Skip to content

feat(symbolic-learning): add SL-5/6/7 — NeuroSymbolic, KG-ILP, LLM-Symbolic (#1404) #1633

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
jsboige merged 7 commits into from
May 28, 2026
Merged

feat(symbolic-learning): add SL-5/6/7 — NeuroSymbolic, KG-ILP, LLM-Symbolic (#1404) #1633

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
3fe57ee
feat(symbolic-learning): add SL-5/6/7 notebooks — NeuroSymbolic, KG-I…
jsboige May 27, 2026
03e64e5
fix(sl-5): fix DeepProbLog variable lookup bug + execute notebook + f…
jsboige May 27, 2026
7eb38e0
chore(catalog): refresh markers for SL-5/6/7 (ALL + SymbolicAI)
jsboige May 27, 2026
8c0f0df
fix(slides): replace openai TODO comments with clean doc references i…
jsboige May 27, 2026
5bdab7c
chore(catalog): refresh markers after rebase on main
jsboige May 27, 2026
40d490e
fix(sl-7): re-execute notebook to restore execution_count after cell …
jsboige May 27, 2026
39788bf
chore(catalog): refresh markers post SL-7 re-execution
jsboige May 27, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
92 changes: 82 additions & 10 deletions COURSE_CATALOG.generated.json
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4019,7 +4019,7 @@
"owner_logique": "po-2023",
"last_validation": "2026-05-27",
"last_validator": "jsboige@gmail.com",
"issue_pr_associee": "#999",
"issue_pr_associee": "#999, #1632",
"cells_total": 22,
"cells_code": 10,
"cells_markdown": 12,
Expand All @@ -4043,7 +4043,7 @@
"owner_logique": "po-2023",
"last_validation": "2026-05-27",
"last_validator": "jsboige@gmail.com",
"issue_pr_associee": "#999",
"issue_pr_associee": "#999, #1632",
"cells_total": 14,
"cells_code": 6,
"cells_markdown": 8,
Expand All @@ -4067,7 +4067,7 @@
"owner_logique": "po-2023",
"last_validation": "2026-05-27",
"last_validator": "jsboige@gmail.com",
"issue_pr_associee": "#999",
"issue_pr_associee": "#999, #1632",
"cells_total": 77,
"cells_code": 38,
"cells_markdown": 39,
Expand All @@ -4091,7 +4091,7 @@
"owner_logique": "po-2023",
"last_validation": "2026-05-27",
"last_validator": "jsboige@gmail.com",
"issue_pr_associee": "#999",
"issue_pr_associee": "#999, #1632",
"cells_total": 29,
"cells_code": 10,
"cells_markdown": 19,
Expand Down Expand Up @@ -11123,7 +11123,7 @@
"owner_logique": "po-2024",
"last_validation": "2026-05-27",
"last_validator": "jsboige@gmail.com",
"issue_pr_associee": "#1611",
"issue_pr_associee": "#1619",
"cells_total": 43,
"cells_code": 15,
"cells_markdown": 28,
Expand Down Expand Up @@ -11507,7 +11507,7 @@
"owner_logique": "po-2024",
"last_validation": "2026-05-27",
"last_validator": "jsboige@gmail.com",
"issue_pr_associee": "#1611",
"issue_pr_associee": "#1619",
"cells_total": 32,
"cells_code": 9,
"cells_markdown": 23,
Expand Down Expand Up @@ -11625,9 +11625,9 @@
"maturity": "BETA",
"duree_estimee": "30min",
"owner_logique": "po-2024",
"last_validation": "2026-05-26",
"last_validation": "2026-05-27",
"last_validator": "jsboige@gmail.com",
"issue_pr_associee": "#1404",
"issue_pr_associee": "#1614",
"cells_total": 28,
"cells_code": 13,
"cells_markdown": 15,
Expand All @@ -11649,9 +11649,9 @@
"maturity": "BETA",
"duree_estimee": "30min",
"owner_logique": "po-2024",
"last_validation": "2026-05-26",
"last_validation": "2026-05-27",
"last_validator": "jsboige@gmail.com",
"issue_pr_associee": "#1404",
"issue_pr_associee": "#1614",
"cells_total": 26,
"cells_code": 11,
"cells_markdown": 15,
Expand All @@ -11663,6 +11663,78 @@
"requires_wsl": false,
"executable_locally": true
},
{
"path": "SymbolicAI/SymbolicLearning/SL-5-NeuroSymbolic.ipynb",
"title": "SL-5 : Integration Neuro-Symbolique",
"serie": "SymbolicAI",
"sous_serie": "SymbolicLearning",
"kernel": "Python 3",
"status": "READY",
"maturity": "BETA",
"duree_estimee": "30min",
"owner_logique": "po-2024",
"last_validation": "2026-05-27",
"last_validator": "jsboige@gmail.com",
"issue_pr_associee": "",
"cells_total": 28,
"cells_code": 9,
"cells_markdown": 19,
"cells_with_outputs": 9,
"cells_without_outputs": 0,
"requires_api": false,
"requires_gpu": false,
"requires_cloud": false,
"requires_wsl": false,
"executable_locally": true
},
{
"path": "SymbolicAI/SymbolicLearning/SL-6-KnowledgeGraphs-ILP.ipynb",
"title": "SL-6 - ILP Moderne et Knowledge Graphs",
"serie": "SymbolicAI",
"sous_serie": "SymbolicLearning",
"kernel": "Python 3",
"status": "READY",
"maturity": "BETA",
"duree_estimee": "30min",
"owner_logique": "po-2024",
"last_validation": "2026-05-27",
"last_validator": "jsboige@gmail.com",
"issue_pr_associee": "#1404",
"cells_total": 39,
"cells_code": 13,
"cells_markdown": 26,
"cells_with_outputs": 13,
"cells_without_outputs": 0,
"requires_api": false,
"requires_gpu": false,
"requires_cloud": false,
"requires_wsl": false,
"executable_locally": true
},
{
"path": "SymbolicAI/SymbolicLearning/SL-7-LLM-SymbolicLearning.ipynb",
"title": "SL-7 - LLMs et Apprentissage Symbolique : Generation et Verification d'Hypotheses",
"serie": "SymbolicAI",
"sous_serie": "SymbolicLearning",
"kernel": "Python 3",
"status": "READY",
"maturity": "BETA",
"duree_estimee": "45min",
"owner_logique": "po-2024",
"last_validation": "2026-05-27",
"last_validator": "jsboige@gmail.com",
"issue_pr_associee": "",
"cells_total": 38,
"cells_code": 16,
"cells_markdown": 22,
"cells_with_outputs": 15,
"cells_without_outputs": 0,
"requires_api": true,
"requires_gpu": false,
"requires_cloud": false,
"requires_wsl": false,
"executable_locally": false
},
{
"path": "SymbolicAI/Tweety/Tweety-1-Setup.ipynb",
"title": "Configuration et Installation TweetyProject",
Expand Down
6 changes: 3 additions & 3 deletions MyIA.AI.Notebooks/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,9 +4,9 @@ Ecosysteme complet de **555 notebooks** Jupyter pour l'apprentissage des technol

