D12-1038 shows the overall procedure of the iterative training method . The loop of lines 6-13
D12-1038 Predict-Self Reestimation We adopt the iterative training strategy to the baseline annotation
D12-1038 source classifier . We propose an iterative training procedure to gradually improve
D12-1038 little worse than that brought by iterative training . Figure 3 shows the performance
D11-1128 variant of SOUNDEX along with iterative training was proposed by Darwish ( 2010
D15-1039 of density , together with an iterative training procedure that makes use of these
D12-1038 reestimation brings improvement to the iterative training at each iteration . The maximum
D12-1038 two optimization strategies , iterative training and predict-self reestimation
D15-1105 source corpus . They observed iterative training to improve training-set perplexity
D12-1038 corpora of the next iteration . The iterative training terminates when the performance
D08-1092 The full training setup used the iterative training procedure on all 2298 training
D12-1038 two optimization strategies , iterative training and predict-self reestimation
D13-1016 from the source domain during an iterative training process . Representation learning
D12-1038 human-annotated knowledge . The iterative training procedure proposed in this work
D13-1062 two optimization strategies , iterative training and predict-self re-estimation
D12-1038 optimization strategies in details . 3.2 Iterative Training for Annotation Transformation
D08-1039 likelihood es - timate , we resort to iterative training via the EM algorithm ( Dempster
A00-1014 , which were determined by an iterative training procedure using a corpus of transcribed
D12-1038 two optimization strategies , iterative training and predict-self reesti - mation
D12-1038 and predict-self reestimation . Iterative training takes a global view , it conducts
hide detail