AIIDE StarCraft AI Competition - Survey Feel free to answer as many questions as you like, but it would be great if everyone answered everything! Many people are interested in learning as much as possible about the bots that competed! Please feel free to provide external references/links as necessary Bot Name: SAIDA Bot Race: Terran Author Name(s): Changhyeon Bae(Leader), Iljoo Yoon, Daehun Jun, Junseung Lee, Hyunjin Choi, Hyunjae Lee, Yonghyun Jeong, Uk Jo Affiliation(s): Samsung SDS Nationality(s): Korea, South Occupation(s): Software Engineer, AI Research Engineer (These will be listed on the competition website) Bot URL: https://github.com/TeamSAIDA/SAIDA Personal URL: Affiliation URL: https://www.samsungsds.com/ Questions about your bot (please answer as many as you can, especially Q 1-3) Q: What is the overall strategy/strategies of your bot? Why did you choose them? A: SAIDA basically plays the mechanic terran in all games. it starts with a stable defence-first strategy and after mid stage of game, it seeks the best rush timing and win the game with a powerful one-shot attack. We think that the mechanic terran can cope with the most opponent's strategies and has the least weakness. That's why we chose this one. Q: Did you incorporate any of the following AI techniques in your bot? If you did, please be as specific as possible a) Search-Based AI (Path-Finding, A*, MiniMax, MCTS, etc) b) Offline Machine Learning (Supervised or Unsupervised, but not RL) A: We made the CNN & Encorder-Decode that has learned progamers replays to decide when to fight. (Not in submitted version of AIIDE because of some performance issue. But it's result paper is submitted to 2019 AAAI Conference.) c) Offline Reinforcement Learning A: We studied the multi-agent reinforcement learning in micro unit control in several mini-games of Starcraft. (Not in submitted version of AIIDE. But it's result paper is submitted to 2019 AAAI Conference.) d) Online Learning of any kind (Including competition file IO for strategy selection) e) Influence Maps f) Custom Map Analysis g) Hard-coded or rule-based strategy / tactics A: We have implemented an adaptive strategy. SAIDA was implemented to select very precisely the relative strategy and my strategy to win the strategy based on the information gathered through scouting and scanning rather than the pre-established build order or strategy. h) Analysis of bots from previous competitions / hard-coded specific bot counter strategies A: SAIDA has no hard-coded about map or oppenent bot. i) Any techniques not mentioned here A: We use a finite-state machine to control units and buildings. Every units and buildings has a specific state in every situation. And we use several search algorithm to find enemy base or place to build on. Q: How did you become interested in Starcraft AI? A: Starcraft is so famous in Korea that everyone knows it. In the meantime, last year there was an starcraft ai contest in our company. So I and my team members participated in the contest and gained the good results. After the contest, some developers who had a good grade in the contest gathered to make worldwide champion bot. They proposed this idea to the company and fortunately it is accepted. That is the start of 'SAIDA'. Q: How long have you been working on your bot? A: Since the begining of this year. Q: About how many lines of code is your bot? A: Total Valid Lines : 37027 Total Comment Lines : 3820 Total Blank Lines : 8855 Total Lines : 49702 Q: Why did you choose the race of your bot? A: Because terran is much more powerful than other races when it goes with superior unit controls that can be performed by AI. And with similar reasons, we thought the terran has the most possibilities which can be first AI beyond human in starcraft. Q: Did you use any existing code as the basis for your bot? If so, why, and what did you change? A: We started with UAlberata bot & BWEM. It seemed well engineered neither too much nor too little to start. Q: What do you feel are the strongest and weakest parts of your bot's overall performance? A: SAIDA changes its strategy flexibly in response to opponent's strategy like human does when they play it. But it lacks aggresive strategy in eary stage of game. Q: If you competed in previous tournaments, what did you change for this year's entry? A: SAIDA is new entry this year. Q: Have you tested your bot against humans? If so, how did it go? A: Yes we did several times and beated amateures. Even when we tested against progamer, We couldn't win but saw the possibilities that it is not that long way to beat the progamer. Q: Any fun or interesting stories about the development / testing of your bot? Q: Any other projects you're working on that you'd like to advertise? Optional Opinion Questions: Q: What is your opinion on the current state of StarCraft AI? How long do you think before computers can beat humans in a best-of-7 match? A: I think the level of StarCraft AI is about high level amateur (about Battle.net ladder 1800 points). I think it will take at least two years for the AI ??to win in the one of the top seven human players (Battle.net ladder 3000 points or more). Q: What do you feel is the biggest hurdle (technological or otherwise) in improving your bot's AI? A: I think it is a difficult way to respond appropriately to various anomaly strategies of people. I think that it is only possible to learn the response of various progamers playing. We have a plan to study imitation learning to solve this problem. Q: Which bots are the most interesting to you and why? A: CherryPi. We want to know how much progress it has made since Facebook developed and did not confront the new update version this year. AIIDE Specific Question: Q: Do you feel that the current format of iterated round-robin win percentage is a good indicator of bot skill ranking? If not, how would you change it? A: I think the current method is the most fair.