<!-- CATALOG-STATUS
series: ALL
total: 496
breakdown: GenAI=114, QuantConnect=102, SymbolicAI=94, Search=45, Probas=43, Sudoku=32, ML=30, GameTheory=25, RL=6, CaseStudies=4, IIT=1
maturity: BETA=287, PRODUCTION=152, ALPHA=43, DRAFT=10, TEMPLATE=4
total: 499
breakdown: GenAI=114, QuantConnect=102, SymbolicAI=97, Search=45, Probas=43, Sudoku=32, ML=30, GameTheory=25, RL=6, CaseStudies=4, IIT=1
maturity: BETA=290, PRODUCTION=152, ALPHA=43, DRAFT=10, TEMPLATE=4
-->

Dernière mise à jour : 2026-05-23
Expand Down
6 changes: 3 additions & 3 deletions MyIA.AI.Notebooks/SymbolicAI/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,9 +2,9 @@

<!-- CATALOG-STATUS
series: SymbolicAI
pedagogical_count: 94
breakdown: SmartContracts=27, SemanticWeb=18, Lean=14, Planners=13, Tweety=10, Argument_Analysis=6, SymbolicLearning=4, =2
maturity: BETA=80, ALPHA=7, PRODUCTION=7
pedagogical_count: 97
breakdown: SmartContracts=27, SemanticWeb=18, Lean=14, Planners=13, Tweety=10, SymbolicLearning=7, Argument_Analysis=6, =2
maturity: BETA=83, ALPHA=7, PRODUCTION=7
-->

