AWS AI/ML Tech Flashcards
What are The 5 ML Development Life Cycle Stages
- Planning: Defining the problem and business goals. [citation:4]
- Data Preparation: Gathering, cleaning, and organizing data.
- Model Development: Selecting, training, and tuning the model. [citation:4]
- Deployment: Integrating the model into a live environment.
-
Monitoring: Tracking performance and making adjustments as needed. [citation:4]
1.
How to identifying opportunities
1. Business Needs
- Identify Pain Points and Challenges: Begin by thoroughly understanding your organization’s current challenges and pain points. This could involve analyzing operational inefficiencies, customer churn, or areas where manual processes hinder productivity.
- Explore Growth Opportunities: Look beyond immediate challenges and consider how AI can unlock new growth avenues. This might involve personalizing customer experiences, automating tasks, or developing innovative products and services.
2. Focus on Value Creation:
- Quantify Potential Impact: Estimate the potential ROI of AI initiatives by considering factors like cost savings, revenue growth, and improved efficiency.
- Prioritize High-Impact Use Cases: Focus on projects that offer the greatest potential for positive business outcomes and align with strategic goals.
3. Consider Feasibility:
- Data Availability and Quality: Assess the availability and quality of data required to train and operate AI models effectively.
- Resource Requirements: Evaluate the need for infrastructure, computing power, and skilled personnel to support AI initiatives.
- **Expertise and Skillset: **Determine if your team possesses the necessary expertise or if additional training or external support is required.
Examples:
Start with readily available data: If you have existing customer data or operational logs, explore how they can be used to build initial AI models.
Utilize AWS Free Tier: Leverage AWS Free Tier to experiment with AI services and build proof-of-concept projects without significant upfront costs.
Engage with AWS Partner Network: Collaborate with AWS Partners to access specialized expertise and accelerate AI implementation.
**What are the 3 evaluation typs of entripise data ? **
1. Data Aware: Primarily records and stores data, without active use in decision-making; responsibility for data is not clearly defined.
2. Data Informed: Analyzes data to inform decisions, with specific roles assigned for managing data; employs interactive data tools for insights.
3. Data Driven: Integrates data into all decision-making processes, with a company-wide commitment to data; utilizes advanced technologies like AI for strategic actions.
1. Data Aware:
**Capability: **Focus is on the collection of data with an emphasis on knowing what has happened, signifying a basic level of data utilization.
**Ownership: **There’s no designated responsibility for data management, indicating a lack of strategic importance placed on data assets.
**Technology: **Relies on legacy systems (old guard databases) that support simple data storage without advanced analysis or integration capabilities.
2. Data Informed:
**Capability: **Uses insights gained from data analysis to understand why events have occurred, pointing to a more analytical approach to data.
Ownership: Data is managed by designated individuals or teams, reflecting an organizational move towards recognizing data as a valuable asset.
Technology: Incorporates more sophisticated tools like data warehouses and interactive queries, allowing for deeper analysis and pattern recognition.
3. Data Driven:
**Capability: **Prioritizes action based on data analytics, with the aim to influence future outcomes, demonstrating the highest level of data maturity.
**Ownership: **Data is a responsibility shared across the organization, indicating a culture that values and utilizes data in decision-making processes.
Technology: Employs cutting-edge technologies like cloud computing, artificial intelligence, machine learning, and generative AI, enabling real-time analysis and proactive decision-making.
Define High-Impact AI Initiatives (HI-AIs)
- high-impact AI initiative (HI-AI) is an AI opportunity that is feasible, has a clear short-term and long-term business impact, and minimizes risks.
High-Impact AI Initiatives (HI-AIs)
It involves:
1. Identifying or recognizing potential AI initiatives (PAIs)
- Framing potential AI initiatives for clarity on benefits and measurability
- Scoring initiatives to shortlist and prioritize HI-AI initiatives
- Verifying with experts on the viability of the initiatives
What is Amazon CodeWhisperer?
Amazon CodeWhisperer is an AI-powered coding assistant that provides real-time, context-aware code suggestions to enhance developer productivity
What is ML?
