What is AI? TonkaBI brings clarity to your projects.

Artificial Intelligence (AI) two words, two separate meanings: Artificial means ‘made by people’ and Intelligence means ‘the ability to learn, understand and make judgments or have opinions that are based on reason’.

Each AI domain has its own use case. For example, a reinforcement learning model would not be the preferred choice to perform a standard computer vision task such as defects in medical imagery. In this case, the preferred paradigm is of supervised learning.

Genetic Algorithms are adaptive heuristic search algorithms that are inspired by Charles Darwin’s theory of natural selection to solve optimisation problems. They are robust, provide optimisation over large space states and unlike traditional AI, do not break on slight change in input or presence of noise.

Validating an AI is subject to the purpose of the AI, and the success rate should be clearly defined prior testing. If an NLP AI has only been trained on financial customer care data from existing customers, and the test questions are all about financial forecasting then the AI would in most cases failed the test.

Supervised Learning

Supervised learning is the machine learning task of learning an activity or function which maps an input to an output based on example input-output pairs (input image of a 5 pound note – output, 5 pound note). Input labelled data is used to train the algorithm which can then map new data.

How can we validate the accuracy of a supervised learning algorithm? A new set of data is required which the model has not been trained on before – this is known as training and testing the data. A perfect outcome is when the algorithm correctly defines the classes of the input data.

Unsupervised Learning Model

Unsupervised learning is a type of machine learning that looks for previously undetected patterns in a data set with no pre-existing labels and with minimum human supervision.

How can we validate an unsupervised learning algorithm? The results can be validated by the accuracy of clustering the data. A tester can take an unlabelled set of data and perform the validation test with the algorithm in question.

Reinforcement Learning

Reinforcement learning is an area of machine learning where a software agent makes the best possible decisions and/or takes actions in an environment and tries to maximise the potential in getting a reward (the correct answer). How does reinforcement learning differ from supervised learning? A supervised learning model is trained with the correct answer(s) – labelled data, whereas reinforcement learning is not provided with the correct answer. The agent decides how to best perform the given task.

How can we validate a reinforcement learning algorithm? The trained model can be benchmarked against other models (if they exist) or against a human. The model should be validated against the situation it is trained for.

Natural Language Processing

Natural language processing or NLP is broadly defined as the manipulation of language like speech and text by software.

How can we validate the accuracy of a NLP algorithm? The accuracy of the algorithm can be evaluated through a test dataset. Defined key words or phrases are extracted, and then the results are reviewed and evaluated by human(s).

Deep Learning

Deep learning is a simulation of a human brain that has large neural networks and can process large amounts of data with very high performance. Deep learning works extremely well with supervised learning due to the use of labelled data.

Get In Touch

London

12 Hay Hill, London, W1J
6DQ UK

New York

26 Broadway, New York, NY
10004

India

Bangalore, Mumbai & Trivandrum