GPT2

5 atoms · 跨 2 天 · 首见 2026-05-05 · 最近 2026-06-07

三色: 🟦 fact 1 · 🟥 take 4 · stance ▲0/▼3/◆2

来源: X 5

时态: aging:4 · fresh:1

标签: 好思考:4 · 好信源:1

叙事 (narrative) (4)

• 2026-05-05 [老化中] · 过拟合训练数据后训练准确率陷入陷阱
  even then after some period of time it overfits its training data, and the training accuracy traps
  tags: 好思考 · → daily
• 2026-05-05 [老化中] · 权重矩阵中 Correlation Traps 的外特征值越大，对模型泛化影响越大
  Such traps screw up the correlated model, and keep it from generalizing well. and the larger the outlier, the bigger an effect they have.
  tags: 好思考 · → daily
• 2026-05-05 [老化中] · Correlation Traps 具有非零方差 O(1)，导致测试误差失去集中性，违反泛化界限
  the traps have non-vanishing variance of order O(1), and cause the test error to lose concentration, therefore violating any generalization bound you can think of.
  tags: 好思考 · → daily
• 2026-05-05 [老化中] · 检测过拟合不需要训练或测试数据，只需权重矩阵
  And all of this can be done without needing access to the training or test data. All you need are the weights.
  tags: 好思考 · → daily

事实 (fact) (1)

• 2026-06-07 · 代码和权重发布状态
  GPT2 did not release source code, just weights.
  tags: 好信源 · → daily

时间轴 (近 20)

• 2026-06-07 · fact · 代码和权重发布状态 · → daily
• 2026-05-05 · narrative · 过拟合训练数据后训练准确率陷入陷阱 · → daily
• 2026-05-05 · narrative · 权重矩阵中 Correlation Traps 的外特征值越大，对模型泛化影响越大 · → daily
• 2026-05-05 · narrative · Correlation Traps 具有非零方差 O(1)，导致测试误差失去集中性，违反泛化界限 · → daily
• 2026-05-05 · narrative · 检测过拟合不需要训练或测试数据，只需权重矩阵 · → daily

← 实体目录 · 系统日志