Skip to main content

The Basics

Terminology​

Artificial Intelligence (AI)​

A concept in which machines can mimics human intelligence. General Artificial Intelligence is the current aim where one General AI model capable of making several intellectual task, but for the most part, AI refers to automated decision making from machine that resembles human counterpart.

Deep Learning​

A subset of AI that focuses on decision making based on how neuron in brain behaves. The 'deep' part refers to the complexity of neural architecture in learning the situation, understand the problem and spit out the solution.

Data Science​

Applying scientific methods based on available data to gain a better insight, thus come up with value. A mixture of analyst and statistician.

Machine Learning​

Considers a subset of AI but unlike deep learning which mimics neuron behavior, machine learning focuses on developing model to only solve task at hand. This model is very dependant on training data to generate accurate outputs.

Framework​

Refers to Deep Learning Framework where users can utilize to develop their deep learning architecture and model. Example of framework:

Architecture​

A blueprint or design of the neural network that we are going to build. Common architectures:

  • Recurrent Neural Network
  • Convolutional Neural Network
  • Long Short Term Memory

Some of the more advanced architecture and its variants are build upon these. For example, Residual Network (ResNet) is a very complex CNN with the so-called "identity shortcut connections" added to address the shrinking spatial resolution in CNN.

Methodology​

The AI expansion has allowed multiple companies to leverage this term and use it various marketing campaign. The Wall Street Journal reported that 40% of new startup in UK did not use AI at all.

Identify Data at Hand​

In order to effectively build an AI solution, the consensus is always to go back to data and problem at hand. From data, formulation of work needed to produce wanted output can be made. From supervised or unsupervised learning, classification or regression work, the evaluation matrix, model deployment and scalability, these are all things to consider when looking to build an AI or machine learning solution.

Finding Proper Talent​

Attracting the wrong talent is still a big issue. This is due to inability to differentiate function needed for task at hand. For example, hiring data scientist to manage database or hiring data analyst to create machine learning model.

MLOps Pipeline​

MLOps refers to the need of designing, developing and deploying AIML solution. Creating a continuous and integrated AIML pipeline helps in streamlining the input, process, model testing, validation and deployment.

Documentation​

An important often neglected process, documentation helps in tracking changes, identifying problems and replicating solutions to ensure the model developed is robust.

Famous Work​

Aritificial Intelligence has a very rapid growth where the work is built on top of one another. OpenAI is famous for its GPT-3, an NLP model that was trained with 175 billion parameters!

Google's Deepmind also is doing superb work with the creation of AlphaZero, a system that thought itself on how to play chess, shogi and go.

Resources​