Certifiable Unlearning Pipelines for Logistic Regression: An Experimental Study
Round 1
Reviewer 1 Report
The paper is devoted to the un-learning of machine learning models using linear regression as a case study. The topic is quite interesting due to the various possible applications in different fields. First of all, I want to note that the abstract is written in a very clear way. The concept of unlearning and the main points of the research become clear to me even before reading the main paper. The main methods and results presented in the paper are adequate. The source codes are provided, which improves the reproducibility of the results. Of course, it is more interesting to analyse the effectiveness of un-learning for other ML models too. However, the paper is quite large already and can be considered self-sufficient. So, I think the paper can be accepted to MAKE in its current form.Author Response
Please see the attachment for the responses
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The authors present the theories and applications of real datasets in terms of machine unlearning, while comparing with two of effectiveness, efficiency and certifiability. In particular, they propose a practical online strategy under the condition that unlearning errors gets larger enough to warrant a full retraining of the ML model.
As a whole, the structure and propositions including the pipeline of unlearning in figure 1 are technically sound, and the completeness of the manuscript would be enhanced by further explanation in considerations of feasible deletion distributions in Figure 2, e.g. selectively-deleted over irregular spots. Also, the authors may present the feasibility of unlearning pipelines about other cases besides logistic regression. Minor polishing of the entire manuscript will be better for the readability of the researchers in academic and practical ML fields.
Author Response
Please see the attachment for the responses
Author Response File: Author Response.pdf
