"Wilderness Survival" hit Neuraips, and there are 10,000 yuan bonus. You can take AI with AI

You dare to believe that the serious game of Neurips for machine learning is actually letting you "play games" and the kind of money.

All you have to do is to adjust a team of 8 AI to let them learn to "survive the wilderness" in such a map.

In this process, the other 15 teams of the same scale will compete with your AI for survival resources, so you have to fight, engage in equipment, even run poison, do business ...

If your AI is particularly resistant, you can compete for the "tank award" (manual dog head) in the single prize.

If you don't want to fight and kill, you can also consider taking the "economic route": the trading system is added in the competition system, so that the "small people" will focus on money (buying and selling equipment). Farmer Award.

You say you are a novice? It doesn't matter, you can play a machine first and then rank ~

It can be said that the difference between the games that we often play is not very large -just use your own code to compare.

Of course, don't worry about technical level. As long as you can program, you can try it. There are many ways to win.

As for the bonus, the organizer prepared $ 20,000, and the top sixteen were available.

Want to try? Start up ~

Massive AI fights with the University on the same Taiwan, watching AI to play thirty -six plans

This game is called Neurips 2022-Neural MMO Massive AI team survival challenge. The organizers are Super Portable Technology, Massachusetts Institute of Technology, Tsinghua University Shenzhen International Graduate School, and data science challenge platform AicrowD.

Just as its name, its "tunnel" comes from a platform called Neural MMO.

The platform is an open source environment released by OPENAI in 2019 to study massive intelligence strategies.

Author Joseph Suarez is a doctoral student of MIT, inspired by large multiplayer online role -playing games (MMORPG).

It supports a large number of players to compete and confront in a long -lasting (non -remake) and vast environment, and a desktop -level CPU can run. It is an excellent enhanced learning training field.

Challenge based on Neural MMO is for global AI technology practitioners, researchers and enthusiasts.

As long as you can program it, you can come and play. Of course, it will be better to strengthen the learning algorithm or the optimization of the operation and the inspiration algorithm ~

In terms of specific rules, each contestant needs to control a team of 8 intelligence, starting free confrontation with 15 other teams on the 128x128 map.

Each intelligence can only see things within 15X15.

There are 16 basic elements on the map:

The "water" used for drinking, the "forest" that can get food, the "grassland" that can be trampled normally, the "stone" that blocks the road, and the "magma" that will die;

It can be converted into a stored food resource "fish" and "mushrooms", which can be made into three ammunition "ore", "crystal", "pine tree";

And the "bushes" that can be degraded after being collected by the collection of resources, it will have a certain probability to re -grow back;

and many more.

Each step can be performed many actions: moving, selecting attack objects, selecting attack methods, using a certain equipment in the backpack, buying items in the market, selling it with itself and bidding it freely (within the fixed range).

The launch of attacks include melee attacks, long -range attacks and magic attacks. They all have corresponding weapons and ammunition to enhance force value, as well as defensive armor and other equipment to increase their defensive values.

After the start of the game, each team is randomly born on the edge of the map. The poison circle mechanism will allow everyone to slowly gather in the middle of the map, so that the opponents and high -strength NPCs will be forced to fight positively. And survive longer.

Trading system, poison circle mechanism ... interesting and hard core

Yes, for the excitement of the game, not only the original game duration was limited to 1024 steps. In addition to the survival and attack rules provided by the platform itself, the event also increased the trading system, expanded the equipment category, and designed multi -occupation division of labor. And poison circle mechanism.

in:

Equipment system allows intelligence to obtain items that improve their defense and attack power by attacking the enemy and NPC, including ammunition, weapons, armor, and so on.

The trading system is the biggest highlight of this competition.

In every STEP, the intelligent body can freely buy and sell the equipment in the hand. Compared with the enemy, the teammates have the right to purchase priority (non-delivery). When selling, the number of numbers between 1-100 is priced by themselves.

The money for purchasing equipment can be obtained by constantly killing NPCs or other players.

Because the backpack capacity of each intelligent body is limited, it cannot be discarded after the equipment is full, so how to sell for a price becomes a problem (such as high -end equipment selling it by the enemy, which will threaten itself).

This setting not only provides new ways to interacting for group intelligence, but also gives them higher decision -making complexity.

For the multi -occupational division of labor mechanism, its setting is to win the competition by playing the role.

For example, those who are good at melee attackers have been overcome the close -off capabilities, only buying melee equipment, selling the remote and magical attack equipment obtained;

Smarts who are good at collecting will sell more advanced equipment to their teammates at low prices. Finally, the poison circle mechanism stipulates that the game starts to reduce poisoning when the game is carried out at 240, so as to force the smart body to enter the security area in the middle of the map as much as possible, and the higher -level NPC and other opponents front Battle.

