Glossaries - AI & ML Services

1.  What is Artificial Intelligence (AI)?

Artificial Intelligence (AI) refers to the capability of computers and other types of hardware to mimic human cognition by simulating the way that humans think through the use of computerized algorithms and rules. As such, these machines are capable of achieving this intelligence by utilizing complex algorithms to automate, perform, and learn from human decisions and conduct complex analyses.

2. What is Machine Learning (ML)?

Machine Learning (ML) involves the ability for a computer system to autonomously learn to perform tasks without being explicitly programmed. This ability enables these systems to build their knowledge by learning from experience and developing their capabilities through experience. ML is used in all types of applications to create machine-led analytical systems.

3. What is Deep Learning?

Deep learning refers to a category of machine learning employing artificial neural networks with numerous layers of neurons for analyzing larger amounts of data. The ability of deep learning systems to analyze large volumes of different forms of unstructured data (e.g., photographs, sound recordings, text) means that many practical applications of deep learning technology exist: image detection, natural language processing, and autonomous vehicles.

4. What is Natural Language Processing (NLP)?

Natural Language Processing (NLP) uses the tools of artificial intelligence to enable machines to recognize, understand, and create language. Natural language processing also has many applications of its own such as chatbots, sentiment detection, translation, voice assistants, summarisers, and many more. It supports users with more natural interactive experiences with computers than earlier methods.

5. What is Data Remodeling in the AI/ML context?

Data Remodeling is a collective process to reconfigure original data, down into a form that is structured, clean, and usable by means of a model developed with the algorithms and methods of machine learning and artificial intelligence. The process includes removing unwanted content from the raw data, engineering the best characteristics to use for measuring accuracy and performance, creating the proper metrics, assembling related information into a complete, high-quality dataset, and then delivering the high-quality datasets back to the model.

6. What is the difference between Supervised and Unsupervised Learning?

Supervised Learning uses labelled data in order to provide training to the model to achieve a specific predicted output or classification, such as spam detection or price prediction. 

On the other hand, Unsupervised Learning uses unlabeled data to automatically discover hidden patterns and other relationships, such as segmenting customers or detecting anomalies, without a predefined outcome.

7. What is Computer Vision?

Computer Vision is a branch of Artificial Intelligence used to help machines interpret and understand images or video. Examples of computer vision technology include 

• facial recognition, 
• object detection, 
• medical imaging analysis,
•  manufacturing quality inspection, and 
• navigating autonomous vehicles.

8. What is Business Intelligence in the AI context?

Artificial Intelligence-enabled Business Intelligence integrates the methods of analysing data in traditional business analysis with Machine Learning algorithms to provide enhanced analytical capabilities and insights, predictive analytics capabilities, and the ability to automate decision-making. It allows businesses to take advantage of analysing large data volumes and to be able to identify trends within their data and make informed strategic business decisions more quickly and accurately.

9. What is Data Forecasting?

Data Forecasting utilizes the available data and the machine learning algorithms to make predictions regarding upcoming trends, results, behaviors, and similar elements. These include, but are not limited to: 

• sales forecasting
• demand planning
• financial forecasting
• resource optimisation, and 
• risk assessment

This helps companies to take proactive action.

10. What is Model Training in Machine Learning?

Model Training is the process of feeding data to machine learning algorithms so they can learn patterns and relationships. While this process is occurring, the algorithm will continuously be adjusting (Changing) the value of Internal Parameters to eliminate errors associated with Training, so that it has a greater accuracy when predicting output than the errors produced by the previous attempts to predict output. The most traditional method to decrease error in a Machine Learning Model is through the combination of both the Backpropagation Algorithm and the use of Gradient Descent.

11. What is Feature Engineering?

When building machine learning models through the application of feature engineering, you will identify the features (data elements) in your data set that will be relevant for the model to predict the output. Through the use of domain knowledge, statistical analysis, and creativity, you will create or transform raw feature(s) into additional features to create new.

12. What is Transfer Learning?

Transfer learning is a unique process that allows an ML Model to benefit from previous work accomplished by the same or other ML Models, so that less time and fewer computational resources are necessary for developing an ML Model for a different but related task. Transfer Learning makes it feasible to build ML Models with very small amounts of data.

13. What are Chatbots and Conversational AI?

Powered by artificial intelligence, Chatbots are designed as computer programs that mimic human conversation via written text and/or spoken words. Organizations have begun to utilize advanced Natural Language Processing (NLP) capabilities, Context Awareness, and Personalized Responsiveness to improve the overall service, response quality, and user engagement levels of their Chatbot implementations.

14. What is Predictive Analytics?

Predictive analytics relies on both machine learning and statistical algorithms, which allow us to examine historical data to anticipate upcoming occurrences or behaviours using predictive models. Businesses frequently employ predictive analytics to project customer attrition, predict when an item will need to be repaired, detect fraud and establish market trends.

15. What is Reinforcement Learning?

In reinforcement learning, the learner is considered an agent who acts within its environment based upon feedback received from actions taken and subsequent rewards received or penalties assigned. Applications of reinforcement learning include robotics, games, autonomous systems, and optimization.

16. What is the difference between Batch Processing and Real-time Processing in AI?

Batch processing is the method of processing a great deal of information acquired over time in batches according to a set time interval. Batch processing allows for the implementation of historical analysis and the generation of reports from processed data. 

