and applications are discovered and created over time; combining the three types of learning and aggregating large datasets and models.
Key Terms in Machine Learning
For a comprehensive understanding of ML, it is essential to grasp its unique terminology, including “features,” “labels,” “training data,” and “inference.” Each term serves a distinct role in the complete lifecycle of a Machine Learning project.
- Features are the input characteristics considered for an ML learning model in a dataset.
- Labels are pre-defined results out of existing known data
- Machine Learning Algorithm is the specific computational procedure that is fed with training data as an input and outpurs a training model that can later be used to make inferences on new data.
- Training data or dataset is the data used to feed a ML algorithm to obtain a trained model
- Training Model is the result of applying a machine learning algorithm to one or more rounds of training data and enables our system to make inferences on new data.
- Inference refers to the results obtained through the ML models, wether it is a particular label (such as “has cats”) or a particular cluster-set definition for our data or a combination of them.
- Generative AI is the use of AI hybrid techniques to create systems capable of generating content such as text, audio and video.
Practical Implementations
It’s crucial to recognize that Machine Learning is not a theoretical concept but a practical technology. It is embedded in everyday applications, from the smartphone cameras that we use to capture moments, to advanced home security systems that protect us.
AWS and Machine Learning
Amazon Web Services (AWS) acts as a powerful ally for deploying Machine Learning models and creating applications that benefit from those models. Offering scalability and a comprehensive suite of tools and services, AWS allows businesses to take their ML projects from concept to implementation with ease. The following “Top-5” list provides a glance of AWS Services available in the ML/AI field.
1. Amazon SageMaker
Purpose: End-to-End Machine Learning Platform
Key Features: SageMaker provides an integrated set of tools for the entire ML lifecycle, including data labeling, model training, optimization, and deployment. It simplifies the process of building, training, and deploying machine learning models at scale.
2. AWS DeepLens
Purpose: Deep Learning-Enabled Video Camera
Key Features: DeepLens is designed to run deep learning models locally, providing a hardware-software integrated solution for developing computer vision applications. It allows developers to deploy models directly to the device.
3. Amazon Comprehend
Purpose: Natural Language Processing (NLP) Service
Key Features: Comprehend uses machine learning to find insights and relationships in text, capable of performing sentiment analysis, entity recognition, language detection, and more.
4. Amazon Rekognition
Purpose: Image and Video Analysis
Key Features: Rekognition provides highly accurate facial recognition, object and scene detection, and text extraction features, among others. It is particularly useful for security applications, content filtering, and user verification.
5. Amazon Bedrock
Purpose: Fully managed service for foundation models
Key Features: Amazon Bedrock offers access to a variety of foundation models such as Ai21, Anthropic, Cohere, Meta, Stability AI and Amazon; for the implementation of generative AI solutions.
Upcoming Topics
Stay tuned to Curiosity Point for future posts that will delve deeper into the specifics of ML algorithms, explore the services offered by AWS, and discuss real-world applications and ethical considerations in the expanding landscape of Machine Learning.