This paper proposes a method called BiC (Bias Correction) to address the data imbalance issue in large-scale incremental learning. The method focuses on correcting the bias in the last fully connected layer of a convolutional neural network (CNN), which tends to favor new classes due to data imbalance. The BiC method uses a simple linear model with two parameters to correct this bias. The method is trained in two stages: first, the convolution layers and fully connected layer are trained using knowledge distillation; second, a bias correction layer is learned using a small validation set. The validation set is used to estimate the bias parameters and correct the bias in the fully connected layer. The BiC method is tested on two large datasets: ImageNet (1000 classes) and MS-Celeb1M (10000 classes), and outperforms state-of-the-art algorithms by 11.1% and 13.2% respectively. The method is effective in handling the data imbalance issue and improves classification accuracy on large-scale incremental learning tasks.This paper proposes a method called BiC (Bias Correction) to address the data imbalance issue in large-scale incremental learning. The method focuses on correcting the bias in the last fully connected layer of a convolutional neural network (CNN), which tends to favor new classes due to data imbalance. The BiC method uses a simple linear model with two parameters to correct this bias. The method is trained in two stages: first, the convolution layers and fully connected layer are trained using knowledge distillation; second, a bias correction layer is learned using a small validation set. The validation set is used to estimate the bias parameters and correct the bias in the fully connected layer. The BiC method is tested on two large datasets: ImageNet (1000 classes) and MS-Celeb1M (10000 classes), and outperforms state-of-the-art algorithms by 11.1% and 13.2% respectively. The method is effective in handling the data imbalance issue and improves classification accuracy on large-scale incremental learning tasks.