Real-time processing allows for processing incoming information immediately, allowing for the ability to derive immediate insights and to take action upon the insights gathered as they occur. Real-time processing is often found in applications such as

• fraud detection systems, 
• recommendation systems, 
• live monitoring applications, etc.

17. What are AI Model Deployment strategies?

The four primary strategies for deploying an artificial intelligence (AI) model are: 

1. cloud-based deployment (e.g., Amazon Web Services, Microsoft Azure, Google Cloud Platform);
2. edge (deployment for low-latency) deployment; 
3. containerization (Docker/Kubernetes); and 
4. serverless deployment. 

The deployment method to be employed depends upon critical performance criteria, scalability requirements, and security policies.

18. What is Data Annotation?

Data annotation is the process of creating training datasets for supervised machine learning by tagging data (such as images, text, audio or video) with relevant labels. 

Examples of the scope of the data annotation process include 

• tagging items in a photo
• transcribing audio recordings
• classifying text and marking outlines of pictures. 

All of these actions help ensure the machine learning models are accurate when trained on the data.

19. What are AI Ethics and Responsible AI?

AI ethics is concerned with how AI systems should be developed fairly, transparently and responsibly. It addresses concerns about mitigating bias, protecting privacy, being responsible for one’s actions, providing explanations, and addressing social impacts, so that AI can improve society and limit the risk of harm.

20. What is Model Validation?

Model validation is the process of determining how well a trained machine learning model performs on data that it hasn’t seen before, using methods such as 

• Cross-validation
• hold-out validation and 
• k-fold validation. 

Model validation ensures that the machine learning model is generalising well to new datasets and is not overfitting to its training set.

21. What are AI APIs?

AI APIs are Application Programming Interfaces that allow developers to use prebuilt artificial intelligence (AI) functionality without having to create the AI models on their own. AI APIs are offered by vendors such as 

• Google
• Amazon Web Services, and 
• Microsoft Azure.

Examples of AI APIs include 

• Vision APIs
• Language APIs
• Speech APIs, and 
• Recommendation APIs.

22. What is Auto ML (Automated Machine Learning)?

Automated Machine Learning (Auto ML) allows us to quickly and easily apply Automated Machine Learning to solve real-world problems. Automated Machine Learning includes developing features, choosing models, adjusting hyperparameters, and deploying your models. 

Automated Machine Learning allows Machine Learning professionals to accelerate their development processes.

23. What is Edge AI?

Edge AI involves deploying AI at the “edge”, i.e., within smart devices (e.g., smartphones or IoT sensors) and processing the information as soon as it is received, instead of sending the information to the cloud (servers) for processing. This results in faster processing, less latency, and a better level of privacy for applications such as autonomous vehicles and smart devices. Edge AI allows for offline processing.

24. What is the difference between AI Model Training and Inference?

Model training is the process of teaching an AI model with a large dataset. This process takes a significant amount of computational resources and time. 

Inference uses the trained AI model to predict outcomes for new data. The inference process is usually much quicker and requires far fewer computational resources than the model training process.

25. What are Generative AI Models?

Generative AI models create unique content (e.g., text, images, music, code) based on learned patterns from the models’ training data. Examples of models that generate content are 

• GPT for creating text
• DALL-E for creating images, and 
• Generative models that produce new and original creative content.

26. What are Hyperparameters in Machine Learning?

Hyperparameters are the settings that determine how machine learning algorithms learn. Examples include 

• the learning rate
• number of layers
• batch size, and 
• regularisation strength. 

Finding the right hyperparameter values is key to maximising the performance of a model.

27. What is Data Augmentation?

Data Augmentation helps to produce synthetic training data by changing existing training data (for example, rotating images, adding noise, and changing the wording of a sentence) so that there are more data points available to train on and more variation in what the model sees. This makes the model more robust and helps to reduce the risk of overfitting when there is not enough training data.

28. What is Object Detection?

Object Detection is the process of detecting and locating multiple objects within images or videos by marking them in rectangles (bounding boxes) and identifying what type of object they are. Object Detection can be used in 

• self-driving cars
• security systems
• retailer analytics, and
• In the field of Robotics.

29. What is Clustering in Machine Learning?

Clustering is the process of grouping similar data points without defining labels ahead of time to see how the data naturally breaks down. The most commonly used algorithms for clustering are K-means, hierarchical clustering, and DBSCAN, which are used for 

• Segmentation of customers
• Pattern recognition, and 
• Data exploration.

30. What is the ROI of implementing AI/ML solutions?

The Return on Investment (ROI) that organizations get when they implement Artificial Intelligence (AI) and Machine Learning (ML) technologies, which include 

• Improvements in operational efficiency (generally up to 30%) 
• Improved customer satisfaction, 
• Faster product innovation, 
• Better decision-making through predictive insights, and  
• A competitive advantage through data-based strategies.

31. What security measures are essential for AI/ML applications?

Some of the essential security measures to protect your business are: 

• Encryption of data
• Control over who has access to it
• Safe deployment of models
• Regular audits of the security of the systems
• Techniques for preserving privacy
• Prevention of adversarial attacks
• Monitoring for bias
• Meeting regulatory standards (such as GDPR and CCPA), and 
• Having an API management process that is secure.