HW2
Intelligent System Design for Competitive Arenas
In this homework, you will design an intelligent system to compete against built-in agents across multiple stages.
Your Arena
- Othello Arena (for HW2): [Link]
What is changed?
In the Othello arena, you will build an intelligent system to spawn a strategy (skill-program) for given stage. The major difference is that your intelligent system could face multiple stages where the board structure is slightly modified.
- Public Stages:
- Some stages are publicly available, and you (along with your intelligent system) will have essentially unlimited time to play against the provided agents. You may also practice using self-play to refine your strategy.
- Some stages are publicly available, and you (along with your intelligent system) will have essentially unlimited time to play against the provided agents. You may also practice using self-play to refine your strategy.
- Private Stages:
- There are also private stages that only staff can access. Although the overall game structure remains largely unchanged, these stages have subtle tweaks that may affect performance.
- As a problem solver, you may already possess full information about the game’s general structure. However, your intelligent system will be given only 60 seconds to adapt to a new stage before competing. In this period, it should “wisely use” the available time to generate a tailored strategy.
- Note: It is acceptable to submit just the strategy without a fully built intelligent system. Keep in mind that even an optimized fixed strategy might not achieve 100% success on unknown stages.
- Evaluation:
- As explained before, a new “Upload Intelligent System (with Interaction)” button is available. This will let your intelligent system to spend 60 seconds to play on given stages. Use this feature to prepare for the private stages.
-
Similar to HW1, your strategy will compete against built-in strategies and those submitted by your colleagues on the leaderboard.
- Time Limit for Strategies:
- Each strategy (whether built-in, manually created, or generated by your intelligent system) must complete each game within a total of 5 seconds of computation time. If a strategy exceeds the 5-second limit at any point during the game, it will automatically lose by time forfeit, regardless of the current board position.
- Plan your algorithms efficiently, as complex calculations that take too long will result in automatic defeat.
- Each strategy (whether built-in, manually created, or generated by your intelligent system) must complete each game within a total of 5 seconds of computation time. If a strategy exceeds the 5-second limit at any point during the game, it will automatically lose by time forfeit, regardless of the current board position.
Submission Guidelines
- Submit your work at Gradescope by May 18, 11:59pm (KST)
- Your submission should be:
- A paper (PDF up to 8 page) using the overleaf template (Click) describing:
- Detailed description of the intelligent system you implemented (How it interacts with the stage, then come up with the strategy)
- Efforts to generalize to unseen situation
- Challenges faced and future improvements
- How your intelligent system spawned strategy performs against the built-in strategies on public stages
- Downloadable link to your intelligent system (javascript file) explained below
- Justify and prove that your work and its implementation is a way towards general intelligence
- (Optional) Anything to make your paper more comprehensive, e.g., Hyperparameter tuning, perhaps relaxing the time limit of 5 seconds or 60 seconds
- A JavaScript (.js) file containing your intelligent system
- The submitted function should be verified by you, to work in the web arena.
- File name should follow this format: 20261135-HW2.js (Your ID-HW#.js)
- Provide a downloadable link inside the PDF (Gradescope does not allow the .zip submission!)
- A paper (PDF up to 8 page) using the overleaf template (Click) describing:
HW2 will take 30% of the overall grade, and the evaluation criteria of the PDF and the algorithm will be determined during grading.