Topic outline

• Intelligent Systems for Pattern Recognition A.A. 2018-19 (ISPR)

Code: 651AA, Credits (ECTS): 6, Semester: 2, Official Language: English

Instructor: Davide Bacciu

Contact: email - phone 050 2212749

Office: Room 367, Dipartimento di Informatica, Largo B. Pontecorvo 3, Pisa

Office Hours: Thursday, 16-18 (email to arrange meeting)

• This topic

• Course Information

Weekly Schedule

The course is held on the second term. The preliminary schedule for A.A. 2018/19 is provided in table below.

The first lecture of the course will be on Thursday 21/02/2019.

Day
Time    Room
Thursday
14-16
C1
Friday 11-13    L1

Course Prerequisites

Course prerequisites include knowledge of machine learning fundamentals (e.g. covered through ML course). Knowledge of elements of probability and statistics, calculus and optimization algorithms are also expected. Previous programming experience with Python is a plus for the practical lectures.

Course Overview

The course introduces students to the analysis and design of advanced machine learning models for modern pattern recognition problems and discusses how to realize advanced applications exploiting computational intelligence techniques.

The course is articulated in four parts. The first part introduces basic concepts and algorithms concerning pattern recognition, in particular as pertains sequence and image analysis. The next two parts introduce advanced models from two major learning paradigms, that are (deep) neural networks and generative models, and their use in pattern recognition applications. The last part will go into the details of the realization of selected recent applications of AI models. Presentation of the theoretical models and associated algorithms will be complemented by introductory classes on the most popular software libraries used to implement them.

The course hosts guest seminars by national and international researchers working on the field as well as by companies that are engaged in the development of advanced applications using machine learning models.

Topics covered - Bayesian learning, graphical models, deep learning models and paradigms, deep learning for machine vision and signal processing, advanced neural network models (recurrent, recursive, etc.), (deep) reinforcement learning, signal processing and time-series analysis, image processing, filters and visual feature detectors, pattern recognition applications (machine vision, bio-informatics, robotics, medical imaging, etc), introduction to machine learning and deep learning libraries.

Textbook and Teaching Materials

The course does not have an official textbook covering all its contents. However, a list of reference books covering parts of the course is listed at the bottom of this section (note that all of them have an electronic version freely available online).

[BRML] David Barber, Bayesian Reasoning and Machine Learning, Cambridge University Press

[DL] Ian Goodfellow and Yoshua Bengio and Aaron Courville , Deep Learning, MIT Press

• Lectures and Calendar

The official language of the course is English: all materials, references and books are in English.

Date Room Topic References Additional Material
1 21/02/2019
(14-16)
C1 Introduction to the course: motivations and aim; course housekeeping (exams, timetable, materials); introduction to modern pattern recognition applications
(slides)

2
22/02/2019
(11-13)
L1
Signal processing: timeseries; time domain analysis (statistics, correlation); spectral analysis; fourier analysis. (slides)

Demo
3
28/02/2019
(14-16)
C1
Image processing: feature descriptors (color histograms, SIFT), spectral analysis, feature detectors (edge, blobs and segments). (slides)

[1,2] Two high-level surveys on visual feature extraction and representation
4
01/03/2019
(11-13)
L1
Generative and Graphical Models: probability refresher; graphical model representation; directed and undirected models: Bayesian networks; Markov networks; conditional independence and d-separation. (slides)
[Video Bayesian Network Part2]
[BRML] Ch. 1 and 2 (Refresher)
[BRML] Sect. 3.1, 3.2 and 3.3.1
(conditional independence)
[BRML] Sect. 3.3 (Directed Models)
[BRML] Sect. 4.1, 4.2.0-4.2.2 (Undirected Models)
[BRML] Sect. 4.5 (Expressiveness)
Sofware
• Pyro - Python library based on PyTorch
• PyMC3 - Python library based on Theano
• Edward - Python library based on TensorFlow
5  7/03/2019
(14-16)
C1  Hidden Markov Models:  learning in directed graphical models; forward-backward algorithm;  generative models for sequential data
(slides)
[Video HMM Part1]
[BRML] Sect. 23.1.0 (Markov Models)
[BRML] Sect. 23.2.0-23.2.4 (HMM and forward backward)
[3]  A classical tutorial introduction to HMMs
6  8/03/2019
(11-13)
L1  Hidden Markov Models: part 2
Notes on the derivation of Jensen inequality for EM
[Video HMM Part2]
[Video HMM Part3]

