Verril-Learn 🧠✂️
A Surgical Machine Unlearning Library for Edge Devices
🎯 What is Verril-Learn?
Verril-Learn is a novel machine unlearning library designed specifically for edge devices. It enables you to surgically remove the influence of specific data points from a trained model without full retraining.
Key Features
- 🔬 Surgical Unlearning: Remove specific data influence using Fisher Information + Gradient Ascent
- 🎭 Data Poisoning Simulation: Built-in tools to simulate and study data attacks
- 📱 Edge-Device Optimized: Lightweight algorithms suitable for resource-constrained environments
- 🔒 Privacy-Compliant: Implements concepts aligned with GDPR's "Right to be Forgotten"
📦 Installation
pip install -e .
Or install dependencies directly:
pip install -r requirements.txt
🚀 Quick Start
from verril_learn import get_poisoned_mnist, SimpleCNN, surgical_unlearn
# Step 1: Load poisoned MNIST (10% of '7's mislabeled as '1's)
train_loader, test_loader, poison_indices = get_poisoned_mnist(poison_ratio=0.1)
# Step 2: Train your model (standard PyTorch training loop)
model = SimpleCNN()
# ... train the model ...
# Step 3: Surgically unlearn the poisoned data
unlearned_model = surgical_unlearn(
model=model,
poisoned_data=poisoned_loader,
retain_data=clean_loader,
fisher_weight=0.1,
unlearn_lr=0.001,
unlearn_steps=10
)
🧪 The Algorithm
1. Fisher Information Matrix (Diagonal Approximation)
Identifies which weights are critical for the retain data:
F_ii = E[(∂L/∂θ_i)²]
2. Gradient Ascent on Poisoned Data
Maximizes the loss to make the model "forget":
θ = θ + α * ∇L(θ; D_poison)
3. Fisher-Weighted Regularization
Prevents catastrophic forgetting by penalizing large changes to critical weights:
L_total = -L_poison + λ * Σ F_ii * (θ_i - θ_i^original)²
📁 Project Structure
verril-learn/
├── setup.py # Package configuration
├── requirements.txt # Dependencies
├── README.md # This file
└── verril_learn/
├── __init__.py # Public API
├── data_loader.py # MNIST loading + poisoning
├── model.py # CNN architecture
└── core.py # Surgical unlearning algorithm
📚 References
📄 License
MIT License - See LICENSE file for details.
Built with ❤️ for the ML Security & Privacy community
Metadata
Release files for verril-learn 0.1.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| verril_learn-0.1.0.tar.gz | 12.8 kB | Without provenance | Details for verril_learn-0.1.0.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| verril_learn-0.1.0-py3-none-any.whl | Python 3 | none | any | Details for verril_learn-0.1.0-py3-none-any.whl |
Total release size: 25.4 kB
Release files / verril_learn-0.1.0.tar.gz
| Download URL | verril_learn-0.1.0.tar.gz |
|---|---|
| Size | 12.8 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
52369d0e7851925562331be68a9c0e5211563f0632d802051ab70b1dab639cc1
|
|
BLAKE2b-256 checksum How to use checksums |
84cafb6f939b7c54fb2dff909ccc759fe4e0e657c2cd7aad4f14bdb3d9ae9a20
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.12.0
|
Release files / verril_learn-0.1.0-py3-none-any.whl
| Download URL | verril_learn-0.1.0-py3-none-any.whl |
|---|---|
| Size | 12.6 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
1c79fc4d67c82ce6d950aefe38e3036d2f19ecb73ddf41bee5d5bd43d37da62d
|
|
BLAKE2b-256 checksum How to use checksums |
da1f294594a3cab0afcbeaaedaa3cffc1c8d97890b8ce533fd2b6d72c805c6a6
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.12.0
|
Security for verril-learn 0.1.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.