AI

Question 1

Who coined the term AI?

Answer 1

John McCarthy

Question 2

What halted AI development in the late 20th century? What caused it to pick up pace?

Answer 2

AI development was slower for years because of data and compute → called the AI winter.

AI development is now accelerating due to more powerful microchips and more data (the Internet)

Question 3

What is aleatoric uncertainty?

Answer 3

This is the inherent noise in the data. For example, variability in measurements or natural randomness

Question 4

What is Artificial Narrow Intelligence?

Answer 4

AI that is good at one specific task

Question 5

What is a Bayesian network?

Answer 5

Graphical network that maps the factors that contributed to a result and gives the factor that likely contributed the most

Question 6

What is Blob or object storage / Datalakes?

Answer 6

Most flexible DB and good for handling unstructured data such as images. Often cheaper than traditional storage

Question 7

What are Knowledge Graphs?

Answer 7

Summary panel of information (such as those given on Google for a famous person). Could be useful when building a DB of knowledge.

Question 8

What is a FLOP?

Answer 8

Floating Point operation. One FLOP is a single arithmetic operation information floating point numbers, such as addition, substraction etc.

Question 9

What is compute?

Answer 9

Compute is the number of transistors. This determines FLOPs

Question 10

What is Moore’s Law?

Answer 10

Moore’s law dictates that the number of transistors can occupy the same space halves every two years.

Question 11

What is Reinforcement learning?

Answer 11

The AI model receives inputs and tries to attain a goal. At the end, it is given positive of negative feedback and the AI model has to figure out what to do to improve the result.

Right behaviour receives positive feedback, the wrong behaviour receives negative. Up to AI to determine exactly what to input to get the optimal result. Very immature field.

Question 12

What are Generative Adversarial Networks (GANs)?

Answer 12

AI that can produce high-quality images from scratch

Question 13

What is Machine Learning

Answer 13

Algorthms that allow computer systems to learn and adapt without explicitly instruction.

Supervised, unsupervised, semi-supervised, and reinforcement learning are all types of machine learning

One way to deliver AI. Field of study that gives computers the ability to learn without being explicitly programmed quote from Arthur Samuel). Give an AI input and ML model gives output

Input and output can seem limited, but depending on the context, it can be precious

A high-performance AI needs large training data and fast and bountiful compute.

What is the input and output depend on the goal, data, and business

Supervised data can learn from unstructured and structured data

“Machine learning is an application of AI that enables systems to learn and improve from experience without being explicitly programmed.

Give the AI model the rules to distinguish X from Y, then it works from there and improves with practice and feedback

Question 14

What is semi-supervised learning?

Answer 14

Builds a model using labelled and non-labelled data

Question 15

What is deep learning

Answer 15

A type of machine learning that is based on artificial neural networks to extract insights from data.

Good when there are a lot of inputs.

Deep learning and NN’s are interchangeable terms. NN’s facilitate DL, which is a subcategory of ML.

A neural network using artificial neurons to create output from input. A neural network is a vast collection of artificial neurons.

They can assimilate multiple inputs to get an output. NN typically use unsupervised learning to figure out what drives the best output.

A neural network using artificial neurons to create output from input. A neural network is a vast collection of artificial neurons. They can assimilate multiple inputs to get an output.

In training, NN will figure out what drives the best output.

A form of ML. It uses a neural network to extract features and classify them based on patterns to provide an output. DL recognises similar features and groups images together - called clustering.

Question 16

What are Cloud deployments?

Answer 16

When you rent space on someone else’s servers to run AI model

Question 17

What is Transfer learning?

Answer 17

Learnings and infrastructure of an AI model trained on one tasks is helpful in achieving another task. The tasks often share metrics or processes.

Question 18

What is Computer vision?

Answer 18

Image recognition and classification, object detection, object positional detection, Image segmentation (object recognition with exact boundaries showing what pixels belong to each object), racking (objects over time)

Question 19

What is a CPU?

Answer 19

Made by Intel and AMD - compute in laptop

Question 20

What is Data science?

Answer 20

Manually finding trends from data

Question 21

What is Deep Evidential Learning?

Answer 21

Deep learning that provides predictions and the uncertainty of those predictions. It can give the confidence of a model’s output.

DEL gives information on risk and robustness of an answer, leading to better decision making.

Applications include medical diagnosis, autonomous vehicles, and financial forecasting.

Question 22

What is an Edge deployment?

Answer 22

On-location compute. AI model makes decision on location.

Question 23

What is an Epistemic Uncertainty?

Answer 23

This stems from the model’s lack of knowledge. For instance, uncertainty due to insufficient training data or model parameters

Question 24

What is Generative AI ?

Answer 24

AI that can produce high-quality text, images, and audio

Question 25

What are Graphical models?

Answer 25

Probabilistic, graphical representation of data points that show a specific distribution. Two common examples are Bayesian networks and Markov random fields

Question 26

What is a GPU?