“ML” stands for Machine Learning, which is a branch of artificial intelligence (AI). It involves training algorithms to make predictions or decisions based on data. Machine learning models automatically improve their performance as they are exposed to more data over time. It’s used in a wide range of applications, from recommendation systems in online platforms to autonomous vehicles.
What are the 3 types of ML algorithms?
- Reinforcement Learning (RL):
* Dynamic Decision Processes: RL is suitable for scenarios involving dynamic decision-making where an agent interacts with an environment to achieve a specific goal through trial and error.
- Sequential Decision-Making: Ideal for tasks that require sequential decision-making, learning from feedback to optimize actions over time.
- Delayed Rewards:Effective for problems with delayed rewards, where the consequences of actions are not immediate.
- Real-World Applications: Widely used in autonomous driving, robotics manipulation, NLP, finance, and industry automation.
- Example: Training robots for complex movements or optimizing resource allocation in dynamic environments.
- Unsupervised Learning:
* Pattern Discovery: Unsupervised learning is valuable for tasks focused on finding patterns, relationships, or structures within data without explicit guidance.
* Exploratory Data Analysis: Commonly used for clustering, dimensionality reduction, and anomaly detection.
* Data Segmentation: Ideal for segmenting data into meaningful groups based on underlying patterns.
Real-World Applications: Useful when there is a large amount of unlabeled data or to understand the inherent structure of the data.
* Example: Segmenting customer preferences for targeted marketing or reducing data dimensionality for visualization. - Supervised Learning:
* Labeled Data: Supervised learning requires labeled training data where the model learns from input-output pairs to make predictions.
Predictive Modeling: Ideal for tasks focused on making predictions or classifying new data points based on historical data.
* Regression & Classification: Commonly used for regression tasks (predicting continuous values) and classification tasks (predicting categories).
* Real-World Applications: Widely applied in image recognition, natural language processing, and customer behavior prediction.
* Example: Spam email detection, medical diagnosis, image classification, and predictive maintenance.
Examples:
Supervised Learning
* Spam prediction, Fraudulent transaction detectior
* Customer churn prediction
* Machine failure prediction
* Forecasting staffing levels
* Forecasting raw material prices
* Forecasting consumer demand
Unsupervised Learning
* Micro-segmentation of custoers
* Recommendations of products to purchase
* Customer behavior analysis (market basket
What is Deep Learning?
- Deep learning is a special class of ML that uses neural networks behind the scenes. Neural networks try to simulate the way the human brain works, with many densely interconnected brain cells. The idea is to replicate this inside a computer, so you can get it to learn things, recognize patterns, and make decisions in a humanlike way.
- Unlike traditional ML, the beauty of deep learning is that it can automatically uncover features in data that it should use for learning to make optimal predictions.
Let’s take the house price prediction problem that we talked about earlier. With neural networks, you can present all possible input -from zip code to neighborhood to the average median price of houses in the city. Neural networks can decide which features are essential and which ones can be excluded or relied on less. These networks will also learn which combination of input can make the best prediction. In contrast, with traditional ML, you’d have to experiment with different combinations of features. Even though a neural network can automatically learn which features to use, for it to work as expected, the network needs to be designed correctly and fed large volumes of high-quality training data.
Provide two types of GenAI models
Transformers
Description: Transformers are an innovative type of neural network architecture that can understand and process the context of words within a sentence or text sequence. They employ mechanisms such as self-attention to assign appropriate weights to the words based on their context within the input sequence.
GANs (Generative Adversarial Networks)
Description: GANs are a type of generative model characterized by their two main components: a generator and a discriminator. The generator attempts to produce synthetic data, such as images, while the discriminator aims to differentiate between real and generated data. GANs are built on an adversarial training paradigm whereby the generator and discriminator engage in a competitive dialogue, resulting in the generator improving its capacity to create more realistic outputs.
How do large language models work?
- A key factor in how LLMs work is the way they represent words. Earlier forms of machine learning used a numerical table to represent each word. But, this form of representation could not recognize relationships between words such as words with similar meanings.
