HNHacker News
TopNewBestAskShowJobs

w_t_payne

1,510 karma · joined May 23, 2012

blog.wtpayne.info
submissionscomments
w_t_payne··on Nvidia to Acquire Arm for $40B
Sounds like we need someone to found another ARM.
w_t_payne··on My Diplomatic Life
Still with their mother on posting. I had a breakdown and returned to the UK about 18 months ago.
w_t_payne··on My Diplomatic Life
Most diplomatic careers deliver rather less in the way of storytelling materiel than the author's. Nonetheless, I'm glad that I am no longer a diplomatic spouse. A life in which we regularly uprooted ourselves every few years to move to a new country was never one that I wanted, and one that I am more than happy to leave behind. (Although I do miss my children terribly, and wish there were some way I could maintain contact with them).
w_t_payne··on Tech firms face growing resentment of parent employees during Covid-19
As a parent, I fear that even in the best of times, parenthood is simply incompatible with the level of time and effort that working in tech demands.

I really wish I knew how to resolve this problem, but it feels more like a physical problem, to do with the limits of human attention and learning, than a social one. I suspect there is no easy fix.

w_t_payne··on XLS: Accelerated HW Synthesis
I've got a Kahn-process-network based "simulation" framework, intended to provide a smooth conveyor belt of product maturation from prototypes written in high level scripting languages like Python or MATLAB through to production code written in C or Ada. (Sort of like Simulink, but with a different set of warts). Having some hardware synthesis capability is very much on the roadmap, and this looks like it's going to be worth investigating for that. Very excited to dive into it!
w_t_payne··on Python setuptools v50 breaks pip installation on Debian/Ubuntu
And this is why I vendor every single dependency manually and build them all from source.
w_t_payne··on Lithium ranked the highest performance web application platform
110 ms is not faster than human perception. It's over a tenth of a second, and is very noticeable. The flicker fusion threshold for human vision lies in the 60Hz-90Hz range, corresponding to between 11 and 17ms. If I recall correctly, EEG experiments show differences up to around 100-120Hz, so on the order of 8ms. In any case, network latency itself adds on the order of 10ms per 1,000km of distance; 35ms or so for a London to New York round-trip, so the point about framework speed being a bit of a moot point other than in exceptionally high traffic scenarios still stands.
w_t_payne··on Build tools around workflows, not workflows around tools
Unix had it right. Provide a toolbox with modular legos that you can customise and fit together to make the tools that you need.
w_t_payne··on Ask HN: Examples of good technical design docs
Here's a good resource that I find useful:- http://everyspec.com/USAF/USAF-SMC/SMC-S-012_16JAN2015_52130...
w_t_payne··on YC Startup School: Build Sprint and Equity-Free Grants
Startup school has been amazing for me ... it's really helped to give the psychological push to get going and keep going. I think it's really incredible that it exists, is free, and is as useful as it is.
w_t_payne··on Ask HN: What is your current side project?
The side project is an implementation of a general purpose autonomous problem solving system and decision support toolset based on rules based techniques for governance of machine learning systems that we are developing. We see it as an approach to AGI that will facilitate rules based governance. ("Asimov's three laws" is our elevator pitch).

The idea has evolved over the last few years during the course of regular weekly discussions held by our group of founders.

Our group is planning on turning our technical developments into a startup, probably initially focusing on systems engineering tools.

We are currently implementing our prototype and looking at potential application areas.

w_t_payne··on Starship SN5 150m Hop
Is the engine supposed to be on fire like that?
w_t_payne··on Starting a Business Around GPT-3 Is a Bad Idea
I suspect that 6 months from now the situation will be different, and technology will have moved on. The potential for these very large models is just too tempting, and I can see a lot of different organisations piling on the bandwagon.
w_t_payne··on Starting a Business Around GPT-3 Is a Bad Idea
One thing that I think GPT-3 does show is that NLP technology has now reached a level of sophistication where it becomes possible to use it in a wide range of new and different applications.
w_t_payne··on Ask HN: What's the worst piece of software you use everyday?
+1 for Teams.

It's horrible.

w_t_payne··on Yoloface-500k: ultra-light real-time face detection model, 500kb
You might want to investigate the Movidius Neural Compute stick (which you can use with a RPi), or the Nvidia Jetson Nano, which has a lot more oomph.
w_t_payne··on Synthetic Data for Deep Learning (2019)
In my mind it's because you have a more direct route to the underlying physics of the problem than either hand-coding features or trying to learn them from real data (at least when that real data is as poorly managed and controlled as it usually is).

Our ML algorithms are really good at finding correlations -- but we don't necessarily know if the correlations in our data are actually the ones we want our system to learn. When we're using synthetic data, we have many more levers at our disposal to ensure that this is the case.

w_t_payne··on Synthetic Data for Deep Learning (2019)
I can't see your email address in your profile. :-( Mine is my username (without the underscores, all one word) at gmail dot com.
w_t_payne··on Synthetic Data for Deep Learning (2019)
Well, in a sensing-for-autonomous-vehicles type problem, it's actually more important to have simple and easy to specify data distributions than ones which map to reality, which in any case may be so poorly or incompletely understood that it's impossible to write the requirement for.

