AI Course - Week3

# DAY15-17: Hands-on construction of deep learning neural network

# 1. Specific requirements

1. Learn to build a convolutional neural network CNN's composition, convolution layer, pooling layer, fully connected layer, and softmax layer

   • task decomposition：
     • Read the relevant chapters of "Deep Learning" to understand the basic principles of CNN.
     • Implement a simple CNN model using TensorFlow or PyTorch.

2. Learn to build recurrent neural network RNN, long and short-term neural network

   • task decomposition：
     • Read the relevant chapters of "Deep Learning" to understand the basic principles of RNN and LSTM.
     • Implement a simple RNN and LSTM model using TensorFlow or PyTorch.

3. Learn to build a generative adversarial network GAN

   • task decomposition：
     • Read the relevant chapters of "Deep Learning" to understand the basic principles of GAN.
     • Implement a simple GAN model using TensorFlow or PyTorch.

4. Learn to build graph neural network GNN and the definition of convolution on the graph

   • task decomposition：
     • readgraph neural networkopen in new windowThesis, understand the basic principles of GNN.
     • Implement a simple GNN model using TensorFlow or PyTorch.

5. Learn to build Transformer, understand attention mechanism, mask tensor, encoder and decoder

   • task decomposition：
     • watchTransformer《Attention is all your need》open in new windowVideo to understand the basic principles of Transformer.
     • Implement a simple Transformer model using TensorFlow or PyTorch.

6. Learning to build reinforcement learning

   • task decomposition：
     • Read the relevant chapters of "Deep Reinforcement Learning Fundamentals, Research and Applications" to understand the basic principles of reinforcement learning.
     • Implement a simple reinforcement learning model using TensorFlow or PyTorch.

# 2. Display of learning results

Select 2-3 neural network models to practice on public data sets

• task decomposition：
  • Choose the appropriate model for practice.
  • inKaggleopen in new window、UC Irvine Machine Learningopen in new windowandGoogle Datasetopen in new windowSelect a data set.
  • Write experimental reports, including model architecture, training process, result analysis, etc.

# 3. Reference materials

• Book recommendations：

  1. Deep Learning《Deep Learning》 Yoshua Bengio & Ian GoodFellow
  2. Shanghai Jiao Tong University "Hands-on Machine Learning"
  3. graph neural networkopen in new window
  4. Transformer《Attention is all your need》open in new window
  5. Reinforcement Learning《Deep Reinforcement Learning Fundamentals, Research and Applications》
  6. Kaggleopen in new window
  7. Hugging faceopen in new window
  8. Introduction to various neural networksopen in new window
  9. Learn the neural network built by TensorFlowopen in new window
  10. Getting started with machine learning at Microsoftopen in new window
  11. Aurélien Géron 《Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow: Concepts, Tools, and Techniques to Build Intelligent Systems, 3rd Edition》

# DAY18-21: AI EDA training

# 1. Specific requirements

1. target: Able to run through the training process and output the training model.
2. Reference steps：

• Collect and preprocess data.
• Design neural network model architecture.
• Configure training parameters and optimizer.
• Train the model and evaluate it.
• Adjust model parameters to optimize performance.

# 2. Practical training topics

1. Based on the data generated during the layout phase, a network is built to predict delay based on wire length.
   • The topic refers to iEDA practice:M1-Layout line length prediction delay
2. Use the delay calculation training data set to build a network that predicts timing path delays.
   • The topic refers to iEDA practice:M2-timing correction
3. Use the capacitance extraction training data set to build a neural network for calculating 3D capacitance.
   • The topic refers to iEDA practice:M3-3D capacitance extraction

# 3. Display of learning results

1. Output the training model and write a PPT to introduce the design ideas
2. You can choose one AI EDA task

# 4. Reference materials

1. OSCC dataset warehouse iBMopen in new window
2. Related papers

• 《PCT-Cap: Point Cloud Transformer for Accurate 3D Capacitance Extraction》
• 《Accurate_Timing_Path_Delay_Learning_Using_Feature_Enhancer_with_Effective_Capacitance》