HNHacker News
TopNewBestAskShowJobs

tfgg

1,287 karma · joined May 25, 2013

timothy.green+hellohn@gmail.com
submissionscomments
tfgg··on Why we need to create careers for research software engineers
As a recently departed academic, to vastly generalize, the only people who are research software engineers are perennial postdocs who don't realise yet that their career is a dead end and they're being significantly underpaid for their skill level. They also won't develop their skills adequately in such an environment, lacking good standards and mentorship.

Academia needs to stop taking advantage and accept that it should pay double, so they can actually hire inwards from the private sector and gain good knowledge and skills, or somehow increase the prestige and power to be equivalent to other PI positions. I don't the latter is attainable [1], and I'm sure plenty of people would be happy to just be paid well rather than leave for data science/finance/building web apps.

[1] E.g. If you're in physics, and not doing physics, you're fighting a losing battle to get respect and that lectureship/professorship. Everyone doing coding as their main job has to pretend to do research as well.

tfgg··on TensorFlow Benchmarks
I think it makes more sense that:

a) It's Google, they can throw hardware at a problem such as out of memory

b) TensorFlow makes methods development so much easier that it's worth the loss of performance

c) It's early days and the compute graph scheduler has lots of opportunities, and is designed, for optimization, and in a more flexible fashion than other frameworks.

When I work on developing methods for scientific code, I worry more about whether the code is bug free/easy to understand and that it's giving the right answer. I don't usually worry about performance unless I'm actually not able to run things. Especially since when developing stuff you waste way more time on runs with bugs - I dread to think what my (published / unpublished) CPU hour ratio is. If the new approach allows less buggy implementations then that's a resource win.

tfgg··on Quasiparticles and the Miracle of Emergence
> It seems physicists in general have not had much to say in response to this idea

Huh? My impression is that the notion that the current laws of physics are an effective field theory of some more fundamental theory is pretty widespread and mainstream.

tfgg··on $68B California bullet train project likely to overshoot budget and deadline
I've noticed during the HS2 high-speed rail arguments in the UK that one effect of giving more accurate costing up front (which the UK govt appears to have got better at recently) is that people will just double your numbers anyway, due to long-standing lack of trust. Despite the government giving a very generous £40bn cap on the project, opposition groups still quote £80bn as the final price.

Of course, no-one will actually know if the costing was accurate until it's finished. Does being accurate really pay off for a politician?

tfgg··on How the Internet Has Changed Bullying
> In contrast, it is easier to learn to avoid the chatrooms where the trolls hang out. (The article mentions bullying on your facebook page, but that is very easy to fix with facebook's privacy settings).

Bullying isn't just done by random people, it's also done most significantly by people you know at school, often having greater social power than you. When I was still at school a decade ago, being excluded or bullied by people over instant messenger (e.g. MSN) wouldn't have been something you could ignore or write off as nothing like you can with anonymous trolls, it would have been social death.

I feel that, in the end, and this isn't particularly useful advice to everyone, the strategy is to out-social them by winning over your own group of friends and support group who will choose you over the bully. Or just holding on until you can leave and meet better people.

tfgg··on Trap Streets
I understand the history of copyright, but in general I don't understand why copyright can't be an incentive for governments to create and maintain information products such as maps, just as much as it is for companies/individuals. Without copyright and the ability to charge for use, the government might not want to spend the larger amount of money to create or maintain the dataset.
tfgg··on Trap Streets
In the past I've used a trap unicode character in a name in a public wiki-style CC-BY licensed database of politicians to confirm that a site was (not maliciously) failing to credit us.

This resulted in the person who was using our data reverting the unicode character edit in our database, probably because it was breaking their import scripts.

tfgg··on Trap Streets
While I agree with the conclusion that mapping data should be free, since it's useful infrastructure and will probably help the entire economy more than it costs to make it free, I don't see how your argument necessarily follows.

Governments can own things in a way that doesn't mean that everyone has property rights over it -- just a democratic right to influence how it is used. The government owns the pavement, that doesn't mean I can dig it up and take it away (copyright is different from this, but just to illustrate "government propety" != "everyone's property")

In the case of mapping data, a government can choose the invest money in a product, which it then sells to a subset of the population. Charging means that taxpayers pay less. It also means there are direct customers that can drive the development of the product along the economically highest impact lines.

tfgg··on How to know if where you live is “up and coming”: fried chicken vs. coffee shops
That's the thing, you'd need to demonstrate how "coffee shop / chicken shop" density correlates with house prices over time, and whether it's a predictor or a lagging indicator. There might be better correlators - art galleries and cocktail bars are probably other good ones, or most effects might be mostly priced in, or house price increases might be mostly related to investment in repairing shoddy properties, etc. etc.