L'intelligence artificielle n'est pas qu'apprentissage automatique et réseaux de neurones. Une grande partie de l'IA classique repose sur le **raisonnement symbolique** : représenter la connaissance sous forme de propositions, de règles et de structures logiques, puis dériver mécaniquement de nouvelles conclusions. C'est cette tradition — des systèmes experts des années 80 aux assistants de preuve modernes comme Lean 4 — que cette série explore en profondeur.
Expand Down
54 changes: 46 additions & 8 deletions MyIA.AI.Notebooks/SymbolicAI/SymbolicLearning/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,25 +2,25 @@

<!-- CATALOG-STATUS
series: SymbolicLearning
pedagogical_count: 4
breakdown: =4
maturity: BETA=4
pedagogical_count: 7
breakdown: =7
maturity: BETA=7
-->

Comment un agent peut-il apprendre a partir de connaissances existantes plutot que de donnees brutes ? Cette serie explore l'apprentissage symbolique tel que decrit dans le chapitre 19 d'AIMA (Russell & Norvig), depuis l'apprentissage inductif pur (CBH, Version Space) jusqu'aux methodes guidees par la connaissance (EBL, RBL).

Le premier notebook pose les bases : representation d'hypotheses comme conjonctions de contraintes, algorithmes Current-Best-Hypothesis et Candidate Elimination (Version Space), et leurs limites face au bruit et aux concepts disjonctifs. Le second notebook montre comment la connaissance du domaine accelere l'apprentissage : l'apprentissage base sur les explications (EBL) compile les theories en heuristiques operationnelles, et l'apprentissage base sur la pertinence (RBL) identifie les attributs determinant via les determinations. Le troisieme notebook approfondit le RBL avec le treillis des determinations, l'algorithme MINIMAL-CONSISTENT-DET et une comparaison avec sklearn. Le quatrieme notebook couvre la programmation logique inductive (ILP) : l'algorithme FOIL (top-down), la resolution inverse (bottom-up) et la connexion avec les knowledge graphs.

**A qui s'adresse cette serie** : etudiants en IA, infoirmaticiens interesses par le raisonnement symbolique, et chercheurs en apprentissage automatique souhaitant comprendre les approches non-statistiques. Les notebooks (~2h total) ne necessitent que Python 3.10+ standard library. Une familiarite avec la logique propositionnelle suffit. Ils constituent un complement theorique aux series [Tweety](../Tweety/README.md) (argumentation computationnelle) et [SemanticWeb](../SemanticWeb/README.md) (representation de connaissances).
**A qui s'adresse cette serie** : etudiants en IA, infoirmaticiens interesses par le raisonnement symbolique, et chercheurs en apprentissage automatique souhaitant comprendre les approches non-statistiques. Les notebooks (~6h total) ne necessitent que Python 3.10+ standard library. Une familiarite avec la logique propositionnelle suffit. Ils constituent un complement theorique aux series [Tweety](../Tweety/README.md) (argumentation computationnelle) et [SemanticWeb](../SemanticWeb/README.md) (representation de connaissances).

## Vue d'ensemble

| Statistique | Valeur |
|-------------|--------|
| Notebooks | 4 |
| Notebooks | 7 |
| Kernel | Python 3 |
| Duree estimee | ~210 min |
| prerequis | Python 3.10+ (standard library + sklearn pour SL-3/SL-4) |
| Duree estimee | ~370 min |
| prerequis | Python 3.10+ (standard library + sklearn pour SL-3/SL-4, rdflib pour SL-6) |

## Notebooks

Expand All @@ -30,6 +30,9 @@ Le premier notebook pose les bases : representation d'hypotheses comme conjoncti
| 2 | [SL-2 - Apprentissage et Connaissance](SL-2-KnowledgeBasedLearning.ipynb) | EBL, RBL, determinations | 55 min |
| 3 | [SL-3 - Apprentissage Base sur la Pertinence](SL-3-RelevanceLearning.ipynb) | Treillis des determinations, MINIMAL-CONSISTENT-DET, RBL vs sklearn | 50 min |
| 4 | [SL-4 - Programmation Logique Inductive](SL-4-InductiveLogicProgramming.ipynb) | FOIL, resolution inverse, clauses Horn, knowledge graphs | 55 min |
| 5 | [SL-5 - Integration Neuro-Symbolique](SL-5-NeuroSymbolic.ipynb) | T-norms, predicats neuronaux, LTN, DeepProbLog | 55 min |
| 6 | [SL-6 - ILP Moderne et Knowledge Graphs](SL-6-KnowledgeGraphs-ILP.ipynb) | rdflib, AMIE rule mining, completion KG | 55 min |
| 7 | [SL-7 - LLMs et Apprentissage Symbolique](SL-7-LLM-SymbolicLearning.ipynb) | Prompting, extraction de regles, verification symbolique | 50 min |

