Neural Networks and Deep Learning
Jacopo Pegoraro Daniele Mari
[Link]@[Link] [Link]@[Link]
2022-2023
Instructors
● Jacopo Pegoraro ([Link]@[Link])
● 16 theory lessons
● Applications
● 32 hours
● Daniele Mari ([Link]@[Link])
● 8 lab sessions
● 16 hours
Contact us
● For any question regarding theory lessons or the exam
● Email [Link]@[Link]
● Start the object of the email with «[NNDL]»
● For any questions regarding lab sessions
● Email [Link]@[Link]
● Cc [Link]@[Link]
● Start the object of the email with «[NNDL]»
(Artificial) Neural Networks (ANN)
● Network of artificial neurons
● Input signals propagate to outputs
● Layered structure
● Tunable connectivity
● ANN vs Human brain
● Extremely simplified topology
● Discrete vs continuous inputs
Course outline (1)
1. Neural Networks basics
● Learning problems and optimization methods
● Regression and classification
● Overfitting and regularization
● Feedforward Neural Networks (FFNN)
● Training neural networks: backpropagation
● Convolutional Neural Networks (CNN)
● Overfitting in NNs and advanced optimization
Course outline (2)
2. Advanced NN architectures
● Batch normalization
● Residual networks
● Auto-encoders and variational auto-encoders
● Generative Adversarial Networks
● Continual Learning
Lab sessions (1)
● Python programming language
● library for scientific computing
● library to build NN models
● Guided lab coding sessions
● Google Colab
● Nice and clean text/code integration
● Easy access to GPUs
Lab sessions (2)
1. Intro to Python & NumPy + Least Squares
2. Intro to PyTorch
3. Regression and classification with a NN (PyTorch)
4. Neural Network implementation from scratch (NumPy)
5. Convolutional NN and Residual Networks
6. Auto-encoders and Variational Auto-encoders
7. Generative Adversarial Networks
8. Continual Learning
Prerequisites
● For the theory lessons
● Calculus - derivatives, functions, integrals, …
● Probability - distributions, expectation, variance, …
● Linear Algebra - matrix/vector operations, matrix properties, …
● For the labs
● Everything will be covered from scratch
● Basics of programming can help
Relation to other courses
● Machine Learning (ICT)
● Highly recommended Deep
Learning
● Learning theory
● Non NN-based algorithms
Neural Networks
● Human Data Analytics (ICT)
● Advanced ML&DL course Machine Learning
● Focus on sequential data
Course material
● Lecture slides
● [Link]
● Enrollment key: nndl22
● Cover theory and (some) applications
● Lab material
● Code (notebooks)
● Links to documentation and tutorials
● Lecture recordings
Self study material
● Books
[1] J. Watt, R. Borhani, A. K. Katsaggelos, Machine learning refined: foundations,
algorithms and applications, Cambridge University Press, 2016.
[2] S. S. Shwartz, S. Ben-David, Understanding machine learning: from theory to
algorithms, Cambridge University Press, 2014.
[3] Ian Goodfellow, Yoshua Bengio, and Aaron Courville. Deep learning. MIT press,
2016. [Link]
[4] Christopher M. Bishop, and Nasser M. Nasrabadi. Pattern recognition and machine
learning. New York: Springer, 2006.
● Research papers provided after each lecture
Exam modalities
● STEP 1: GROUP PROJECT (MAX 3 people)
● Implement and test a neural network on a public dataset
● Projects provided by the instructors (or proposed by you, after approval)
● STEP 2: PROJECT OUTPUT
● Report: max 10 pages + contribution description for each member
● Complete source code
● To be delivered 1 week before the oral exam in step 3
● STEP 3: ORAL PRESENTATION
● 15 min. per group, each member gets to present something
● 10 min. questions on the project and/or on the course topics
Grading
● PROJECT – 50% ● Originality
● Preprocessing
● Learning model
● Comparison with other methods
● WRITTEN REPORT - 25% ● Clarity
● Completeness
● Results
● ORAL – 25% ● Clarity of the presentation
● Questions
Research
● Neuromorphic computing
● Deep learning applied to Radio Frequency signals
Neuromorphic computing
Synaptic
weights
Information is
encoded in
continuous
time pulses
Leaky integrate-and-fire model
Sensory Neural networks
inputs that mimic the
brain in hardware
DL for joint communication and sensing
● Indoor human detection and sensing through a testbed implementing the new
IEEE 802.11ay Wi-Fi standard at 60 GHz
processed Wi-Fi data
HAR classifier Identification classifier
• Walking • Person 0
•
Softmax
Softmax
Running
Dense
Dense
• Person 1
Residual Block • Sitting Residual Block • Person 2
• Waving hands
X4 X4 • …
• Standing
Walking spectrogram
micro-Doppler input
If you are curious…
● M. Canil*, J. Pegoraro*, and M. Rossi, “MilliTRACE-IR: Contact Tracing and Temperature
Screening via mmWave and Infrared Sensing”, IEEE Journal of Selected Topics in Signal
Processing, vol. 16, no. 2, pp. 208-223, May 2022.
● J. Pegoraro and M. Rossi, “Human Tracking with mmWave Radars: a Deep Learning
Approach with Uncertainty Estimation”, IEEE International Workshop on Signal Processing
Advances in Wireless Communications (SPAWC), Oulu, Finland, July 2022.
● J. Pegoraro, J. O. Lacruz, F. Meneghello, E. Bashirov, M. Rossi, and J. Widmer, “RAPID:
Retrofitting IEEE 802.11ay Access Points for Indoor Human Detection and Sensing”, under
submission, 2022.
● Furber Steve and Temple Steve, Neural systems engineering Journal of the Royal Society
Interface, 4193–206, 2007
