Active learning algorithm based on historical evaluation result
A historical evaluation and active learning technology, applied in the field of active learning algorithms based on historical evaluation results, can solve the problems of no data, no information mining, ignoring knowledge and information, etc., to achieve the effect of improving effect and efficiency, and easy to implement
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0037] Such as figure 2 As shown, it can be seen that these samples have the same evaluation results in the last iteration, but have completely different performances in the historical evaluation process, and the change curve is similar to figure 2 (a) and figure 2 The sample in (c) is evaluated many times as containing a large amount of information during the iterative training of the model, and the change curve is similar to figure 2 The sample in (b) only gets a higher evaluation score in the current round of evaluation. For the training of the model, it is obvious that the former has more information and value than the latter.
[0038] Based on the above description, the present invention provides an active learning algorithm based on historical evaluation results, comprising the following steps:
[0039] 1) Use the labeled sample set to initialize the task model.
[0040] 2) Use the current task model to evaluate samples, and select some unlabeled samples in the un...
Embodiment 2
[0053] Uncertainty is one of the most commonly used evaluation indicators in active learning. Uncertainty-based sampling methods tend to select samples that the model cannot accurately predict or have low confidence in the predicted results. At the t-th iteration, each unlabeled sample has a corresponding historical evaluation sequence, and each historical evaluation result is calculated based on a specific uncertainty sampling method S. From the perspective of sequence volatility, it can be found that the main figure 2 (a) and figure 2 (d) Two types of typical sequences, in which the change curve is the same as figure 2 (a) The performance of similar samples is relatively stable, while the change curve is similar to figure 2 (d) Similar samples have greater volatility. With the iterative update of the model, samples with low uncertainty and relatively stable changes in the historical evaluation sequence are easy for the model to learn and judge. However, samples with ...
Embodiment 3
[0063] The first and second embodiments above are two heuristic active learning algorithms that use the historical evaluation results of unlabeled samples in the unlabeled sample set, and use the information in the historical evaluation sequence to select unlabeled samples in the unlabeled sample set, but may still There is a lot of information not covered. This embodiment further proposes an active learning algorithm based on learning from historical evaluation sequences, and regards the process of active learning to select samples as a sorting problem, using historical evaluation sequences as training data to train a sorting model in each iteration All unlabeled samples are sorted, and the top unlabeled samples are selected. Therefore, the active learning algorithm based on learning from the historical evaluation sequence of this embodiment is the same as that of Embodiment 1 and Embodiment 2 except for training the ranking model. Basically the same, including the following ...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More