Double track, take a seat according to strength

So what is the specific method?

In order to undertake the enthusiasm of novice players, two tracks were set up intimately: PVE and PVP.

PVE is "Player vs Environment". In the track, the player competed with the built -in AI, a total of ten games, ranking 0.1 points first.

There are two stages in the competition, and the difficulty of built -in AI strategy has gradually increased.

However, its strategy is fixed, so participants can get reference feedback, so as to continuously optimize their strategies according to this feedback, let the effect be visible, and boost confidence.

(Think about it, if the real player PK at the beginning, both parties optimized the strategy at the same time. How can I know if their optimization has grown?)

If you get 0.5 points in Stage 1 in the PVE session, then congratulations, you can participate in the main arena-

PVP, "Player VS Player", is a competition between real contestants, and the bonus is here.

In this track, everything is unknown, and no one knows what strategies they will encounter.

And your way to achieve good results in the PVE may stand out in a random fighting with other players, or it may be flowing flowers in seconds.

The PVP link is also not a finalist. In order to ensure fairness, the competition will use a ladder -like ranking algorithm to roll the ranking.

This is not over yet.

In order to encourage diversification and more interesting strategies, two single prizes have been set up in the competition:

"Gold Farmer Award" and "Tank Award".

That's right, this tank refers to T. At the beginning, it has been mentioned that the intelligences need to learn to jump in the inside and outside of the poison circle, and go crazy to take drugs and return the blood to be the most fleshy one in the game.

The "Gold Farmer Award" only needs to be obsessed with money.

I have to say that the setting of these two awards is very taste.

There are so many about the rules of the competition.

Testing the rules and understanding, but also the depth of decision -making

In order to reduce the cost of the contestants' familiarity with the environment and start training, the organizer deliberately prepared the Starter Kit (Starter Kit).

And a Baseline based on Facebook's TORCHBEAST framework, players can run the Neural MMO environment with just a few lines of code to start training.

In addition, they have developed a lightweight Web Viewer, which can use a very simplified process to allow players to observe the performance of all smart parties in each game like human observer.

Then, the text mentioned in the text that it is simple to get started, and the smart body can survive when drinking water and eating. It has become the first reason to recommend participation by Dr. Chen Jiaxin, the person in charge of the competition.

(If you are not a game, you can see it as a task with specific rules, which can be used to enhance learning.)

The second recommendation reason: You say it is simple to get started, but there is also a challenge of "100 million points".

For example, the Neural MMO Challenge has tested the players' two abilities: understanding the game itself, and the ability to use decision -making algorithms.

In view of the 16 teams in each environment, a total of 128 intelligences are learning at the same time. To win, the smartrs not only need to consider what rewards they expect to get, how to cooperate with teammates, but also consider what strategies they may adopt.

In addition, each game is a multi -tasking mode. The decision -making chain is long, and the final winning award is very sparse.

Behind the layers of settings, the "choice" faced by each smart body has higher decision -making complexity and fit the complex decision -making environment of reality.

For example, they may be solved at the same time:

How to find more time -saving channels? How to keep attacking stronger NPCs to get more powerful equipment? At the beginning of the game, did you choose to accumulate the killing points or get equipment? When the game starts to reducing poisoning 240, is it important to protect the life or the enemy and the NPC who continue to chase the outer circle and make a click -to -kill and equipment? When the backpack is full, is it selling equipment to store ammunition and potion, or does it sell high -level equipment to the enemy? Native

How to make a specific decision depends on how you play it ~

In the end, Dr. Chen also said that the new trading system of this competition was also very interesting and highly attractive. The interaction between AI changed from basic robbing resources and mutual attacks to more enrichment.

Under this setting, the strategy of learning to buy and sell between smart parties will slowly evolve the economic behavior of AI, and this settings are rare in the previous research environment.

So everyone is playing for the first time. Without the pre -cars, it is worth looking forward to what strategies can emerge.

In addition to the above three points, we will focus on the multi -intelligent decision -making technology behind the game.

In fact, the results of this research can come in handy in autonomous driving, drone cluster confrontation, smart cities, and industrial fields.

To go far away, this technology also has a booster effect on the development of universal artificial intelligence.So, such an interesting and valuable game, you really don't try it?

After all, the organizer provides rich material rewards.

In addition to material rewards, the winners will also be invited to write the Competition Report as a co -author, published on domestic and foreign technical blogs, and may be published as Neurips conferences.

Participate in DDL on October 30th

Finally, the schedule is arranged:

August 18, 2022-Phase 1 competition starts, release entry kits, open submission system August 31, 2022-PVP September 22, 2022-Phase 2, Phase 2Establishing the deadline with the team October 30, 2022-The final submission deadline (focus !!!), October 31, 2022 to November 15, 2022result

- END -