I do think it's a cute way for someone who doesn't know London that well to quickly find likely up-and-coming areas, though and a good starting hypothesis.

tfgg··on Modeling Molecules with Recurrent Neural Networks
The list of speakers at the "Machine Learning Methods in Materials Modeling" stream at this recent conference [1] are a few people in the area, though some of those are more predicting properties than making atomic potentials.

I chatted to Gabor Csanyi (Cambridge [2]) during the conference, who I think is probably one of the furthest ahead in the area, and they've recently moved away from their gaussian process based methods to kernel methods. With regard to NNs, he seemed of the opinion that CNNs (the more obvious NN model) were too expensive and ultimately unnecessary compared to carefully chosen, physically motivated kernels. I have to admit I didn't quite understand everything he presented, and I can't seem to find a recent publication, but I'm sure there's one out there.

Despite enthusiastic presentations with lovely results, I suspect from the slow progress in this area that transferability is the main problem plaguing these methods. You want a local atomic potential which doesn't depend on its environment beyond a certain radius, sort of like a convolutional kernel, but a lot of this sort of materials modelling/quantum chemistry is pretty inherently delocalised. Machine learning isn't magic and ultimately has to reflect and represent the underlying physics.

[1] http://nano-bio.ehu.es/psik2015/programme.html

[2] https://camtools.cam.ac.uk/wiki/site/5b59f819-0806-4a4d-0046...

tfgg··on MH17 Report
I think it's just so unlikely that a group that couldn't be held properly responsible would have AA missiles capable of reaching to normal cruising altitude that it's not normally considered a problem and why MH17 was exceptional. Or, at least, that's how it was explained to me by the captain of a 747 while we were flying over Afghanistan.
tfgg··on Quantum effects in relatively large systems
Everything is being conserved perfectly well in MW/UWF, since you're just evolving a pure wave function with the Hamiltonian operator, guaranteeing conservation of everything that's meant to be conserved.
tfgg··on Quantum effects in relatively large systems
I don't think Many Worlds/Universal Wave Function is actually unpopular amongst physicists. There was a survey (I'll have to find it) that showed it was one of the most popular interpretations.

It's appealing for its mathematical simplicity, giving unitary wave function evolution with the reasonable assumption that the universe has a wave function. It eliminates wave function collapse from the axioms, and possibly eliminates the Born rule.

Talking about its 'complexity' in the 'creation' of parallel worlds isn't much of an issue, you can look at simple systems like the von Neumann measurement model to see how they naturally arise through superposition.

The main issue is how the Born rule arises. That is, why the probability of an 'observer' ending up in a particular branch of the wave function is |psi|^2 isn't clear, but then it isn't clear what probability means in this context...

tfgg··on Tales of coding from a beginner
"software engineering" feels broader, like it includes knowing the right aspects of project management and risk for the given project and business needs, beyond just getting something to work.
tfgg··on Linearized map of Lake Michigan: a matter of perspective
Lovely map.

> I once briefly understood why we were doing what we were doing, but that knowledge has since vanished into the ether

Looks like generalized barycentric coordinate over the quadrilateral? Sort of like UV mapping, though it'd be good to hear from someone that knows more.

tfgg··on Where are the missing gravitational waves?
> We are using different definitions of mass. I'm calling invariant energy mass, you are calling rest energy mass.

Modern physics only refers to rest mass [1] as 'mass', otherwise people tend to get confused about how 'relativistic mass' (γm) affects gravity.

[1] E/c^2 in a centre of momentum frame, which doesn't exist for an object travelling at c

tfgg··on Where are the missing gravitational waves?
> I suspect that a combination of those processes adds so much noise that there is no detectable signal left.

You can calculate what the non-linearities do to gravitational waves and, in short, this isn't correct.

tfgg··on Netflix Data Reveals When TV Shows Hook Viewers
Do you think that those design decisions are not backed by data demonstrating better engagement levels versus other interface layouts? I'm not saying that their interface is perfect, but I don't see how you can make sweeping claims about their data science capabilities based on your one data point of disliking their interface.
tfgg··on “On Accident” or “By Accident”?
Perhaps that's the subtle humour behind this linguistic change?

"I broke your mug on accident", at the fifth word, the listener/reader thinks the other person is about to say "on purpose", but suddenly they say "on accident". Hence it's funny, reduces the tension about them breaking your mug.

tfgg··on Neural Networks, Types, and Functional Programming
This is an unformed thought, but to me the philosophy of emergence, intelligence and deep learning have something to do with each other. In emergent systems, you have a stack of levels of phenomena, and you can describe the phenomena at one level without reference to the details of the lower levels. It's like those details have been blurred out and you're left with a compressed representation with just enough information to be useful at the new level (e.g. pixels->edges, renormalization in physics). The success of architectures with multiple layers, including bottlenecks in layer size, seems very similar, and show that this is a good way to look at the world.

