# Lexique et définitions rencontrées dans la bibliographie
## Activation Patching
Méthode d'explication/interprétation de DNN
([Chen et al., 2024] - Activation Patching for IR) "replaces the output of a component from one forward pass (e.g., an attention layer) with that from a similar input (‘patching’)"
Autres noms pour cette méthode : causal mediation analysis, causal tracing, interchange interventions
## Continual Learning
[Hou et al., 2025] "Continual learning refers to the capability of a machine learning model to learn and adapt to new
information, without compromising its performance on previously learned tasks"
...
...
@@ -14,7 +20,8 @@ a specific user’s information need from a corpus of documents.
## Interprétabilité
## Interprétabilité mécanique
"Mechanistic interpretability is an emerging diagnostic approach for neural models that has gained traction in broader natural language processing domains. This paradigm aims to provide attribution to components of neural systems" [Parry et al., 2025]
"Mechanistic interpretability is an emerging diagnostic approach for neural models that has gained traction in broader natural language processing domains. This paradigm aims to provide attribution to components of neural systems" ([Parry et al., 2025] - MechIR)
([Chen et al. 2024] - Activation Patching for IR) " Mechanistic Interpretability focuses primarily on understanding learned behaviors of the Transformer architecture underlying modern NLP systems"
## Question Answering (et variantes)
Tâche d'extraction de document (IR) et de RAG qui vise à répondre à une question à l'aide des documents de la base de données. On retrouve plusieurs définitions et différentes variantes
...
...
@@ -29,8 +36,8 @@ Tâche d'extraction de document (IR) et de RAG qui vise à répondre à une ques
([Xu et al., 2024] - SearChain) "first plan each sub-question and then solve them, they are not suitable for scenarios where the next sub-question needs the answer of the previous sub-questions to generate, which is common for complex knowledge-intensive tasks"
### Open Domain QA
([Thakur et al., 2021] - BEIR) "Retrieval in open domain question answering [8]is the task of retrieving the correct answer for a question, without a predefined location for the answer. In open-domain tasks, model must retrieve over an entire knowledge source (such as Wikipedia). We consider the question as input and the passage containing the answer as output"
([Lewis et al., 2020]) "“Closed-Book QA” approaches[52], which, like RAG, generate answers, but which do not exploit retrieval, instead relying purely on parametric knowledge"
([Thakur et al., 2021] - BEIR) "Retrieval in open domain question answering is the task of retrieving the correct answer for a question, without a predefined location for the answer. In open-domain tasks, model must retrieve over an entire knowledge source (such as Wikipedia). We consider the question as input and the passage containing the answer as output"
([Lewis et al., 2020]) "“Closed-Book QA” approaches, which, like RAG, generate answers, but which do not exploit retrieval, instead relying purely on parametric knowledge"
([Petroni et al., 2020] - KILT)
- "in open-domain question answering (Chen et al.,2017) models need to find answers within a large corpus of text"
- "Open domain Question Answering (Chen et al., 2017) is the task of producing the correct answer