## Contenu detaille

Expand Down Expand Up @@ -80,6 +83,33 @@ Le premier notebook pose les bases : representation d'hypotheses comme conjoncti
| Knowledge Graphs | Regles AMIE, triples RDF, SPARQL CONSTRUCT |
| Exercices | sibling, operateur W, regles sur KG |

### SL-5-NeuroSymbolic.ipynb

| Section | Contenu |
|---------|---------|
| T-norms / T-conorms | Operateurs logiques differentiables |
| Predicats neuronaux | Fonctions P(x) -> [0,1] apprises |
| LTN | Logique Tensorielle simplifiee |
| Raisonnement guide | Regles logiques guidant l'entrainement neuronal |
| DeepProbLog | Programmation logique probabiliste + predicats neuronaux |

### SL-6-KnowledgeGraphs-ILP.ipynb

| Section | Contenu |
|---------|---------|
| Knowledge Graphs | Construction avec rdflib |
| AMIE rule mining | Decouverte de regles de Horn sur KG |
| Completion | Inference de nouveaux triples |

### SL-7-LLM-SymbolicLearning.ipynb

| Section | Contenu |
|---------|---------|
| LLMs et raisonnement | Forces et limites pour le raisonnement symbolique |
| Parseur de regles | Extraction de regles IF-THEN depuis du texte |
| Verification symbolique | Coherence formelle des sorties LLM |
| Boucle LLM-Symbolique | Generation + verification + feedback |

## Concepts cles

| Concept | Explication | Notebook |
Expand All @@ -98,6 +128,11 @@ Le premier notebook pose les bases : representation d'hypotheses comme conjoncti
| **Clause Horn** | Regle logique avec au plus un litteral positif | SL-4 |
| **Unification** | Trouve une substitution rendant deux termes egaux | SL-4 |
| **ILP** | Apprentissage de programmes logiques a partir d'exemples | SL-4 |
| **T-norm** | Generalisation differentiable de AND | SL-5 |
| **DeepProbLog** | Programmation logique probabiliste + predicats neuronaux | SL-5 |
| **Knowledge Graph** | Graphe oriente de triples (sujet, predicat, objet) | SL-6 |
| **AMIE** | Rule mining sur knowledge graphs incomplets | SL-6 |
| **LLM-Symbolique** | Boucle de retroaction LLM + verification formelle | SL-7 |

## prerequis

Expand All @@ -109,7 +144,7 @@ Le premier notebook pose les bases : representation d'hypotheses comme conjoncti

### Environnement Python

Aucune dependance externe pour SL-1 et SL-2 (bibliothèque standard Python 3.10+ uniquement). SL-3 utilise `scikit-learn` et `numpy` pour la comparaison avec la selection statistique. SL-4 utilise uniquement la bibliothèque standard.
Aucune dependance externe pour SL-1 et SL-2 (bibliothèque standard Python 3.10+ uniquement). SL-3 utilise `scikit-learn` et `numpy` pour la comparaison avec la selection statistique. SL-4 utilise uniquement la bibliotheque standard. SL-5 utilise uniquement la bibliotheque standard. SL-6 utilise `rdflib`. SL-7 utilise uniquement la bibliotheque standard.

## Ressources

Expand All @@ -127,6 +162,9 @@ SymbolicLearning/
├── SL-2-KnowledgeBasedLearning.ipynb # EBL, RBL
├── SL-3-RelevanceLearning.ipynb # Treillis, MINIMAL-CONSISTENT-DET, RBL vs sklearn
├── SL-4-InductiveLogicProgramming.ipynb # FOIL, resolution inverse, knowledge graphs
├── SL-5-NeuroSymbolic.ipynb # T-norms, LTN, DeepProbLog
├── SL-6-KnowledgeGraphs-ILP.ipynb # rdflib, AMIE rule mining
├── SL-7-LLM-SymbolicLearning.ipynb # LLMs + verification symbolique
├── reference/
│ └── AIMA_Ch19_Knowledge_in_Learning.md # Notes de reference
└── README.md # Cette documentation
Expand Down
Loading
Loading