- hw1:
- incorrect linear backward for gradb: should sum across batches
- good report format and analysis
- 9.5 / 10
- hw2:
- when pad != 0: [PASS] conv2d_forward: all correct [PASS] conv2d_backward: all correct [ERROR] avgpool2d_forward: output value is not correct [PASS] avgpool2d_backward: all correct
- good report
- 10.5 / 10
- hw3:
- running mean and var are only used for normalization when testing, not in training
- resnet +1
- 10 / 10
- hw4: 10 / 10
- hw5: 10 / 10
luoyt14 / deeplearninghw Goto Github PK
View Code? Open in Web Editor NEW深度学习课程作业