14/03/2019
(14-16)
Lesson CANCELLED - Will be recovered in April

15/03/2019
(11-13)
Lesson CANCELLED - Will be recovered in April
7  21/03/2019
(14-16)
C1  Markov Random Fields: learning in undirected graphical  models; conditional random fields; pattern recognition applications
(slides)
[Video MRF Part1, Incomplete due to recording problems]

[BRML] Sect. 4.2.2, 4.2.5 (MRF)
[BRML] Sect. 4.4 (Factor Graphs)
[BRML] Sect. 5.1.1 (Variable Elimination and Inference on Chain)
[BRML] Sect. 9.6.0, 9.6.1, 9.6.4, 9.6.5 (Learning in MRF/CRF)
[4,5] Two comprehensive tutorials on CRF ([4] more introductory and [5] more focused on vision)
[6] A nice application of CRF to image segmentation

Sofware
Check out pgmpy: it has Python notebooks to introduce to working with MRF/CRF
8  22/03/2019
(11-13)
L1 Boltzmann Machines: bridging neural networks and generative models; stochastic neuron; restricted Boltzmann machine; contrastive divergence
(slides)
[Video MRF Part 2 + BM1]
[7] A clean and clear introduction to RBM from its author

Sofware
Matlab code for Deep Belief Networks (i.e. stacked RBM) and Deep Boltzmann Machines.
9  28/03/2019
(14-16)
C1 Bayesian Learning: non-parametric models; variational learning
(slides)
[Video RBM]
[Video Bayesian Learning]
[8] LDA foundation paper
[9] A gentle introduction to latent topic models
[10] Foundations of bag of words image representation

Sofware
10  29/03/2019
(11-13)
L1  Bayesian Learning: sampling methods
(Guest Lecture by Daniele Castellana)
(slides)
[BRML] Sect. 27.1-27.3, 27.4.1, 27.6.2

M1  01/04/2019
(13.30-16)
L1
Midterm 1 discussions

11  08/04/2019
(14-16)
L1
Learning in Structured Domain I: Recursive Neural Networks
(Guest lecture by Alessio Micheli)
(slides)

12 11/04/2019
(14-16)
C1 Learning in Structured Domain II: Neural Networks for Graphs
(Guest lecture by Alessio Micheli)
(slides)

13 12/04/2019
(11-13)
L1 Randomized Recurrent Neural Networks
(Guest lecture by Claudio Gallicchio)

(slides)

M2  29/04/2019
(13.30-16)
O1
Midterm 2 discussions
14 02/05/2019
(14-16)
C1 Convolutional Neural Networks (part I):
introduction to CNN; basic CNN elements
(slides)
[11-15] Original papers for LeNet, AlexNet, VGGNet, GoogLeNet and ResNet.
15 03/05/2019
(11-13)
L1
Convolutional Neural Networks (part II):
CNN architectures for image recognition; convolution visualization; advanced topics (deconvolution, dense nets); applications and code
[Video CNN part 2]
[16] Complete summary of convolution arithmetics
[17] Seminal paper on batch normalization
[18] CNN interpretation using deconvolutions
16 06/05/2019
(14-16)
(Guest seminar by Jan Göpfert)
(slides)

17 09/05/2019
(14-16)
C1 Deep Autoencoders: introduction to the deep learning module, sparse, denoising and contractive AE; deep RBM.
(slides) [video]
[DL] Chapter 14, Sect 20.3, 20.4.0 (from 20.4.1 onwards not needed)
[19] DBN: the paper that started deep learning
[20] Deep Boltzmann machines paper
[21] Review paper on deep generative models
[22] Long review paper on autoencoders from the perspective of representation learning
[23] Paper discussing regularized autoencoder as approximations of likelihood gradient
18  10/05/2019
(11-13)
L1 Gated Recurrent Networks: deep learning for sequence processing; gradient issues; long-short term memory; gated recurrent units; generative use of RNN
(
slides) [video]
[25] Original LSTM paper
[26] An historical view on gated RNN
[27] Gated recurren units paper
[28] Seminal paper on dropout regularization

Sofware
19  16/05/2019
(14-16)
C1 Advanced Recurrent Architectures: sequence-to-sequence;  attention models; multiscale network; memory networks; neural reasoning.
(slides) [video]
[29,30] Models of sequence-to-sequence and image-to-sequence transduction with attention
[31,32] Models optimizing dynamic memory usage (clockwork RNN, zoneout)
[33] Differentiable memory networks
[34,35] Neural Turing Machines and follow-up paper on pondering networks
[36] Transformer networks: a paper on the power of attention without recurrence
20  17/05/2019
(11-13)
L1 PyTorch – neural networks in Python:
python; pytorch; RNN
(Coding practice by Antonio Carta)
(slides)

