Interactive Concept Bottleneck Generative Models aims to enhance interpretability and user interaction within the domain of generative AI. The project introduces interactive concept bottleneck layers into generative models, enabling users to manipulate and explore human-understandable concepts during the generative process. Our objective is also to discover and influence emergent concepts actively useful not only for generation purpose but also to perform downstream tasks such as classification.
- Motivation: The drive to make generative models more understandable and interactive, allowing for a direct manipulation of the generative process.
- Why: Explore how interactive concept bottlenecks can facilitate a deeper understanding of generative mechanisms.
- Problem Solved: Address the challenge of limited interpretability and user control in generative models by introducing a layer of interaction.
If your README is long, add a table of contents to make it easy for users to find what they need.
git clone https://github.com/ltronchin/interactive-cem.git
cd interactive-cem
pip install -r requirements.txt...
This project is inspired by the pioneering work in Concept Bottleneck Generative Models and further extends the idea into an interactive realm. Key references include:
- Concept Bottleneck Generative Models by Aya Abdelsalam Ismail et al.
- Concept Bottleneck Models by Pang Wei Koh et al.
This project is available under the MIT License. See the LICENSE file for more details.
- Real-time manipulation and exploration of generative models via interactive concept bottlenecks.
- Enhanced model interpretability and user engagement.
- Applicable across diverse generative model architectures and datasets.
...
...