Technical note: systematic bias in stochastic learning
研究了随机学习模型中参数估计的系统性偏差，发现基于早期数据估计的模型平均而言对未来成本降低的预测过于乐观。
The learning curve is a fundamental model used by engineers in cost estimating. In industry, it is typical to use the deterministic model for projecting cost, which is also suggested in standard textbooks. However, the parameters for the model are obtained from actual data, which usually come from a stochastic process. In this technical note, we investigate a particular phenomenon of the stochastic learning model that indicates that a bias may exist in the parameter estimates simply due to random behaviour in learning. The findings suggest that, on average, projections of cost from a model whose parameters are estimated from early data points are, on average, optimistic about the future cost reduction.