21 23/05/2019
(14-16)
C1 An introduction to Tensorflow and Keras
python; numpy, tensorflow, keras;

(Coding practice by Luca Pedrelli)

(slides)

22 24/05/2019
(11-13)
L1 Generative and Unsupervised Deep Learning: explicit distribution models; variational autoencoders; adversarial learning.
(slides)
[Video part1 VAE] [Video part2 GAN]
[37] PixelCNN - Explict likelihood model
[38] Tutorial on VAE
[39] Tutorial on GAN (here another online resource with GAN tips)
[40] Wasserstein GAN
[41] Tutorial on sampling neural networks

Sofware
23 30/05/2019
(14-16)
C1 Advanced research topics: new recurrent models, graph neural networks, tensor-based neural models and structured output prediction
(Quick seminars by Antonio Carta, Daniele Castellana, Federico Errica and Marco Podda)
[Slides deep learning for graphs]

31/05/2019
(11-13)
L1 Lesson cancelled due to strike

24 03/06/2019
(16-18)
F1 Final lecture: course wrap-up; research themes; final projects; exam modalities
M3 06/06/2018 (13.30-16.30)
L1  Midterm 3 discussions

Course examination for students attending the lectures is performed in 3 stages: midterm assignments, a final project and an oral presentation. Midterms provide bonus points for the final grade.

Midterm Assignment

Midterm assignments consist in a very short presentation (5 minutes per person) to be given in front of the class, presenting the key/most-interesting aspects of one of the following tasks:

• A quick and dirty (but working) implementation of a simple pattern recognition algorithm
• A report concerning the experience of installing and running a demo application realized using available deep learning and machine learning libraries
• A summary of a recent research paper on topics/models related to the course content.

Students might be given some amount of freedom in the choice of assignments, pending a reasonable choice of the topic. The assignments will roughly be scheduled every 1 month.

Final project

The final project concerns preparing a presentation concerning a topic relevant to the course content or the realization of a software implementing a non-trivial learning model and/or a PR application relevant for the course.

Oral Presentation

Prepare a seminar on the project to be discussed in front of the instructor and anybody interested. Students are expected to prepare slides for a 20 minutes presentation which should summarize the ideas, models and results in the report. The exposition should demonstrate a solid understanding of the main ideas in the report. The second part of the oral examination will test knowledge of the course contents (models, algorithms and applications).

The final exam vote $$G$$ is determined as

$$G = G_P + G_O + \sum_{i=1}^{3} G_{M}^{i}$$

where

• $$G_P \in [1,15]$$ is the final project grade
• $$G_O \in [1,14]$$ is the oral grade
• $$G_{M}^i \in \{0,1\}$$ is the increment for the i-th midterm

Non attending students

Working students and those not attending course lectures will deliver an extended final project making up for the missing midterms. They will be subject to an oral examination like attending students.  Contact the instructor by mail to arrange project topics and examination dates.

Grade for this exam modality is determined as

$$G = 0.5 \cdot (G_P + G_O)$$

where $$G_P \in [1,32]$$ is the project grade and $$G_O \in [1,30]$$ is the oral grade

