yoav_hollander

11 ms·

61.
by yoav_hollander 10y ago
A few comments on this: First, as etendue says, it is not easy. The problem of mixing “Boolean” verification with probabilistic, less-deterministic verification is especially hard. I discussed this a bit in [1], if you care to take a look.
62.
Using Machine Learning to Verify Machine Learning?(blog.foretellix.com)
2 points by yoav_hollander 10y ago|0 comments
63.
Using Machine Learning to Verify Machine Learning?(blog.foretellix.com)
2 points by yoav_hollander 10y ago|0 comments
64.
by yoav_hollander 10y ago
Right. This paper ([0]) is actually mentioned in the DARPA BAA ([1]) as an example of a possible direction. A somewhat-similar scheme is [2]. Both seem to do some kind of sensitivity analysis, so as to show the user which parts of the input
65.
by yoav_hollander 10y ago
If there is real progress towards Explainable AI, this would also be very useful for _verifying_ machine-learning-based systems (i.e. finding the bugs in them). I wrote about this in [1], but I am not a machine-learning expert (I am coming
66.
by yoav_hollander 10y ago
Well yes, you _can_ verify ML-based systems (and I discuss briefly how in the linked article). It is just that this is much harder with such systems than with HW or SW systems for which "you have the source code". What I meant by
67.
Machine Learning Verification and Explainable AI(blog.foretellix.com)
3 points by yoav_hollander 10y ago|0 comments
68.
by yoav_hollander 10y ago
I think this issue of "Explainable Machine Learning" and interpretability is just going to get more and more important as ML grows. It will also be important for verifying ML-based systems - another problem area. See [1] for a dis
69.
Machine Learning Verification and Explainable AI(blog.foretellix.com)
2 points by yoav_hollander 10y ago|0 comments
70.
by yoav_hollander 10y ago
I like the idea of using fuzzing "to show a direct correlation between programs that score low in their algorithmic code analysis and ones shown by fuzzing to have actual flaws". I hope they'll use AFL, and publish the parame
71.
The Tesla crash, Tsunamis and spec errors(blog.foretellix.com)
1 points by yoav_hollander 10y ago|0 comments
72.
Trying to understand automation and unemployment(blog.foretellix.com)
1 points by yoav_hollander 10y ago|0 comments
73.
Finding bugs in autonomous vehicles(blog.foretellix.com)
2 points by yoav_hollander 10y ago|0 comments
74.
Finding bugs in autonomous vehicles(blog.foretellix.com)
3 points by yoav_hollander 10y ago|0 comments
75.
Automation and unemployment are coming, but I guess we’ll manage(blog.foretellix.com)
2 points by yoav_hollander 10y ago|0 comments
76.
The next 20 years of automation(blog.foretellix.com)
2 points by yoav_hollander 10y ago|0 comments
77.
The next 20 years of automation(blog.foretellix.com)
1 points by yoav_hollander 10y ago|0 comments
78.
by yoav_hollander 10y ago
See also [2] for a description of the verification challenges of a fictitious (but informative) Amazon last-mile autonomous delivery vehicle. [2] https://www.cs.york.ac.uk/ftpdir/reports/2015/YCS/496&#x2F
79.
by yoav_hollander 10y ago
"Last mile" delivery may perhaps be done with much smaller, mostly-autonomous vehicles (with remote operators getting involved only when needed). See related article and thread in [1] - article also discusses some startup ideas fo
80.
The rise of mostly-autonomous systems(blog.foretellix.com)
37 points by yoav_hollander 10y ago|6 comments
81.
The rise of mostly-autonomous systems(blog.foretellix.com)
3 points by yoav_hollander 10y ago|0 comments
82.
Is a Universal Verification Framework possible?(blog.foretellix.com)
1 points by yoav_hollander 11y ago|0 comments
83.
by yoav_hollander 11y ago
I think you are absolutely correct. It is the price/performance of creating these "interactive specs" that will determine whether this idea is any good. From what I have seen (I looked a bit at the what people are doing in UA
84.
by yoav_hollander 11y ago
Right. Cutting-edge military systems are, indeed, risky business, and one has to balance (as you say) more testing against not having those systems when you need them. This puts a high premium on the efficiency of bug-finding, especially sp
85.
by yoav_hollander 11y ago
Absolutely. What I meant was that there should have been some "are you sure" confirmation in this case.
86.
by yoav_hollander 11y ago
To me, the really-interesting question here is "How can we find such issues in the design before they hit". In the article, I explain why this is going to be a bigger problem in the near future, and speculate that some simulation-
87.
by yoav_hollander 11y ago
To Matthew Wilkes: It is true that in military situations the tradeoffs of safety vs. utility are quite different. However, I think this is one of those (many) cases where it is an "operator error" which a better design could have
88.
by yoav_hollander 11y ago
The archived Washington Post article is still here: https://web.archive.org/web/20130624110018/http:/www.gpsnavi...
89.
It’s the spec bugs that kill you(blog.foretellix.com)
67 points by yoav_hollander 11y ago|38 comments