GitHub repository for DORA: Data-agnOstic Representation Analysis paper. DORA allows to find outlier representations in Deep Neural Networks.
-
Updated
Apr 23, 2026 - Python
GitHub repository for DORA: Data-agnOstic Representation Analysis paper. DORA allows to find outlier representations in Deep Neural Networks.
Interpretability study of LoRA fine-tuning in LLMs: where LoRA puts its capacity (per-module update norm + effective rank) and how much it shifts representations (layer-wise drift), with rank/module ablations. Qwen2.5-1.5B, paper + code.
Attribution-based analysis of French grammatical gender encoding in FlauBERT embeddings using SHAP, LIME, Random Forest, and cross-validation methods across multiple architectures
Code and artifacts for "The Cost of Language: Centroid Erasure Exposes and Exploits Modal Competition in Multimodal Language Models" (COLM 2026). Centroid replacement, a training-free probe for text-vs-visual dependence in multimodal LMs, and text centroid contrastive decoding (TCCD) are included.
Applied research in representation and decision behavior
Add a description, image, and links to the representation-analysis topic page so that developers can more easily learn about it.
To associate your repository with the representation-analysis topic, visit your repo's landing page and select "manage topics."