Genμ 2.0
Call for submissions
How the 2026 challenge is structured
Teams can work on the Face Unlearning Track, the Speech Unlearning Track, the legacy concept benchmark, or all three. The challenge is built around forgetting the target without damaging the surrounding capability of the generator.
Track summary
Face Unlearning
Model-agnostic face identity unlearning using CelebA and, where applicable, FFHQ. Participants may use any suitable model and must report the exact checkpoint.
Speech Unlearning
Model-agnostic speaker identity unlearning using LibriTTS / LibriTTS-R. Any compatible TTS or voice-cloning model may be used.
Concept Unlearning
The retained concept-unlearning benchmark with the published ESD, CA, FMN, and FADE baselines.
Paper submission and challenge report
Regular workshop paper
Teams may submit a separate paper describing their method to the U&ME Workshop through regular peer review by 9 July 2026. This can be done in parallel with the challenge artifact submission due on 10 July.
Competition report
The two highest-ranked eligible teams will be invited to co-author the Genμ 2.0 competition report. Eligibility requires both a top-two finish and a score above the relevant official track baseline. The report will be prepared after evaluation for the workshop camera-ready stage.
Required submission package
A submission is complete only when all three components below are accessible for evaluation.
Drive link
- Final unlearned checkpoint, weights, or adapter
- Any accessible cloud-storage provider may be used
- Download permission must be enabled for evaluation
- Use clear filenames and include any required configuration files
GitHub repository
- Complete unlearning/training code
- Inference and evaluation entry points
- Environment or dependency file
- Exact commit hash or release tag
Testing instructions
- Installation and data preparation
- Exact commands to reproduce outputs
- Hardware and runtime requirements
- Expected output structure and troubleshooting notes
Revision policy
Revisions are accepted until 10 July 2026. Submit the same track form again and identify it as a revision; the latest complete response received before the deadline is treated as the official submission for that team and track.
Track submission forms
Use one form per track. Teams entering multiple tracks must submit the corresponding form separately for each track.
Task 1 · Face
Submit model details, CelebA/FFHQ validation results, weights link, GitHub code, and testing instructions.
Face Track Google Form — link pendingTask 2 · Speech
Submit model details, LibriTTS validation results, weights link, GitHub code, and testing instructions.
Speech Track Google Form — link pendingTask 3 · Concept
Submit method details, benchmark outputs, weights link, GitHub code, and testing instructions.
Concept Track Google Form — link pendingFor any submission-related query, contact genmu.challenge.2025@gmail.com.
Site administrators: add the three published Google Form URLs once in submission-forms.js; every button on the site will update automatically.
Notes for participants
- Face and speech model choice is open; the suggested Hugging Face checkpoints are optional starting points only.
- Use the released public validation identity groups before submitting.
- We evaluate submitted code and weights and update the website leaderboard as verified results become available.
- Do not submit private credentials, access tokens, or datasets that cannot be legally shared.