Lesson 1 - Introduction

Question 1

Deductive Reasoning

Answer 1

• Foundation of math proofs and theorems
• From true assumption infers logical consequences
  
  • RULE(C→R): All men are mortal
  • CASE(C1): Socrate is a man
  • RESULT(R1): Socrate is mortal

Question 2

Inductive Reasoning

Answer 2

• Used in ML and by scientists
• From particular cases we try to generalize, no guarantee on truth
  
  • CASE(C1): Socrate is a man
  • RESULT(R1): Socrate is mortal
  • RULE(C→R): All men are mortal

Question 3

Abductive Reasoning

Answer 3

• Used in medical and investigative fields
• Assume that an implication (rule) is valid even on the contrary, no guarantee on truth
  
  • RULE(C→R): All men are mortal
  • RESULT(R1): Socrate is mortal
  • CASE(C1): Socrate is a man

Question 4

Algorithm

Answer 4

• ordered and finite sequence of instructions (understandable to the executor)
• leads to a result in a finite time
• procedure to solve a class of specific problems or to perform a computation
• not always possible to find an algorithmic approach, ML has to be used
  
  • difficult problem formalization
  • noise/uncertainty in data
  • high complexity of solution formulation

Question 5

When to choose ML?

Answer 5

• Learning is important when the system has to:
  
  • Adapt to the environment in which it operates (automatic personalizzation)
  • To improve its own performance with respect to a particular task
  • To discover regularities and new information (knowledge) from empirical data

Question 6

Relationship between Data and Knowledge in ML?

Answer 6

• ML methods are used to transform empirical information (present in the data) into new knowledge.
• Data is nowadays present everywhere and are abundant (Web pages content, Social networks, etc.)

Question 7

The Foundamental Assumption

Answer 7

• A relatively simple stochastic process exists that explains the observed data
  
  • we may not know the details, but we assume it exists [social behavior is not random, images of the same subject share common patterns]
• In ML we are interested in building a good or useful approximation of this stochastic process
  
  • we are not interested in the process itself [reverse engineering]

Question 8

What is the main goal of ML?

Answer 8

• to define some performance criteria and optimize it using datas or previous experience or knowledge about the domain.
• Models are defined over some parameters that will be learned by optimizing a given criterion
• Two types of models:
  
  • Descriptive Models -> obtain new knowledge
  • Predictive Models -> make predictions about the future

Question 9

2/3 examples of ML use

Answer 9

• Face Recognition
• Named Entity Recognition -> The problem of identifying entities in a sentence: places, titles, names, actions, etc.
• Document Classification
• Games and Adversary Profiling -> predict missing information based on the strategies the adversary used in the past (threats, reactions, etc.)
• Bioinformatics -> determine how likely is that a patient react positively to a given therapy
• Recommender Systems
• Speech Recognition
• Handwritten Recognition
• Social Network Analysis
• Generative Adversarial Learning