Paper explained
Forgetting Machines: Responsibility by Design in Machine Unlearning
What would it take to build models that can forget by design?
The question
Why this work
This PhD proposal connects the meaning of forgetting to the engineering of systems that can unlearn. It asks how to integrate unlearning into data workflows, test its limits, and make future removal intentional and accountable.
The approach
A visual guide
Three connected research directions
Integrate
Make unlearning part of data preparation, model repair, and evaluation.
Investigate
Probe residual influence, semantic dependencies, and lost capabilities.
Design
Build traceability, source separation, and verification into future systems.
What the work contributes
The proposed agenda combines statistical and mechanistic verification with architectural choices such as source separation, provenance tracking, and retained training updates. Mechanistic interpretability is proposed as one way to examine traces that behavior alone may miss.
Original paper & resources
PhD Topic Proposal, UCU Applied Sciences Faculty (2025).
Citation
Makovska, V. (2025). Forgetting Machines: Responsibility by Design in Machine Unlearning [PhD topic proposal]. Ukrainian Catholic University.