Answer 26

Historically made to process images and graphics. Good for AI. Developing more powerful GPUs to develop more powerful AI models.

Question 27

What is a Knowledge Graph?

Answer 27

Summary panel of place, person or thing containing key information (i.e. name, DOB, DOD of a person). This could be useful in building a DB of knowledge

Question 28

What are Markov random fields?

Answer 28

Graphical representation of probabilities that are non-directional

Question 29

What is Natural Language Processing (NLP)?

Answer 29

AI understanding natural language - Text classification (input text and classify), information Retrieval (ability to search within company documents), name entity recognition (finding all words belonging to a particular category in a sentence), Machine Translation - translating to and from other natural languages

Question 30

What are non-relational (NoSQL) databases?

Answer 30

DBs with dynamic schemes to store different data types (i;.e. documentation, oriented by column, key value pairs, and nodes)?

Question 31

What is on-prem?

Answer 31

Your company owns the servers and AI model is run internally

Question 32

What is a prediction?

Answer 32

Output of AI model or data science model. Using past data to make predictions for the future

Question 33

What is Robotic Process Automation (RPA)?

Answer 33

A set of technologies that replicates the manual interactions of human agents withvisualinterfaces.

Question 34

What is a Relational Databases?

Answer 34

Tables with relationships based on common key fields

Question 35

What is semi-structured data?

Answer 35

Some form of structure, but not a schema (fully structured).

Question 36

What is Speech recognition (speech to text)?

Answer 36

Trigger/wake word speaker ID (listen to the speech and figure out identity), speech synthesis (text-to-speech or TTS)

Question 37

What is structured data?

Answer 37

Pure data sets. Data in columns and rows with some standardisation (format is called schema).

Question 38

What is supervised learning?

Answer 38

Supervised learning is learning using data labelled by a human. This gradually guides the development and output of the AI model

Give an input (labelled data) and get an output. Feedback is given on the output and AI model adjusts weighting accordingly.

Give an input of labelled data. and AI outputs this is supervised learning. Examples are spam filtering, speech recognition, translation, online advertising, generative AI. It seems limited, but depending on the context, it can be precious. What is the input and output depend on the goal, data, and business. Supervised learning can learn from unstructured and structured data It uses human knowledge to looks for patterns. The quality of AI model depends on the quality of the data and the annotations.

In supervised learning, one needs to highlight the problem and determine inputs and outputs

Question 39

What is Text/Image/Data Mining?

Answer 39

Detecting specific information from pieces of text

Question 40

What is a TPU?

Answer 40

A computer chip developed by Google specifically for neural network machine learning

Question 41

What is Traditional AI?

Answer 41

A model that uses data analysis to identify patterns, make predictions, and perform specific tasks. It’s a highly efficient problem solver. It’s scope of operations is limited.

Question 42

What is Transfer Learning?

Answer 42

Learning from task A helps with a related task B. The tasks may share metrics or similarities

Question 43

What is unstructured data ?

Answer 43

Formats that consist of many data types. Humans find these very easy to comprehend, but it can be tricky for AI. Examples are images, audio, and text.

Question 44

What is Unsupervised learning?

Answer 44

Most common use is clustering - mapping data points onto a graph with two axis and searching for clusters of points to determine insights. Good for marketing segmentation

Do not have to give the AI a goal and can leave it to find insights itself

Thought to be the way the human brain learns and perhaps a massive area of development in AI.

Giving an AI model data without a hypothesis and asking it to find it’s own conclusion. The most common application is finding clusters of data points. It automatically segments data which is good for finding trends and marketing segemtnation

Thought to be the way the human brain learns and perhaps a massive area of development in AI

Question 45

What is a diffusion model?

Answer 45

GenAI model that recognises the sequential (Markhov) pattern of a dataset, then can generate new elements that are distributed in the same manner.

Diffusion models are typically used in computer vision tasks such as vide and image generation, denoising (noise reduction), and inpainting.

Question 46

What is inpainting?

Answer 46

A conservation process in which missing or damaged parts of an image are reinstated to create a complete image.

Question 47

What is a convolutional neural network?

Answer 47

A reguarlisaed, feed-forward neural network that learns by itself via kernal optimisation. Traditional backpropagation and rapid cycles led to very extreme results. A CNN uses regularised weights with fewer connections.

Question 48

What is a diffusion model?

Answer 48

Diffusion Models are generative models, meaning that they are used to generate data similar to the data on which they are trained. Fundamentally, Diffusion Models work by destroying training data through the successive addition of Gaussian noise, and then learning to recover the data by reversing this noising process

Question 49

What is a Language Processing Unit (LPU)?

Answer 49

Developed by Groq, LPUs are specifically designed to handle the unique speed and memory demands of LLMs. AnLPU is an especially fast processor designed for computationally intensive applications that are sequential in nature rather than parallel.