[**PRACE AUTUMN SCHOOL 2021**](https://events.prace-ri.eu/event/1188/)
[**INTRODUCTION TO DEEP LEARNING**](https://hackmd.io/@pdl/Skn7I48MY)

# NOTEBOOKS EXERCISES

## Exercise 2 (02-tf2-mnist-mlp)

### Using MNIST data

| Submitter | Model description | Test accuracy |
| --------- | ----------------- | ------------- |
| Markus | original mlp_model | 95.54% |
| Markus | mlp_model, epochs=20 | 96.01% |
| Sreeram | mlp_model, epochs=20 with 3 dense layer with units 100,50 and 20. dropout = 0.2 | 97.41% |
| Neha | mlp_model, epochs=10, Dense layer1 =100, layer2 = 50 and dropout=0.2 | 98.02%|
| Markus | example answer two-layer model | 97.30% |
|A Submitter|0.21 Dropout rate|97.17%|
|Rajendra| 0.1 Dropout rate |97.17%|
|Charles| layer1=160 layer2=120 dropout=0.5 |98.25%|
|Murali | 3 dense layers 150, 100, 50, dropout (0.2,0.2,0.1) | 98.30%|

## Exercise 3 (03-tf2-mnist-cnn)

| Submitter | Model description | Test accuracy |
| --------- | ----------------- | ------------- |
| Markus | original cnn_model | 98.36% |
| Markus | original cnn_model, 10 epochs |98.74% |
| Rajendra|first layer 3, 3 second 2, 3 0.01 Dropout rate, Epoc 5 |0.9919|
|Pekka|2 layers, Adam, 30 epochs, batch 128|0.9933|
|Rajendra| first layer 3, 3 second 2, 3 0.01 Dropout rate, Epoc 11 |0.9964|
| ~ilja | 1. C2D 32, (7x7); 2. C2D 62 (3x3); MaxPool 2x2 ; Drop 0.2; Flat; Dense 256; Dropout 0.5; 20 epochs | 99.39% |
| Neha | 2 dense layers= 128,100,25 epochs | 99.08% |
| Sreeram | 2 dense layers= 128,64,25 epochs | 98.72% |
| Murali | 2 conv2d layers, 10 epochs, 32 batch size | 99.30% |
| Markus | example answer better_cnn_model; batch_size 128 | 98.95% |
| Febrian | Task 1 parameters, batch_size 32 | 99.23% |
|Dr. Leipa| batch size 24| 99.14%|

https://paperswithcode.com/sota/image-classification-on-mnist : 99.87%

## Exercise 4 (04-tf2-imdb-rnn)

| Submitter | Model description | Test accuracy |
| --------- | ------------------------------------ | ------------- |
| Mats | original RNN (single LSTM), 5 epochs | 82.78% |
|Rajendra| first layer 32, second 16, 0.01 Dropout rate, Epoc 5 |83.06 %|
| Mats | single CNN layer (optional/tf2-imdb-cnn) | 88.28% |
| Neha | single RNN layer | 80.05% |
| Sreeram | 2 RNN layer | 78.40% |
| Markus | example answer rnn_model_with_two_lstm_layers | 83.62% |
| Mats | original RNN (two bi-directional LSTMs), 5 epochs | 83.32% |
| Febrian | original RNN (one bi-directional LSTMs), 15 epochs, 32 batch_size | 81.45% |