• Aucun résultat trouvé

Introduction to Artificial Intelligence & Machine Learning

N/A
N/A
Protected

Academic year: 2022

Partager "Introduction to Artificial Intelligence & Machine Learning"

Copied!
48
0
0

Texte intégral

(1)

Introduction to Artificial

Intelligence & Machine Learning

Nicolas Thome Professor at Cnam Computer science dpt Machine Learning team

(2)

1.Definition of AI and ML 2.Unsupervised learning 3.Supervised learning

Outline

(3)

Artificial Intelligence

Building machines able to solve problems, work & react like humans

Requiring understanding of the problem

Very general, being able to

Acquire and understand information from the world, environment => perception

Image, audio, text, … and any sensor / measurement (physics )

(4)

Artificial Intelligence

Building machines able to solve problems, work & react like humans

Requiring understanding of the problem

Very general, being able to

Perform action in the world

Robot, chatbot, playing games, etc

(5)

Artificial Intelligence & big data

Big data => huge number of data

Impossible to manually process such data

=> Obvious need for automatic processing

(6)

Big data applications: essentially all data sience domains

Email filtering, Online recommendations

Voice recognition, Face recognition

Medical diagnosis

Autonomous driving

(7)

Artificial Intelligence & Machine learning AI ambiguous

(8)

Historical Artificial Intelligence

Traditional IA (1950-1990): symbolic problems

Constraint satisfaction problem (CSP) => Optimization/ search issues

games (chess, go), Travelling salesman problem, etc

Ex: Travelling salesman problem (TSP)

Find the shortest path to visit all n cities

Exhaustive search: O(n!)

Explodes very quickly with n

(9)

Historical Artificial Intelligence: Expert systems

Knowledge base collected by experts, expressed by if-then rules

Inference: deduce new facts from knowledge

(10)

Artificial Intelligence & Machine learning

Traditional AI: explicit rules, handcrafted programs

Difficult to build and maintain knowledge database

For many pbs: impossible to explicitly express rules (ex: image classification)

ML: rules learned from data, emerged from data

(11)

Machine learning: methods and supervision

Unsupervised vs supervised learning

(12)

Machine learning: methods and supervision

Reinforcement learning

(13)

Machine learning & generalization

Inductive learning: training database => extract rules

Apply to new data

Machine Learning ≠ optimization

Under-fitting vs overfitting

(14)

Machine learning: representation

For many tasks: input representations not adequate

(15)

Deep learning: learning representations

ML on hacrafted features DL on raw data

(16)

1.Definition of AI and ML 2.Unsupervised learning 3.Supervised learning

Outline

(17)

Unsupervised learning

General motivation: learning the structure of data

Useful for:

Clustering

Visualization

Learning representations, manifold learning etc

(18)

K-Means

(19)

K-Means

(20)

K-Means

(21)

K-Means: python example on MNIST for clustering

Cluster with min entropy

(22)

K-Means: python example on MNIST for clustering

Cluster with max entropy

(23)

Principal Component Analysis

(24)

Principal Component Analysis

(25)

Principal Component Analysis

(26)

Unuspervised learning

And many other methods…

Generative models, e.g. Gaussian Mixture Models (GMMs)

Maximum likelihood vs Maximum a Posteriori

(27)

1.Definition of AI and ML 2.Unsupervised learning 3.Supervised learning

Outline

(28)

Supervised learning

General methods

Decision trees and variants (random forest)

K-NN (nearest neighbor): For each test example, simply find its closest example

Or compute k-NN, and apply majority class voting

(29)

Supervised learning

=> Train a model with gradient descent

(30)

Supervised learning: gradient descent

(31)

Supervised learning

(32)

Neural Networks

(33)

Deep Neural Networks

(34)

Deep Neural Networks & expressivity

(35)

Deep Neural Networks: Training with backprop

(36)

Backprop: chain rule

(37)

Deep Neural Networks: specific architectures

(38)

Deep Neural Networks: specific architectures

(39)

Deep Neural Networks: specific architectures

(40)

Deep Neural Networks: specific architectures

(41)

Deep learning History

(42)

Deep learning History

(43)

Deep learning History

(44)

Deep learning since 2012

(45)

Deep learning since 2012

(46)

Deep learning since 2012: ressources

(47)

Deep learning & AI: ongoing issues

(48)

Deep learning & AI:ongoing issues

Références

Documents relatifs

A Bird’s Eye View on Requirements Engineering and Machine learning, the 25th Asia-Pacific Software Engineering Conference 2018 (APSEC 2018), December 4-7, Nara, Japan.. Feldt

In the early days of machine learning, Donald Michie intro- duced two orthogonal dimensions to evaluate performance of machine learning approaches – predictive accuracy

We conclude the section with a general lower bound, via a reduction to PAC learning, relating the query class’s VC dimension, attack accuracy, and number of queries needed for

En th´ eorie, on pourrait se dire que pour contraindre fortement le mod` ele, nous pourrions employer une fonction objectif plus restrictive, notamment en p´ enalisant le mod` ele

Like the face recognition and the natural language processing examples, most works discussing multi-task learning (Caruana 1997) construct ad-hoc combinations justified by a

From these results, we infer (i) that the Nem1-Spo7 module constitutively binds a fraction of Pah1, which is also in line with a previous report that implicated the

Dans la minute qui suit la première bouffée de cigarette, le coeur se met à battre plus rapidement. Les vaisseaux sanguins se contractent, entraînant un plus grand effort

We have shown that in the Biochemical Abstract Machine BIOCHAM, the rule-based language for modeling bio- molecular interactions, and the temporal logics used for formalizing