The National Big Data and Computational Intelligence Challenge is an annual event organized by the Big Data and Decision-Making Laboratory of the School of Systems Engineering, National University of Defense Technology . Team, promote technological innovation in the field of big data and demand-oriented results generation, and promote the formation of an iterative evolution innovation model integrating research, construction and application of "intelligent crowdfunding, joint research, and sharing and sharing".
The theme of the 2023 National Big Data and Computational Intelligence Challenge is "Challenge to publish rankings and gather intelligence to tackle key problems". Open a track for cutting-edge technical difficulties, organize innovation and research competitions in the form of "revealing the rankings", and determine the superior team through the combination of online rankings and on-site evaluation.
organizational structure
Organizational Unit
School of Systems Engineering, National University of Defense Technology (Big Data and Decision Lab)
Initiator of the challenge
Huang Hongbin Director of Big Data and Decision-Making Laboratory, National University of Defense Technology
Challenge Organizing Committee
Tang Jun Liu Lihua Wu Jibing Ge Bin
Zhao Xiang Li Xuan Wang Mao Xiao Kaiming
Chen Haiwen Zeng Weixin
competition platform
DataFountain
Introduction
Problem 1: Controlled Text Generation under Data-To-Text Hard Constraints
Competition link: https://www.datafountain.cn/competitions/633
Competition task: Design two topics of text generation technology for given keywords and text generation technology for given tables, with increasing difficulty. The preliminary round requires a set of keywords in a given order to generate a piece of text containing all keywords, and the generated text needs to be domain-relevant and expressive. The rematch requires that, given a table, generate a piece of text containing the key information of the table. The generated text must be faithful to the table, and meet the requirements of correct grammar, concise and clear expression, and natural and coherent semantics.
Question 2: Multi-granularity timing knowledge graph question answering
Competition link: https://www.datafountain.cn/competitions/634
Competition task: A time series knowledge map is given in the form of a quaternion, and its format is [head entity relationship tail entity time]. For each given natural language question, contestants need to reason and answer based on the information in the time series knowledge graph, which involves time information of various granularities and various types of time constraints.
Problem 3: Fine-grained dense ship target detection task based on high-resolution remote sensing visible light data
Competition link: https://www.datafountain.cn/competitions/635
Competition task: The fine-grained dense ship target detection task based on high-resolution remote sensing visible light data requires the positioning of the ship target and the type identification of the ship. In the remote sensing scene, the gap between fine-grained ships is extremely small, the ships are densely distributed, and the scale of ships varies greatly. How to use detection methods to achieve high-precision fine-grained ship recognition is the research difficulty of this competition.
Problem 4: Track real-time correlation and track fusion tasks based on sensor signals
Competition link: https://www.datafountain.cn/competitions/636
Competition task: The competition provides real-time data from multiple multi-source sensors (2D radar and ESM sensors), and requires participants to design appropriate models to identify different ship targets, identify special targets, and obtain ship latitude and longitude information in real time, and The trajectories of the same ship are merged. At the same time, it is necessary to ensure the accuracy and generalization ability of the model.
Problem 5: Refined target detection based on sub-meter images
Competition link: https://www.datafountain.cn/competitions/637
Competition task: This task belongs to the problem of refined detection and recognition in the field of target detection. Different from ordinary image detection and recognition tasks, the difference between classes in refined detection and recognition tasks is smaller. Subclass. This task has higher requirements for target detection and recognition, and is more difficult, and has a wide range of application values in real scenarios.
Question 6: Joint extraction of domain multi-event information
Competition link: https://www.datafountain.cn/competitions/638
Competition task: This competition builds a field news chapter-level event extraction data set (FNDEE) and releases the field multi-event information joint extraction challenge based on this data set. It is required to extract multiple events as independently, completely and accurately as possible based on chapter-level texts. , including the event's trigger word, event type, argument, and argument role. In addition, in order to better explore the relationship between the extracted multiple events, this competition defines interweaving arguments and encourages attention to the extraction of interweaving arguments. Interweaving arguments are arguments belonging to multiple different events in the same text. meta, which requires a complete extraction of interwoven arguments that belong to different events and play multiple argument roles.