- This limitation was overcome by using multi-dimensional vectors, commonly referred to as word embeddings, to represent words so that words with similar contextual meanings or other relationships are close to each other in the vector space.
- Using word embeddings, transformers can pre-process text as numerical representations through the encoder and understand the context of words and phrases with similar meanings as well as other relationships between words such as parts of speech. It is then possible for LLMs to apply this knowledge of the language through the decoder to produce a unique output
What are applications of large language models?
“There are many practical applications for LLMs.
Copywriting
Apart from GPT-3 and ChatGPT, Claude, Llama 2, Cohere Command, and Jurassiccan write original copy. AI21 Wordspice suggests changes to original sentences to improve style and voice.
Knowledge base answering
Often referred to as knowledge-intensive natural language processing (KI-NLP), the technique refers to LLMs that can answer specific questions from information help in digital archives. An example is the ability of AI21 Studio playground to answer general knowledge questions.
Text classification
Using clustering, LLMs can classify text with similar meanings or sentiments. Uses include measuring customer sentiment, determining the relationship between texts, and document search.
Code generation
LLM are proficient in code generation from natural language prompts. Examples include Amazon CodeWhisperer and Open AI’s codex used in GitHub Copilot, which can code in Python, JavaScript, Ruby and several other programming languages. Other coding applications include creating SQL queries, writing shell commands and website design.
Text generation
Similar to code generation, text generation can complete incomplete sentences, write product documentation or, like Alexa Create, write a short children’s story.”
What is natural language processing?
- Large language models (LLM) are huge deep learning models trained on massive datasets.
- They use transformer neural networks with an encoder and a decoder for self-attention.
- Transformer models are capable of unsupervised training and self-learning to understand languages and knowledge.
- Unlike earlier models, transformers process entire sequences in parallel, reducing training time.
- The large-scale transformer architecture allows the use of models with hundreds of billions of parameters and the ingestion of massive amounts of data.
How Language Models work?
Understanding Language Models
- An essential aspect of how Large Language Models (LLMs) operate is the method of word representation.
- Previous machine learning approaches used a numeric table for word representation, unable to recognize relationships between words like those with similar meanings.
- Multi-dimensional vectors, known as word embeddings, have overcome this limitation by representing words in a manner where words with related contextual meanings or other connections are positioned closely together in the vector space.
- Transformers, utilizing word embeddings, process text as numerical representations via the encoder, comprehending word and phrase contexts, as well as other word relationships, such as parts of speech.
- Subsequently, the LLM applies this language knowledge via the decoder to generate a unique output.
What is Amazon quickSight (Gen AI serverless for BI)
- Amazon QuickSight is a cloud-based business intelligence tool that helps users analyze data and gain insights. It has the following key features:
- QuickSight allows users to connect to data from various sources like AWS, third-party data, spreadsheets and SaaS applications. It can combine different types of data into a single dashboard for reporting and analysis.
- QuickSight authors can export data to SageMaker Canvas to build ML models without any coding. Various algorithms can be used to create predictive models for tasks like forecasting, anomaly detection, and more.
How Amazon Web Services (AWS) Services for GAN Support?
Amazon Web Services (AWS) Services for GAN Support
-
Amazon SageMaker
- Fully managed service for preparing data and building, training, and deploying machine learning models.
- Offers fully managed infrastructure, tools, and workflows for diverse model applications.
- Features tailored to accelerate GAN development and training for various applications.
-
Amazon Bedrock
- Fully managed service providing access to foundation models (FMs) or trained deep neural networks from Amazon and leading AI startups.
- Offers FMs through APIs, allowing flexibility to select the most suitable model for specific requirements.
- Enables swift development and deployment of scalable, reliable, and secure generative AI applications without infrastructure management.
-
AWS DeepComposer
- Offers a creative approach for ML initiation, allowing hands-on experience with a musical keyboard and modern ML techniques to enhance ML skills.
- Regardless of ML or music background, developers can engage in GAN training and optimization to create original music.
