在機器學習中的主動式學習旨在找到最具信息性的實例，以充分降低模型的不穩定性。在本文的研究架構中為了降低重複訓練目前已知標籤所造成的計算損失，本研究放棄每次迭代中的僅僅選擇一個未標籤實例手動標記，改採用批量模式作為尋找下批未標籤實例手動標記，即在每次迭代中選擇前k個最具信息性的實例手動標記。

本研究是關於在圖形結構模型上應用批量模式的主動式學習，研究中自節點集選擇了一些具有高影響力的節點作為參考節點集，在此我們考慮易辛模型(Ising model)圖形結構間的節點關聯，所以基於節點信息和從參考節點集中當前模型導出的費雪信息矩陣（Fisher Information Matrix）提出了費雪信息分數做為量測實例中信息量的標準，在每次迭代中用以選擇最能充分降低模型不確定性的下一批實例。
在本文的末節，我們將此研究方法應用於學習不同圖形結構的易辛模型上，以驗證本研究方法對於降低模型不穩定性的效率。

The active learning aims to find the most informative instances to reduce the uncertainty of the classification model sufficiently. To avoid the computational cost of retraining current labeled instances, instead of selecting a single unlabeled instance in each iteration, the research adopts the batch mode to seek instances, that is, the algorithm selects the top-k informative instances to manual labeled in each iteration.

The study is about applying batch mode active learning on the graphical structure model. The research proposes the criteria based on the node information and the Fisher information matrix (FIM) derived from the current model to select the next batch instances which reduce the uncertainty of the classification model the most sufficiently. In the proposed algorithm, we select some of the high influential nodes in the vertex set and treat them as the reference nodes set. Then for each iteration, the instances with maximal FIM score for the reference nodes are selected and are included into the training set. To show the performance of the proposed algorithm, experimental results on the different graphical Ising models are studied.

Hoi, S., Jin, R., Zhu, J., and Lyu, M. (2006). Batch mode active learning and its application to medical image classiflcation. volume 2006, pages 417–424.
Jalali, A., Johnson, C., and Ravikumar, P. (2011). On learning discrete graphical models using greedy methods.
Lauritzen, S. L. (1996). Graphical Models. Oxford University Press.
Ravikumar, P., Wainwright, M. J., and Lafferty, J. D. (2010). Highdimensional ising model selection using ℓ 1 regularized logistic regression. The Annals of Statistics, 38(3):1287–1319.
Xu, H., Zhao, P., Sheng, V. S., Liu, G., Zhao, L., Wu, J., and Cui, Z. (2015). Batch mode active learning for networked data with optimal subset selection. In Dong, X. L., Yu, X., Li, J., and Sun, Y., editors, WebAge Information Management, pages 96–108, Cham. Springer International Publishing.