Lecture Plan
Here we describe the lecture plan and insert the link to the corresponding material.
Meta-Introduction
A couple of tricks and advices about jupyter-notebook envirornment.
Introduction 🎒
Here we collect the introductory arguments for this course. In this lecture, we review some basic concepts in machine learning, like model construction and logistic regression.
Hence, we illustrate the general working scheme of an Artificial Neural Network, building an instance of such a model making use of the only numpy.
Hence, we show a first example of predictive model making use of neural network. Finally, we give some references and instructions to follow and execute these lectures.
Binary Classification 🛤️
In this section we introduce a binary classification of images. We will build a deep network, and apply it to cat vs non-cat classification.
Error Metrics 🌊
This module is about error metrics.
In particular, we focus on the definition of different metrics and when these are suitable to use, looking at several examples.
Optimisation 🦾
At this stage, our Deep Learning knowledge is mature enough to wonder how to measure model performances and how to solve the possible issues arising. We review the crucial concepts of underfit and overfit and how to face such problems that may afflict models. Furthermore, we introduce specific Deep Learning techniques to handle such issues.
Error Analysis 👨🏫
We introduce an important concept in the general lifecycle of a Data Science model. Error analysis is a crucial activity in order to improve machine learning models.
Sequence Models 🦿
Here we introduce RNNs and sequence models. We are going to explore the several architectures and the cases where these are useful. We will introduce NLP and Time Series applications.
Transfer Learning 🔧
This lecture is aimed to introduce one of the main motivations of the deep learning success, i.e. how a network trained on a task can actually take advantage of the previous acquired knowledge to perform better on another task.
Convolutional Neural Networks 🖼️
In these couple of lectures we introduce and develop Convolutional Neural Networks. The idea is to study the most famous CNN architectures and apply them to the principal image tasks:
- Image classification
- Image recognition
- Object detection
Transformer models 🤖
In this lecture, we face one of the Deep Learning model that revolutionised the field of AI.
We will start by the famous paper Attention is all you need in order to understand how this work is crucial not only for NLP but also for the whole Deep Learning field.
Advanced Topics 🧪
As advanced topics, we focus on techniques and advanced model combinations. In particular a non-exhaustive list of arguments will include:
Final project 🚧
The proposed final project is an application of what we have seen in the course of the lectures.
The idea is to build a whole pipeline from data collection to prediction and possibly deploy our model on a webapp.