Machine Learning

Session 5. Training to Play Pool Using Machine Learning Principals

Sponsor: D.H.

Signatories: A.A., A.G., A.Gu., J.A., M.H., M.M., N.C., S.A., T.K.

Topic: Machine Learning. Applying Machine Learning Principals to Distinguish the Best Bank Shot Player

pool ml.jpg

Taking the approach of Artificial Intelligence and Machine Learning to train a bank shot and using the algorithm to define the best player.

  1. In a retrospective of the World Champion in the game of Go, Lee Sedol’s defeat to a computer AlphaGo, poses the question: “Is this the match between human and the machine or human and humanity?”
  2. Reviews the definition of Artificial Intelligence using Max Tegmark’s maxim that “intelligence is the ability to solve complex tasks,” definition of Machine Learning adopting Arthur Samuel’s quote as “ability to learn without being explicitly programmed.”
  3. Analyses the following branches of Machine Learning as Supervised Learning, Unsupervised Learning and Deep Learning. Supervised Learning tasks being prediction of continuous numerical values and labelling the agent, for example: the former may be useful to answer the question “What is the predicted price of a property in the Downtown Dubai area in one year?” and the latter – “Is this a high-rise residential tower or a villa?” Unsupervised Learning solves the issues of processing the data before it is handed over to a supervised algorithm. Typically computer is able to cluster the property listings by neighborhoods. Deep Learning uses hidden layers in its artificial neural network to process any kind of complex function. It mimics its workings from how the human brain processes light into vision and sounds into hearing.
  4. Reviews Reinforced Learning method to be used for the purpose of the experiment in learning to play the bank shot. This method can be described as series of rewards for correctly performed functions when correct input and output pairs are not presented.
  5. Questions what type of human activity in the present state of the world can not be outsourced to computer algorithms, and analyzes one of the answers to this question: selection of bespoke design objects.
  6. Agrees that Machine Learning may influence the selection.
  7. Highlights the issue of trust in the event of such artificial influence and acknowledges the variability of trust among different social strata.
  8. Brings up the notion of freedom of choice as human, however takes a conservative view whether this freedom of choice is unbiased and not influenced.
  9. Conducts an experiment in Reinforced Learning when the objective to perform the bank shot is given: achieve the best spin, angle and speed, however the participants are provided with only limited instructions by the sponsor. During the trial set the participants received positive feedback on good performance and no feedback on less than satisfactory performance.
  10. Conducts and experiment in defining the features of the best player using the spin, speed and angle criteria. Although the three best players showing the best results in the elements of the bank shot have been identified, the sponsor and participants agree that the data collected is insufficient to make complete judgement for combining the three best performing elements together and achieving sustainable results.

Side Notes:

Dubai, July 24, 2018