Planting Undetectable Backdoors in Machine Learning Models · HackerPapers discussions

Planting Undetectable Backdoors in Machine Learning Models — mentioned by Kiboneu

Mon, 29 Dec 2025 08:13:49 GMT

I don’t know, but your question reminds me of this paper which seems to address it on a lower level: https://arxiv.org/abs/2204.06974 “Planting Undetectable Backdoors in Machine Learning Models” “ … On the surface, such a backdoored classifier behaves normally, but in reality, the learner maintains a mechanism for changing the classification of any input, with only a slight perturbation. Importantly, without the appropriate "backdoor key", the mechanism is hidden and cannot be detected by any computationally-bounded observer. We demonstrate two frameworks for planting undetectable backdoors…

Paper and discussion history. A mention is not an endorsement.

Planting Undetectable Backdoors in Machine Learning Models — mentioned by IncRnd

Sun, 26 Feb 2023 02:09:06 GMT

The actual paper is here: https://arxiv.org/abs/2204.06974

Paper and discussion history. A mention is not an endorsement.

Planting Undetectable Backdoors in Machine Learning Models — mentioned by doomrobo

Sat, 25 Feb 2023 20:31:33 GMT

Preprint: https://arxiv.org/abs/2204.06974

Paper and discussion history. A mention is not an endorsement.

Planting Undetectable Backdoors in Machine Learning Models — mentioned by kkfx

Mon, 06 Jun 2022 13:31:32 GMT

And isn't the better scoring those made locally by yourself? I suggest for ML https://arxiv.org/abs/2204.06974 or the fact you can't even trust a public model, since you can't really control it regularly in training terms. You are right saying there is not much scoring in RSS, beside Gnus I know zero other feed readers who do anything alike. In search engines terms, beside YaCy witch is distributed and not so good, I do not know anything else, about bookmarking there are countless "managers" but I doubt they can be useful: in scoring terms, for discovering, we need the classic Usenet and…

Paper and discussion history. A mention is not an endorsement.

Planting Undetectable Backdoors in Machine Learning Models — mentioned by PeterisP

Tue, 03 May 2022 04:20:45 GMT

I'm not saying that Meta did it, but recent research shows that it is possible and hard to detect - https://arxiv.org/abs/2204.06974 - so if they really wanted to, they could.

Paper and discussion history. A mention is not an endorsement.

Planting Undetectable Backdoors in Machine Learning Models — mentioned by JoeDaDude

Tue, 26 Apr 2022 16:10:49 GMT

This is the kind of stuff I had in mind when making the suggestion about sabotage: https://arxiv.org/abs/2204.06974

Paper and discussion history. A mention is not an endorsement.

Planting Undetectable Backdoors in Machine Learning Models — mentioned by belter

Sun, 17 Apr 2022 21:46:49 GMT

"...We show how a malicious learner can plant an undetectable backdoor into a classifier. On the surface, such a backdoored classifier behaves normally, but in reality, the learner maintains a mechanism for changing the classification of any input, with only a slight perturbation. Importantly, without the appropriate “backdoor key,” the mechanism is hidden and cannot be detected by any computationally-bounded observer. We demonstrate two frameworks for planting undetectable backdoors, with incomparable guarantees..." PDF: https://arxiv.org/pdf/2204.06974.pdf

Paper and discussion history. A mention is not an endorsement.

Planting Undetectable Backdoors in Machine Learning Models — HN submission

Sun, 17 Apr 2022 21:46:48 GMT

Planting Undetectable Backdoors in Machine Learning Models

275 points; 59 discussion replies on HN. Paper and source history. A submission is not an endorsement.