Università degli Studi di Pavia

Facoltà di Ingegneria


Deep Learning

A.A. 2022-2023

Second Semester

Fri: 11:00 a.m. - 1:00 p.m., Aula C6

Fri: 2:00 p.m. - 4:00 p.m., Aula D8

Lectures & Suggested Readings:

  • Reports of errors in the resources below are always welcome
    1. 2023.03.10 (theory)

      Introduction [pdf]
      AI spring? Artificial Intelligence, Machine Learning, Deep Learning: facts, myths and a few reflections.

    2. 2023.03.10 (theory)

      Fundamentals: Artificial Neural Networks [pdf]
      Foundations of machine learning: dataset, representation, evaluation, optimization. Feed-forward neural networks as universal approximators.

    3. 2023.03.17 (theory)

      Flow Graphs and Automatic Differentiation [pdf]
      Tensorial representation, flow graphs. Automatic differentiation: primal graph, adjoint graph.

    4. 2023.03.24 (theory)

      Deep Networks [pdf]
      Deeper networks: potential advantages and new challenges. Tensorial layerwise representation. Softmax and cross-entropy.

      Shannon Entropy (Wikipedia)

      Cross Entropy (Wikipedia)

    5. 2023.03.24 (theory)

      Learning as Optimization [pdf]
      Vanishing and exploding gradients. First and second order optimization, approximations, optimizers. Further tricks.

      Aside 1: Exponential Moving Average [pdf]

      Aside 2: Predictors [pdf]
      From in-sample optimization to out-of-sample generalization.

    6. 2023.03.31 (theory)

      Convolutional Networks [pdf]
      Convolutional filter, filter banks, feature maps, pooling, layerwise gradients.

    7. 2023.04.14 (theory)

      Deep Convolutional Neural Networks and Beyond [pdf]
      Some insight into what happens in convolution layers. DCNN architectures. Transfer learning. Working in reverse: image generation. Generative adversarial networks. Autoencoders and segmentation. Object detection.

      J Yosinski, J Clune, Y Bengio, H Lipson, "How transferable are features in deep neural networks?" in Advances in Neural Information Processing Systems (NIPS 2014) [link]

    8. 2023.04.21 (theory)

      Aside 3: Tensor Broadcasting [pdf]

      Aside 4: Differentiating Algorithms? [pdf]
      Graph-based vs. tape-based automatic differentiation. The engineering solutions in TensorFlow and PyTorch.

      A Paszke et al. "Automatic differentiation in PyTorch" in Advances in Neural Information Processing Systems (NIPS 2017) [link]

    9. 2023.05.05 (theory)

      Aside 6: Word Embedding [pdf]
      Skip-grams, probability distributions of context and center words, training and results, continuous bag of words (CBOW) model.

      Attention and Transformers [pdf]
      Attention as a kernel, attention maps, queries, key and values, attention-based encoder and decoder, transformer architecture, translator.

      A Vaswani, N Shazeer, N Parmar, J Uszkoreit, L Jones, A N Gomez, L Kaiser, I Polosukhin, "Attention Is All You Need" in Advances in Neural Information Processing Systems (NIPS 2017) [link]

    10. 2023.05.14 (theory)

      Reinforcement Learning [pdf]
      A short recap about RL foundations, Markov decision process, state value function, policy, optimality, action value function, Q-learning.

    11. 2023.05.26 (theory)

      Deep Reinforcement Learning [pdf]
      Integrating DNNs into the RL paradigm, DQN algorithm, policy gradient, Actor-Critic methods, NAF algorithm.

    12. 2023.06.09 (theory)

      Monte Carlo Tree Search [pdf]
      Game trees, Monte Carlo strategy, Monte Carlo Tree Search (MCTS), Upper Confidence Bounds applied to Trees (UCT).

    13. 2023.06.16 (theory)

      Alpha Zero [pdf]
      MCTS + DNN, network architecture, replacing MCTS rollout with estimation, network training, AlphaZero in continuous spaces (hints).

      D J Mankowitz et al., "Faster sorting algorithms discovered using deep reinforcement learning", Nature 618, 257:263 (2023) [link]

    Instructor

    1. Marco Piastra

    2. Contact: marco.piastra@unipv.it


    Kiro

    1. Course info


    Exams

    1. See Faculty website


    Further resources:

    All Colab notebooks used in the course are available on Kiro


    Links

    1. Artificial Intelligence Reading Group


    1. Deep Learning, A.A. 2021-2022 and before