meta-code-verify VS FastTreeSHAP

(2022)

https://github.com/facebookincubator/meta-code-verify is the goods, and is MIT

https://github.com/facebookincubator/meta-code-verify#instal... says Safari support is "coming soon" (from 2022) so I guess they think those users don't need to "verify[..] the integrity of a web page."

Code Verify - A browser extension from Meta for verifying the integrity of web pages and detect executed code that’s not included in the site manifest.

> The server can at any time start serving malicious payloads

True, and I call this threat model "Beware Each and Every Fetch" (BEEF) in contrast to the more common TOFU model (although if you trust a desktop app to auto-update itself then these two models might not be all that different).

In any case, I think you're being a little quick to dismiss the idea of server-hosted applications. It's true that browsers don't natively have a nice way of pinning specific versions of a web app, but there is the clever hack of SecureBookmarks[0] (if you're prepared to sacrifice the UX), or, more realistically, you can pin the web app version using some sort of browser extension.

Examples of the latter include the Signed Pages extension[1], and Code Verify[2], which is the result of a collaboration between Meta and Cloudflare (for securing the WhatsApp Web code, currently, but should eventually support other sites like Proton's too). Of course, it would be much better if this capability was natively included in browsers themselves, but hopefully adoption of this technology will pressure browsers and standards bodies to take ownership of this.

[0] https://coins.github.io/secure-bookmark/

[1] https://github.com/tasn/webext-signed-pages

[2] https://github.com/facebookincubator/meta-code-verify

FastTreeSHAP - A Python package from LinkedIn for fast interpretation of the TreeSHAP algorithm.

LinkedIn open sources the FastTreeSHAP Python package for efficient interpretation of tree-based ML models using SHAPLEY. FastTreeSHAP v2 would be 2.5x faster than TreeSHAP. Let's reminder that SHAP (SHapley Additive exPlanation) values quantify the contribution of each feature to the model prediction, a bit like how each player contributes to the success of a sports team. SHAP does it by incorporating concepts from game theory and local explanations. LinkedIn blog post, scientific paper, and GitHub repo with IPython Notebooks.