Safe inference using representation engineering.
xmaster6y / mulsi Goto Github PK
View Code? Open in Web Editor NEW:lock_with_ink_pen: Safe inference using representation engineering.
License: MIT License
mulsi's Introduction
mulsi's People
Contributors
Stargazers
Watchers
Forkers
sourcery-ai-experimentsmulsi's Issues
[EXPERIMENT] Detect concepts using images
Description
Detect concepts activated by the attack using images instead of text when building contrast vectors.
Acceptance Criterion
- The code is documented
- Utilities and class test are written
- The code was reviewed
Tasks
- Select concepts to work on.
- Gather a set of images for each selected concept.
- Build the contrast vectors.
Extend concept reading with datasets
Description
Extend concept reading with text datasets.
Acceptance Criterion
- The code is documented
- Utilities and class test are written
- The code was reviewed
Tasks
- Collect datasets.
- Compute dataset representation with class
Representation. - Extend script to evaluate on a dataset.
[EXPERIMENT] Expand the list of model to evaluate
Description
Expand the list of models to attack and evaluate the detection method on. The idea is to verify the effectiveness of the detection method.
And if possible, see if the some patterns appear when different models are attacked by different methods.
Acceptance Criterion
- The code is documented
- Utilities and class test are written
- The code was reviewed
Tasks
- List models and build processors.
- Attack the models and perform detection.
- Analyze results.
Attack Scripts
Description
Make some scripts and modules to easily reproduce attacks.
Acceptance Criterion
- The code is documented
- Utilities and tests are written
- The code was reviewed
Tasks
- Create a basic FGSM attack
- Create a script to scale the attack
Contrast Reading Script
Description
Make some scripts and modules to easily scale reading.
Acceptance Criterion
- The code is documented
- Utilities and tests are written
- The code was reviewed
Tasks
- Create a basic contrast reading
- Create a script to scale the reading
Add BERT-Attack on text modality
Description
Attacking the text modality with BERT-Attack. The attack was initially developed on BERT, thus it should be adapted to the targeted LLM.
Acceptance Criterion
- The code is documented
- Utilities and class test are written
- The code was reviewed
Tasks
- Develop the attack.
- Test script for the attack.
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.