Problem 7: Named Entity Recognition for Low-Resource and Incremental Types
Competition link: https://www.datafountain.cn/competitions/639
Competition task: the preliminary task is low-resource named entity recognition, that is, given a low-resource training set, in which each entity type only involves about 50 sample cases, participants need to train a named entity recognition model so that it can be used in It achieves better performance on a larger test set. The rematch task is continuous named entity recognition. Participants need to design a systematic model that enables them to continuously learn a sequence of entity recognition tasks, where each task has an independent data set, and each task involves only one entity type.
Question 8: Dataset Privacy Intersection Technology
Competition link: https://www.datafountain.cn/competitions/640
Competition task: This competition requires participants to realize the application function of privacy-seeking technology. The sponsor provides 4 data samples, and the participants calculate the intersection of the two datasets through the privacy intersection algorithm. During the entire calculation process, the original data of the data samples cannot be exposed, and the calculation results are only known to the receiver.
overall schedule
The competition schedule is four months in total, adopting the "three-level competition system" of preliminary competition, semi-finals and finals. The specific schedule is as follows:
preliminary stage
-
On May 5th , the competition questions will be released, and contestants can log on to the official website of the competition to register;
-
On May 12 , the functions of data download and work submission will be opened successively for each competition topic, and the first round of evaluation will be carried out;
-
June 25th (12:00), deadline for team registration and modification of team information;
-
On June 28 (24:00), the deadline for the submission of works in the preliminary competition, the selection of the top 30 teams for each problem (the number of each problem is different) will be shortlisted for the semi-finals.
Quarter-finals
-
From June 30th to July 13th , the teams shortlisted for the semi-finals will conduct the second round of evaluation. According to the competition questions, two tracks will be set up online and offline (School of Systems Engineering, National University of Defense Technology);
-
In mid-to-late July , the anti-cheating review was carried out, and the results of the works were reproduced. For each question, 5 teams were selected to enter the finals.
final stage
-
In August , offline expert review was organized, and the finalist teams defended on-site. Announce the winning team in due course and organize awards.
*special reminder:
-
After the finalist team is determined, the finalist team must designate a representative to participate in the on-site defense.
-
The time of the competition schedule and the number of shortlisted candidates are adjusted according to the actual situation. The schedule of each competition topic is different. For details, please refer to the rules of each competition topic in the competition topic registration module.
Competition incentives
The units of the participating teams that won the first, second and third prizes will receive targeted project research funding to support the team members to further deepen and improve the results of the competition and overcome related technical problems.
| awards |
quantity (per question) |
Award incentive |
| first prize |
1 team |
Award certificate & the unit of the winning team will receive funding support for targeted project research |
| second prize |
2 teams |
|
| third prize |
2 teams |
Sign up to form a team
Competition official website: https://www.datafountain.cn/special/BDSSF
Participants: Industry-leading teams from various industrial departments, scientific research institutes, universities, and private enterprises across the country.
registration requirements
-
Teams must register in the name of the unit , and the same unit can have multiple teams.
-
Contestants can register for multiple competitions , but they can only register for one team in the same competition.
-
When registering, all participating team members need to verify their real names , and ensure the authenticity of the unit information and basic personal information provided.
-
Student contestants must designate at least one scientific researcher in the relevant field with an intermediate professional title or above who works in the unit as the team instructor .
team requirements
-
All contestants who sign up for the same problem can form a team on the PC side .
-
All contestants should complete the team formation by themselves before the deadline, each team has 1-5 people (including instructors), and repeated team formation is not allowed. Contestants need to submit the work materials of each stage as a team . Once they enter the team, they cannot leave the team.
-
When participants from multiple units jointly form a team, the unit of the captain shall be the first unit of the team .
-
In order to ensure that each participating team has a relatively equal opportunity to submit, each team must meet the requirement that the total number of submissions by team members in the competition ≤ the number of start days * 3 times .
captain responsibility system
As the person in charge of the team, the captain of each team needs to carry out the division of labor and coordination within the team, and assume the responsibility of communication with the competition organizing committee (including but not limited to promotion and shortlisting, team information collection, work review, offline activities, etc.).
*The detailed rules are subject to the content of the official website.
Eight contest questions are waiting for you to challenge, come and show your strengths!
Good luck to everyone!
—End—