Pinned Loading

1. HeatMapSae-A-Gradient-Centric-Approach-to-Explaining-CLIP-Through-Sparse-Latent-Activations HeatMapSae-A-Gradient-Centric-Approach-to-Explaining-CLIP-Through-Sparse-Latent-Activations Public

   HeatMapSae merges gradient-based saliency methods with Sparse AutoEncoders in CLIP’s intermediate layers. By altering latent features, we track how attention maps shift, revealing mechanistic cues …

   Jupyter Notebook

2. Interpreting-Language-Models-with-SAEs-and-SHAP-Analyzing-Activations-for-Medical-Terms Interpreting-Language-Models-with-SAEs-and-SHAP-Analyzing-Activations-for-Medical-Terms Public

   This repository contains an implementation for interpreting GPT-2 Small using Sparse Autoencoders (SAEs) and SHAP, focusing on analyzing feature activations associated with medical terms.

   Jupyter Notebook

3. Enhancing-Interpretability-Methods-in-Imaging-Using-the-Segment-Anything-Model-SAM-Algorithm Enhancing-Interpretability-Methods-in-Imaging-Using-the-Segment-Anything-Model-SAM-Algorithm Public

   Through an experimental approach, we evaluate how incorporating SAM-generated superpixels improves the coherence and understandability of explanations in interpretation methods such as LIME, GLIME,…

   Jupyter Notebook

4. Segmentation_medical_images Segmentation_medical_images Public

   In this repository, different advanced segmentation models (2023) for medical images are implemented and compared.

   Jupyter Notebook 2

5. XAI_ARENA XAI_ARENA Public

   Implementation and rigorous comparison of agnostic and introspective interpretability methods for black-box models using defined metrics and frameworks.

   Jupyter Notebook 2

6. VulnerabilityAndPhishingDetection-GPT VulnerabilityAndPhishingDetection-GPT Public

   A repository showcasing experiments and implementations for detecting vulnerabilities in code and phishing attacks using GPT-based models. Includes datasets, scripts, and results with performance m…

   Jupyter Notebook