Theories of Deep Learning

模块目标

State a theorem about representation, training dynamics, or generalisation with its hypotheses, and refuse empirical slogans as substitutes.

课程大纲

1. 01 Approximation by networks 25h

   学习结果 State a universal-approximation claim with architecture hypotheses.

   形成性检查 四选一检查

2. 02 Loss landscapes (taught grain) 25h

   学习结果 Describe a landscape result the synopsis proves and a spurious-critical-point caveat.

   形成性检查 反例

3. 03 Implicit bias and training dynamics 25h

   学习结果 Name an implicit-bias slogan for a named algorithm and model class.

   形成性检查 书面推导

4. 04 Generalisation analyses 25h

   学习结果 Apply a bound only under the measure and complexity hypotheses given.

   形成性检查 书面推导

5. 05 What experiments do not prove 20h

   学习结果 Give an empirical success that the current theorem does not explain.

   形成性检查 四选一检查

6. 06 Synthesis 30h

   学习结果 Write one theorem-plus-hypothesis argument without converting test accuracy into a proof.

   形成性检查 书面推导

学习结果

Quote an approximation or implicit-bias result with function-class and training-model hypotheses, and give a setting where a naive claim fails.

学科技能

Universal approximation at taught grain, NTK/mean-field language as in the synopsis, generalisation bounds that are actually proved there.

可迁移技能

Keep empirical plots out of the theorem statement; name the training model.