So, as a simple example, the illumination in a real data-set might be strongly bimodal, with comparatively few samples at dawn and dusk, but we might in a synthetic dataset want to sample light levels uniformly across a range that is specified in the requirements document.

Similarly, on the road, the majority of other vehicles are seen either head-on or tail-on, but we might want to sample uniformly over different target orientations to ensure that our performance is uniform, easily understood, and does not contain any gaps in coverage.

Similarly, operational experience might highlight certain scenarios as being a particularly high risk. We might want to over-sample in those areas as part of a safety strategy in which we use logging to identify near-miss or elevated-risk scenarios and then bolster our dataset in those areas.

In general, the synthetic dataset should cover the real distribution .. but you may want it to be larger than the real distribution and focus more on edge-cases which may not occur all that often but which either simplify things for your requirements specification, or provide extra safety assurance.

Also, given that it's impossible to make synthetic data that's exactly photo-realistic, you also want enough variation in enough different directions to ensure that you can generalize over the synthetic-to-real gap.

Also, I'm not sure how much sense the concepts of mean and variance make in these very very high dimensional spaces.

w_t_payne··on Synthetic Data for Deep Learning (2019)
There's also some really interesting research from last year on local dimensionality of a manifold that I really really want to try out.
w_t_payne··on Synthetic Data for Deep Learning (2019)
I don't actually think you want to mirror the natural data distribution, but rather to provide a distribution which has a sufficiently high variance in the right directions so that the resulting NN polytope has a chance of being approximately 'correct'.

Because you have this piecewise-linear sort of warping of the feature space going on, the NN is basically a whole bunch of lever-arms. The broader the support that you can give those lever arms, the less they will be influenced by noise and randomness ... hence my obsession with putting enough variance into the dataset along relevant dimensions.

To put this another way, I think that the synthetic data manifold has to be 'fat' in all the right places.

w_t_payne··on Synthetic Data for Deep Learning (2019)
The GTA V dataset doesn't really have enough modes of variation, in my opinion.

To my knowledge, the best public attempt at this is represented by the OpenAI Rubik's Cube / Shadow Robotics dexterous hand demo. https://openai.com/blog/solving-rubiks-cube/ https://arxiv.org/abs/1910.07113

NVIDIA are also doing some interesting work in this area, but again, I'm not really sure they put enough different modes of variation into it. https://research.nvidia.com/publication/2018-04_Training-Dee....

CVEDIA also get really impressive results using similar techniques:- https://www.cvedia.com/

w_t_payne··on Facial Recognition Leads To False Arrest Of Black Man In Detroit
Well .. it looks like it isn't enough, and additional human-factors oriented requirements need to be documented - like deliberately engineering the system to produce a certain number of false alarms so the police always have to do additional work to discard them - as opposed to only sometimes.
w_t_payne··on Synthetic Data for Deep Learning (2019)
From the perspective of using synthetic imagery to train machine vision systems, I think that the idea of fidelity (i.e. how similar synthetic images are to real images) is less than half the story, and has the potential to be dangerously misleading.

Of greater concern are quality measures that look across the entire dataset. Here are some hypothetical metrics which (although impossible to compute in practice) will help get you thinking in the right way.

- How does the synthetic image manifold compare to the natural image manifold?

- Are there any points on the synthetic image manifold where the local number of dimensions is significantly less than at the corresponding point on the natural image manifold? (Would indicate an inability to generalise across that particular mode of variation in that part of feature space).

- For each point on the synthetic image manifold, are there any points where the distance between the synthetic image manifold and the natural image manifold is large AND the variance of the synthetic image manifold in the direction of that difference is small. (Would indicate an inability to generalise across the synthetic-to-real gap at that point in the manifold).

- Does your synthetic data systematically capture the correlations that you wish your learning algorithm to learn?

- Does your synthetic data systematically eliminate the confounding correlations that may be present in nature but which do not necessarily indicate the presence of your target of interest.

Engineering with synthetic data is not data mining. It is much more akin to feature engineering.

w_t_payne··on Synthetic Data for Deep Learning (2019)
I'm very interested in synthetic imagery in the maritime domain, and would be very interested in talking to you, if at all possible.
w_t_payne··on Synthetic Data for Deep Learning (2019)
The point is to be able to control the correlations that you introduce - so you can therefore control what your learning algorithm learns.
w_t_payne··on Synthetic Data for Deep Learning (2019)
For machine vision problems, synthetic data with domain randomisation definitely works, provided, that is, you are able to generate a large enough number of different modes of variation.

In fact, it works so well it feels like cheating.

w_t_payne··on Putin's People – a groundbreaking study that follows the money
That's a very interesting perspective. I shall have to ponder on that for a while.
w_t_payne··on Facial Recognition Leads To False Arrest Of Black Man In Detroit
Perhaps we, as technologists, are going about this the wrong way. Maybe, instead of trying to reduce the false alarm rate to an arbitrarily low number, we instead develop CFAR (Constant false alarm rate) systems, so that users of the system know that they will get some false alarms, and develop procedures for responding appropriately. In that way, we could get the benefit of the technology, whilst also ensuring that the system as a whole (man and machine together) are designed to be robust and have appropriate checks and balances.
w_t_payne··on The Antagonism of Human and Nature in Factorio
I read it as a very cynical message about the inevitable destructiveness and cruelty of human nature.
← PreviousPage 5 of 32Next →