This seems to be quite a natural way that humans see the world, and in fact we might not think of something that doesn't behave this way as being intelligent -- we want there to be some sort of compressed 'trick' that doesn't just brute force the data.

tfgg··on Ask HN: Who wants to be hired? (September 2015)
Location: Oxford / London, UK

Remote: Yes

Willing to relocate: Yes

Technologies: Python, C++, Fortran, MongoDB, PostgreSQL, parallel computation, HTML, Javascript, CSS

Email: timothy.green+hn@gmail.com

Website: http://tfgg.me/ (minimalist/woefully underdeveloped)

Github: https://github.com/tfgg

CV: http://tfgg.me/cv.pdf

I'm a computational physicist with a long history of programming and web development looking for new post-PhD challenges, technical and social. By day I work on developing new methods for quantum chemistry in order to solve problems in physics, chemistry and materials science, by night I work on internet-based projects to improve democracy in the UK. I love learning new things, and if you've got interesting data to play with, or just a crazy idea, I'd love to chat.

tfgg··on Virtual particles aren't particles (2011)
Virtual particles are an artefact of the calculation method usually used, perturbation theory, and the Feynmann diagrams that produces. They needn't exist with other methods.
tfgg··on A visual proof that neural nets can compute any function
I don't think so, since the former is about the occurrence of sequences in stochastic, ergodic processes, and the latter is about approximating deterministic functions.

It's more like given an infinite number of parameters, you have enough degrees of freedom to describe the vector space a function lives in with some set of basis functions, and a neural net supplies that.

tfgg··on How baseball’s tech team built HBO Now
Why didn't they contract the BBC? iPlayer works insanely well for live content with live rewind for all their channels, and they covered the 2012 Olympics with rewindable live high-def feeds for every sport.
tfgg··on Long and short scales
I don't know anyone who uses the long scale scientifically or colloquially in the UK except for the occasional out-of-date pedant. Maybe some of the older generation still think it's ambiguous, I think I remember one of my parents commenting on it about 10 years ago.
tfgg··on [dead]
It is a real word, though it's usually used as an adjective[1], and so sounds a bit strange when used as a verb ("to bedraggle")[2], as a native Brit myself, but I get the meaning ("to make bedraggled").

[1] http://www.oxforddictionaries.com/definition/english/bedragg...

[2] I think, my grammar is a bit weak

tfgg··on Magnetism, Radiation, and Relativity (1999)
Huh? A moving charge still has a charge. Imagine a charge density described by the function over space and time rho(r,t). Current is defined by the continuity equation

Del[J] = - d rho(r,t) / d t

(those are partial derivatives). Basically that says that the rate of increase of the charge density inside an infinitesimal box is related to the total current flowing into the box, i.e. if the charge in a box changes, it must have flowed in/out from just outside the box, i.e. continuity (assuming no sources/sinks). Both exist at the same time. In SR, you can look at the relativistic four-current.

Maybe you're thinking of how a current can flow in a wire, despite the wire having no charge. This is because, on our length scales, the electric field of the positively charged atomic nuclei completely cancels the electric field of the negatively charged electrons.

tfgg··on Show HN: Democracy.io by EFF – Write to your representatives
I like their policy of not allowing templated form emails, on the grounds that it just annoys MPs and they're more likely to reply if they're unique and personal. I also like the general philosophy of "You don't need to know which level of government representative you need to talk to, we'll find all of them and try to help you decide".

To agree with another poster on this thread, in the UK there's also the big problem with MPs getting too many emails to sort through them and reply personally, many from large campaign groups such as 38 Degrees. Democracy Club (http://www.democracyclub.org.uk, I'm a volunteer) also found the same problem in feedback from candidates in the 2015 UK General Election, many having to handle thousands of emails (partially our fault for making them open data ;)) on little to no resources or party support. I think there's definitely a space for a modern product which can group together and help a representative/candidate handle replies to different campaigns. This is much better than the alternative, which is making it harder to contact representatives/candidates.

tfgg··on Datasets You've Likely Never Seen
Couldn't it still be the case, though, that we have too much data, but it's also the wrong sort for actionable insights? As a scientist I find the most actionable data are often in the smallest, custom-made datasets, driven by some question, not trawling through masses of data collected without a goal, hoping that they'll have collected the right thing.
tfgg··on Could dark matter not exist?
> A combination of MOND and dark matter matches most observations very well, far better than dark matter in isolation.

Isn't that unsurprising that a theory with more degrees of freedom can fit the data better? It's hard to say if this is interesting without knowing if it fits data significantly enough better to make up for the extra parameters. Dark matter (well, Lambda-CDM) is appealing for being able to fit a wide range of observations with a fairly simple model that's entirely compatible with GR.

← PreviousPage 4 of 8Next →