denhaus
Example: Consider what PDB has done for our understanding of protein folding, as well as the ML/computational techniques they’ve enabled (eg, Alphafold). Most scientific questions and properties are not as data-rich as protein folding. What if they could be?
Longer answer: The last 15 years in computational/ML + science have shown that structured databases open up entirely new frontiers in discovery (eg Protein Data Bank, Materials Project). But most scientific topics/properties are NOT in structured DBs, they’re scattered about in millions of papers. It’s especially a huge problem in some topics in materials science. It’s not that these problems are data scarce, but that it’s hard to actually collate their data in a structured format. You literally cannot use most ML methods because structured DBs do not exist.
This paper is a way to generate massive structured databases of specialized, intricate, and hierarchical knowledge graphs from scientific literature. Fine tuning works, prompt engineering does not (at the time, perhaps this has changed). Once you have a database, you can analyze an entire subfield or topic in science with ML or stats methods.
However, the alternative of using, for example - experimental data, is that the synthesis procedures, measurement parameters, sample impurities, and even differences between experimental apparatus means training datasets of even modest size are insanely heterogeneous. So models either are either trained to predict differences between materials due to experimental discrepancies, trained on very small datasets, or must have a slew of post-hoc physics-based adjustments added to get reasonable numbers.
Higher order computational methods (including simply more intensive, non-high throughput DFT) are accurate but expensive as you know. Some of them have systematic error in the way DFT does, and are essentially based on user choice of (many!) parameters. Charged defect calculations are on example of this. Finding large (>10^4) training sets with similar parameters for computation is difficult. “ML” for these kinds of calculations usually consists of like, calculating a hundred (or 10) crystals within a narrow chemical system, doing a linear regression on one variable (eg, valence of cation on some site), and getting numbers +\- 10% of a “true” number.
GGA/meta-GGA DFT, on the other hand, can be applied at a sufficient fidelity to get real(ish) numbers in a homogenous way across huge numbers of crystals. So you are correct, you are predicting an approximate number for a property in many cases. But if we know the approximate number is wrong due to systematic error (and we can, in some situations) we can apply corrections or higher order methods to get the right(ish) answer. More, it’s highly dependent on which property you’re interested in. Some properties, like band gap, can be off by a lot. Others, like formation energy, can be calculated pretty accurately even with run-of-the-mill GGA DFT. Elastic moduli are generally ok.
in summary, approximating DFT with ML is just the least messy way to get real-ish answers across a large number of materials. Of course, there’s a point at which low-fidelity DFT calculations are - (1) so cheap and (2) so inaccurate, generally - that having an ML model approximate them is pointless. Most large DBs of materials now use good enough DFT that the numbers they calculate are not pointless for ML to learn from.
In the future, I think models trained on large numbers of DFT calculations will have to be applied to narrow sets of higher fidelity calculations by tuning. Much like you can fine tune a generalized LLM to do specific things. That might be where ML can actually bring real value to materials design.
Also, it’s worth considering that synthesizing novel materials can be insanely difficult. So 1 in 4 is not bad in my opinion.
Point being, I’ve never found sharp edges matter much for most recreational riding. None of this has any relation to being uncool. Maybe in your wifes scenario, having razor sharp edges would have saved her from breaking her arm; almost certainly not.
On true ice, NO ONE is ripping it except racers or ex-racers with good equipment.
On the real however, getting down the mountain safely after dusk etc with dull vs. sharp edges will likely only affect intermediate skiers. Beginner skiers are going to crash no matter what they’re riding if on steep and icy terrain. Expert skiers know when and how to ride conservatively and can basically ride anything in any conditions “safely” (even if that means just sliding a firm patch rather than carving it), as long as they’re aware of the limitations of their gear. 90% of the year I ride pow skis in any conditions (including melt freeze etc) with super dull edges - it’s totally fine. The other 10% is just to have a little more fun on very firm days.
Intermediates, on the other hand, will be overly aggressive beyond their capabilities. They’ll bounce their helmet off a melt-frozen knoll at first opportunity, similar to what you said!
Nothing factual presented. Not a unique, stimulating, or even really coherent viewpoint. Just a self-confident stream of consciousness with no real takeaway.
Eye-catching title though
good luck, jake. may your days be filled with appreciation and love and tranquility.
Anyways props to Sinead for doing the work here. Having a theoretical understanding of why LK99 behaves why it does (if it does) will be important for its optimization and commercialization. Briefly reading over the methods it seems(?) like it was quickly done with a GGA functional (given there is a +U correction) rather than a hybrid functional so this was probably not “heavy compute” as the tweet author would have you believe. OP it may be productive to just link directly to Sineads publication rather than this tweet
For reference I also work at LBL on computational materials, I have met Sinead on several occasions and she is very legit.
- where do you go, and why?
- how do you get your US insurance to cover/work with providers outside the US? does it just automatically work or…?
I just envision a future where there is some near-circular dependency of passwords/phrases/notifications/authenticators/keys/email verifications etc across different devices and services - the end result is that it is absolute PiTA to log into anything or recover any account if anything is ever lost. Sort of an endless personal bureaucracy for authentication. It’s a future I am personally trying to avoid at all costs
Yubikeys etc seem like something I could potentially get behind, but it still doesn’t seem perfect persay… anyway, maybe I am just a geezer