I exchange computer vision and machine learning for goods and services. Formerly of @OxfordTVG and @_FiveAI, now on the market.
Oxford, UK
Joined May 2019
- Tweets56
- Following89
- Followers98
- Likes92
New blog post: many deep nets are *very* similar to one another, and we can demonstrate it with a cute transfer attack.
medium.com/fiveai/many-deep-…
For that special person in your life who needs to understand adversarial examples but is always on the go: medium.com/@nick_62384/a-sim…
Nicholas A. Lord retweeted
Check out our big news at Five on joining forces with Bosch linkedin.com/pulse/five-join…
Hi. Do you care about adversarial attacks? Or not?
Either way, check out our #iclr2022 paper, "Attacking Deep Networks with Surrogate-Based Adversarial Black-Box Methods Is Easy".
Paper: arxiv.org/abs/2203.08725
Video: youtu.be/gKzwEq6j5C8
More below. 👇 1/
Median query counts in the single digits for successful attacks, from an algorithm that just alternates between a "harder" and "softer" form of gradient transfer within a standard optimiser. Works nearly every time. 👍 5/
Work done with @RomainMueller and @lbertinetto.
(Really simple) code will be available shortly at github.com/fiveai/GFCS .
Use it on your own networks and tell us what you find. 🙏 6/6
Nicholas A. Lord retweeted
Are you an undergrad/postgrad interested in robotics/computer vision? Applications currently open for an internship in the Scene Understanding team at Five:
apply.workable.com/five-ai-i…
I'm not sure what my favourite part of the review process is, but I think it might be the part where the reviewer plugs their own paper.
I'd been considering writing a blog post about adversarial examples, but then, in the spirit of quick and dirty first-order approximations that get the job done, I just did this instead:
Just stumbled onto this old blog post: colah.github.io/posts/2015-0…
That's the best primer on information theory I've ever seen. There's a lot of insight in those deceptively simple visuals. I learned a few things.
Nicholas A. Lord retweeted
My supervisor Sid has 2 PhD positions available at University of Edinburgh! A great learning opportunity for those interested in probabilistic generative models.
See more details on Sid's website here:
homepages.inf.ed.ac.uk/snara…
Nicholas A. Lord retweeted
Pleased to announce a new paper from a team at @_FiveAI (@fgirbal, Majd Hawasly, Stefano Albrecht @UoE_Agents and myself) on "A Two-Stage Optimization-based Motion Planner for Safe Urban Driving", to appear in @ieeeras Transactions on Robotics.
five.ai/a-two-stage-optimiza…
Nicholas A. Lord retweeted
A great thread by @ericarbailey got met thinking about managing innovation in industry and academia. I tried bringing processes like scrum to research but it didn't work. Breakthrough ideas requires long periods of "wandering in the woods", with no apparent measurable progress.
If you spend most of your day coding to lofi, there are worse places to be than this: lofi.cafe
Nicholas A. Lord retweeted
Interested in adv. robustness beyond security, or how adv. examples can help us understand deep learning? 🤔
Check our review paper “Optimism in the face of adversity”, just published in the Proc. IEEE 🚀
💎Xplore: bit.ly/3qz3men 📄arXiv: bit.ly/3s5Vz8a
We're super nice, also.
If you are looking for a (6- to 12-month) research internship in computer vision / machine learning, our Oxford team is hiring - the call will close at the end of the week
apply.workable.com/five-ai-i…
For a glimpse of our current research: five.ai/our-research
Nicholas A. Lord retweeted
Tired of obsessing over state-of-the-art results? Looking for hypothesis-based science in AI?
Join our #NeurIPS2020 workshop: “The pre-registration experiment: an alternative publication model for ML research" - from 09:15am (EST) on Dec 11th.
Details at preregister.science
