miniblog.

← Back to all posts
Aug 28, 2021 at 21:27
It's a bittersweet feeling when you trace a crash back to a "TODO: make this robust against X" comment you wrote.

Related Posts

Jan 20, 2020 at 14:33
Invited talk: Safety Verification for Deep Neural Networks: https://popl20.sigplan.org/details/VMCAI-2020-papers/22/Safety-and-Robustness-for-Deep-Learning-with-Provable-Guarantees How do we verify that a DNN is robust to adversarial attacks? How do we quantify safety? This approach looks at image features (Sift) and verifies all perturbations within a region.
Safety and Robustness for Deep Learning with Provable Guarantees (VMCAI 2020 - - 21st International Conference on Verification, Model Checking, and Abstract Interpretation) - POPL 2020
Welcome to the website of the 21st International Conference on Verification, Model Checking, and Abstract Interpretation (VMCAI 2020). VMCAI provides a forum for researchers from the communities of Verification, Model Checking, and Abstract Interpretation, facilitating interaction, cross-fertilization, and advancement of hybrid methods that combine these and related areas. VMCAI 2020 will be the 21st edition in the series. Proceedings https://link.springer.com/book/10.1007/978-3-030-39322-9 Invited Speakers We are pleased to announce the the invited speakers for VMCAI 2020 are: Rajeev ...
Jul 4, 2019 at 00:09
Commonmark.js has a lovely feature that's rare in markdown renderers: it exposes an AST! https://github.com/commonmark/commonmark.js/#usage This makes it so much easier to extend/modify the syntax in a robust manner.
GitHub - commonmark/commonmark.js: CommonMark parser and renderer in JavaScript
GitHub - commonmark/commonmark.js: CommonMark parser and renderer in JavaScript
CommonMark parser and renderer in JavaScript. Contribute to commonmark/commonmark.js development by creating an account on GitHub.
May 13, 2019 at 19:08
Really interesting paper exploring adversarial inputs to ML models: https://arxiv.org/abs/1905.02175 They conclude: * It's a property of the input data, not the training * You can even train a model on non-robust features and obtain a model that works well on the original input data!
Adversarial Examples Are Not Bugs, They Are Features
Adversarial Examples Are Not Bugs, They Are Features
Adversarial examples have attracted significant attention in machine learning, but the reasons for their existence and pervasiveness remain unclear. We demonstrate that adversarial examples can be directly attributed to the presence of non-robust features: features derived from patterns in the data distribution that are highly predictive, yet brittle and incomprehensible to humans. After capturing these features within a theoretical framework, we establish their widespread existence in standard datasets. Finally, we present a simple setting where we can rigorously tie the phenomena we observe in practice to a misalignment between the (human-specified) notion of robustness and the inherent geometry of the data.