• References

Bibliographic References

1. Scott Krigg, Interest Point Detector and Feature Descriptor Survey, Computer Vision Metrics, pp 217-282, Open Access Chapter
2. Tinne Tuytelaars and Krystian Mikolajczyk, Local Invariant Feature Detectors: A Survey, Foundations and Trends in Computer Graphics and Vision, Vol. 3, No. 3 (2007) 177–2, Online Version
3. Lawrence R. Rabiner:a tutorial on hidden Markov models and selected applications in speech recognition. Proceedings of the IEEE, 1989, pages 257-286, Online Version
4. Charles Sutton and Andrew McCallum,  An Introduction to Conditional Random Fields, Arxiv
5. Sebastian Nowozin and Christoph H. Lampert, Structured Learning and Prediction, Foundations and Trends in Computer Graphics and Vision, Online Version
6. Philipp Krahenbuhl, Vladlen Koltun, Efficient Inference in Fully Connected CRFs with Gaussian Edge Potentials, Proc.of NIPS 2011, Arxiv
7. Geoffrey Hinton, A Practical Guide to Training Restricted Boltzmann Machines, Technical Report 2010-003, University of Toronto, 2010
8. D. Blei, A. Y. Ng, M. I. Jordan. Latent Dirichlet Allocation. Journal of Machine Learning Research, 2003
9. D. Blei. Probabilistic topic models. Communications of the ACM, 55(4):77–84, 2012, Free Online Version
10. G. Csurka, C. R. Dance, L. Fan, J. Willamowski, and C. Bray. Visual Categorization with Bags of Keypoints. Workshop on Statistical Learning in Computer Vision. ECCV 2004, Free Online Version
11. Y. LeCun, B. Boser, J. S. Denker, D. Henderson, R. E. Howard, W. Hubbard and L. D. Jackel. Handwritten digit recognition with a back-propagation network, Advances in Neural Information Processing Systems, NIPS, 1989
12. A. Krizhevsky, I. Sutskever, and G. E. Hinton. ImageNet classification with deep convolutional neural networks, Advances in Neural Information Processing Systems, NIPS, 2012
13. S. Simonyan and A. Zisserman.  Very deep convolutional networks for large-scale image recognition, ICLR 2015, Free Online Version
14. C. Szegedy et al,  Going Deeper with Convolutions, CVPR 2015, Free Online Version
15. K. He, X. Zhang, S. Ren, and J. Sun. Deep Residual Learning for Image Recognition. CVPR 2016, Free Online Version
16. V. Dumoulin, F. Visin, A guide to convolution arithmetic for deep learning, Arxiv
17. S. Ioffe, C. Szegedy, Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift, ICML 2013,  Arxiv
18. M.D. Zeiler and R. Fergus, Visualizing and Understanding Convolutional Networks, ICML 2013, Arxiv
19. G.E. Hinton, R. R. Salakhutdinov. Reducing the dimensionality of data with neural networks.Science 313.5786 (2006): 504-507, Free Online Version
20. G.E. Hinton, R. R. Salakhutdinov. Deep Boltzmann Machines. AISTATS 2009, Free online version.
21. R. R. Salakhutdinov. Learning Deep Generative Models, Annual Review of Statistics and Its Application, 2015, Free Online Version
22. Y. Bengio, A. Courville, and P. Vincent. Representation learning: A review and new perspectives. Pattern Analysis and Machine Intelligence, IEEE Transactions on, Vol. 35(8) (2013): 1798-1828, Arxiv.
23. G. Alain, Y. Bengio. What Regularized Auto-Encoders Learn from the Data-Generating Distribution, JMLR, 2014.
24. Y. Bengio, P. Simard and P. Frasconi, Learning long-term dependencies with gradient descent is difficult. TNN, 1994, Free Online Version
25. S. Hochreiter, J. Schmidhuber, Long short-term memory, Neural Computation, 1997, Free Online Version
26. K. Greff et al, LSTM: A Search Space Odyssey, TNNLS 2016, Arxiv
27. C. Kyunghyun et al, Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation, EMNLP 2014, Arxiv
28. N. Srivastava et al, Dropout: A Simple Way to Prevent Neural Networks from Overfitting, JLMR 2014
29. Bahdanau et al, Neural machine translation by jointly learning to align and translate, ICLR 2015, Arxiv
30. Xu et al, Show, Attend and Tell: Neural Image Caption Generation with Visual Attention, ICML 2015, Arxiv
31. Koutník et al, A Clockwork RNN, ICML 2014, Arxiv
32. Krueger, Zoneout: Regularizing RNNs by Randomly Preserving Hidden Activation, ICLR 2018, Arxiv
33. Sukhbaatar et al, End-to-end Memory Networks, NIPS 2015, Arxiv
34. A. Graves et al, Neural Turing Machines, Arxiv
35. A.Graves, Adaptive Computation Time for Recurrent Neural Networks, Arxiv
36. A. Vaswan et al, Attention Is All You Need, NIPS 2017, Arxiv
37. A. van der Oord et al., Pixel Recurrent Neural Networks, 2016, Arxiv
38. C. Doersch, A Tutorial on Variational Autoencoders, 2016, Arxiv
39. Ian Goodfellow, NIPS 2016 Tutorial: Generative Adversarial Networks, 2016, Arxiv
40. Arjovsky et al, Wasserstein GAN, 2017, Arxiv
41. T. White, Sampling Generative Network, NIPS 2016, Arxiv