Multi-task synthetic dataset

A synthetic dataset collection designed for evaluating multi-task learning and transfer learning algorithms under both regression and binary classification settings. It consists of 100 independently generated batches, each initialized with distinct random seeds to promote diversity across tasks. Every batch contains 10 tasks (including two designated outliers), with 300 training and 1,000 test instances per task distributed across five input features. The dataset ensures balanced class representation and controlled task variation through a weighting parameter of w = 0.9.本合成数据集集合专为评估回归与二元分类场景下的多任务学习与迁移学习算法而构建。该集合包含100个独立生成的批次，每个批次均采用不同的随机种子进行初始化，以保障任务间的多样性。每个批次包含10项任务（含2项指定异常任务），每项任务配备300个训练样本与1000个测试样本，且基于5个输入特征生成。该数据集通过设置权重参数w=0.9，实现了均衡的类别分布与可控的任务差异度。