Ceng 491 - Project Kickoff Document Template (Page 2 of 3) in pdf

Detailed Descriptions of High-Level Work packages

WP1 - (Research on ML basics, implementations, literature search and tool and specifying game

types to work on.)

In this work package, the following functionalities / features / work items will be implemented

1. To be aware of what will be done in background, basic ML concepts and algorithms will be

reviewed and implemented.

2. To get the idea and algorithm analysis, similar projects and papers will be examined.

3. The basic tools, frameworks and languages will be learned, investigated and specified

according to the project properties and criterion.

4. These information will be used in a simple game to learn the basics and start implementing

code.

WP2 - (Supervised learning implementations on selected game types.)

In this work package, the following functionalities / features / work items will be implemented

1. According to the needs, games which have different AIs, properties and different

environments will be selected for prototyping.

2. Data sets for training the agent will be generated and if necessary, automations that generate

specific types of outputs will be coded.

3. Supervised learning algorithms starting from neural networks will be implemented

according to the moves in selected games and first prototypes will be created in this stage.

Algorithms in these prototypes will be evaluated and successful ones will be used in further

iterations of the project.

WP3 - (First comprehensive game agent for a specified game will be implemented with supervised

learning algorithms for mid-term demo.)

In this work package, the following functionalities / features / work items will be implemented

1. A comprehensive game will be selected and a good interface or API to get and send data will

be implemented.

2. A proper deep neural network with helper algorithms like feature selection will be

implemented and trained with the big data set which will be retrieved from the game

beforehand.

3. The product will be prepared to be ready to be presented.

WP4 - (Research on Reinforcement learning, Deep Reinforcement learning and implementations.)

In this work package, the following functionalities / features / work items will be implemented

1. To improve the performance and optimization, different reinforcement and deep

reinforcement learning algorithms will be examined and implemented on sample data sets.

2. Games that have already been investigated environments will be chosen as prototypes and

algorithms will be tried in different games and environments.

Ceng 491 - Project Kickoff Document Template Page 2