Crowd-sourcing competition platform framework system and method based on crowd intelligence

Abstract

The invention relates to a crowd-sourcing competition platform framework system and a crowd-sourcing competition platform framework method based on crowd intelligence. The method comprises the following steps: arranging a data labeling platform access module, a crowd-sourcing community access module and an interactive training space module at a server, and presenting the modules to a user in a Webpage form for interaction. According to the invention, the defect that the existing machine learning competition system does not fully utilize the swarm intelligence is solved. The data utilization efficiency and the model training efficiency of an existing machine learning competition system are improved. The advantage of group intelligence can be fully utilized. The competition holding cost ofcompetition organizers is reduced to a great extent. Meanwhile, the competition participation cost of competition participants is saved.

Description

technical field [0001] The invention relates to a swarm intelligence-based swarm intelligence competition platform framework system and method, and belongs to the technical fields of computer network transmission and machine learning. [0002] technical background [0003] At present, the framework system of the traditional machine learning competition platform has a single function and a simple process, and cannot fully combine and release artificial intelligence. Competition tasks and data sets start to analyze data, design algorithms, and finally use private computing resources to complete model training and submit the trained model or data results to the competition platform. The competition publisher uses the test set and the scorer formulated according to the expected scoring standards Give evaluation feedback to the participants' submitted answers, and the competition participants will optimize the algorithm according to the feedback results, and then submit it again f...

AI Technical Summary

Problems solved by technology

[0003] At present, the framework system of the traditional machine learning competition platform has a single function and a simple process, and cannot fully combine and release artificial intelligence. Competition tasks and data sets start to analyze data, design algorithms, and finally use private computing resources to complete model training and submit the trained model or data results to the competition platform. The competition publisher uses the test set and the scorer formulated according to the expected scoring standards Give evaluation feedback to the submitted answers of the participants, and the participants of the competition will optimize the algorithm according to the feedback results, and then submit and run it again. The whole process will be completed until the end of the competition. The organizer will announce the ranking of the players according to the scoring standards

[0004] The holding process of this traditional group intelligence competition does not make full use of group intelligence resources, and does not reflect the concept of interactive machine learning. It pays more attention to the submission results of individuals, and the so-called "optimal" results obtained in the end The model is only the best performing model among the many models submitted by the participants, not the best model that can be trained based on the data set

At the same time, as far as the current situation of the group intelligence platform is concerned, the data marked by the competitors is limited after all, which is only the tip of the iceberg compared with the massive unlabeled and unprocessed raw data, although more marked data is important to the competition participants. It means that there are more choices and even a better model can be trained, but the economic cost and time cost have become the biggest problem in the data labeling of the crowd intelligence competition platform

Images