Announcing the first Machine Unlearning Challenge

Sub: Sci

Sec: Awareness in IT

Context:

Google announced the first Machine Unlearning Challenge, which will be held as part of the NeurIPS 2023 Competition Track.

What is Machine Unlearning:

Machine unlearning refers to the process of selectively removing specific training data points and their influence on an already trained model, making the updated model behave the same as a model that was never trained on that data.
It provides a ‘subtractive’ capability to adapt models by removing unauthorized, malicious, or outdated data points without full retraining.
Machine unlearning plays an important role in ML models.
The field of machine unlearning is related to other areas of machine learning such as differential privacy, life-long learning, and fairness.

What are the challenges of machine unlearning?

The problem of unlearning is complex and multifaceted as it involves several conflicting objectives like forgetting the requested data, maintaining the model’s utility (e.g., accuracy on retained and held-out data), and efficiency.
Also the evaluation of forgetting algorithms in the literature has so far been highly inconsistent.
Another challenge in this space is the effective erasure of data that has been requested to be deleted.
The unlearning process itself can leak information in multiple ways.
Tracing each data point’s influence becomes increasingly difficult as the dataset changes dynamically.

Unlearning challenge:

Google announced the first Machine Unlearning Challenge, which will be held as part of the NeurIPS 2023 Competition Track.
The goal of the competition is twofold.
- Unifying and standardizing the evaluation metrics for unlearning to identify the strengths and weaknesses of different algorithms through apples-to-apples comparisons.
- Opening this competition to everyone, will foster novel solutions and shed light on open challenges and opportunities.
The competition was hosted on Kaggle and ran between mid-July 2023 and mid-September 2023.