rsfern

60 ms·

571.
by rsfern 6y ago
Well even if you take the most symmetric halves by volume, the density of an ideal elephant is not uniform. Some organs don’t come in pairs and all that I guess it depends on how many hairs you want to split and how precise you want the ans
572.
by rsfern 6y ago
Sure you probably don’t want to do full training runs locally, but There’s a lot you can do locally that has a lot of added friction on a gpu cluster or other remote compute resource I like to start a new project by prototyping and debuggin
573.
by rsfern 6y ago
It’s probably workable in terms of material properties if you can manufacture it in the geometry needed, but I’m not an expert in 2d materials. The edge of a graphene sheet is a defect that affects the electronic properties, with strong ori
574.
by rsfern 6y ago
The choice of publication venue is driven by prestige. Researchers want to publish in “high impact,” high visibility venues. In many fields those are closed journals, but if those journals magically disappeared the research community would
575.
by rsfern 6y ago
I haven’t heard of any reviewer that was compensated for a review, though I once won a reviewer award that had a small honorarium. Editing is generally a paid service — the journal will not correct your typos or your prose. They will find i
576.
by rsfern 6y ago
Well there is pubpeer — https://pubpeer.com/static/about I’m not sure how we get away from the current review system without a coordinated effort involving the prominent scholars in whatever field. The editors are ofte
577.
by rsfern 6y ago
Publish a preprint with Sci-hub links then? It doesn’t seem worth it to have to fight and lose during proofing over what amounts to a DOI anyways
578.
by rsfern 6y ago
This matches my experience as well. It’s really hard to break the habit of typing super hard.
579.
by rsfern 6y ago
Where did you see that they used NAS? Their preliminary results show it works even for the baseline model they did a lot of manual hyperparameter optimization, and spend a fair amount of time unpacking the rationale for their choices, inclu
580.
by rsfern 6y ago
They do a little study in section 4.1 comparing batchnorm to adaptive gradient clipping for resnets over a range of hyperparameters, and they also compare perf to batchnorm versions in table 6. The results indicate AGC does give a real boo
581.
by rsfern 6y ago
If you’re looking to get back into the materials research industry doing applied ML, now is a great time — no need to wait five years. In my experience there’s a lot of demand for people conversant in both fields, a lot of interest in adopt
582.
by rsfern 6y ago
I really like this idea of funding students directly, I feel like it’s pretty well aligned with e.g. the NSF’s mission wrt workforce development, and it reduces the power imbalance for students who find themselves in a toxic group environme
583.
by rsfern 6y ago
This applies to all the traditional retirement accounts invested in index funds as well, so i think your criticism is not specific to this particular program
584.
by rsfern 6y ago
In the same tweet she says she’ll work with (almost) any republican on this issue. Just not Ted Cruz (and presumably Hawley and the others pushing election disinformation). I think your fist point is making this into something more than it
585.
by rsfern 6y ago
This paper seems really interesting, and it’s geared towards applications I guess For more fundamental material I like the CS231n course notes [0] and Goodfellow, Bengio, and Courville [1] 0: https://cs231n.github.io/ 1: h
586.
by rsfern 6y ago
As far as I understand those who were infected and recovered are still encouraged to get the vaccine. I’m not sure what that does to their priority in the vaccination queue though. I think the length of time recovery and vaccination confer
587.
by rsfern 6y ago
So just display a warning/confirmation prompt when the user chooses to use a fixed location rather than real-time location tracking. Or add the configured zip code to the data display. This doesn’t seem like a safety critical feature t
588.
by rsfern 6y ago
Sure, I agree with you, it seems unlikely that fuel cell tech will overtake advances in batteries in the near future, and that compressed hydrogen doesn’t seem likely to become competitive. My point is that there are many other hydrogen gen
589.
by rsfern 6y ago
I think your focus is a bit narrow here, and physical impossibility is a really strong statement. Advances in photocatalytic water splitting could easily make this a reality. That said I’m under the impression that progress in battery tech
590.
by rsfern 6y ago
Depends on the population you’re asking about. The postdocs are generally there to pursue a faculty position or to broaden their skillset and springboard to the next thing. 65k seems high for a postdoc in most fields though. My academic pos
591.
by rsfern 6y ago
While interesting, I think this model is only viable for a certain subset of computational workloads. AF2 took like two weeks of computation on 200 TPUs to train. I’m not sure if they used a data parallel or model parallel distributed train
592.
by rsfern 6y ago
Sounds like ICU capacity is roughly appropriate for normal-times peak demand to me. The average occupancy you quote is consistent with the “2 months ago” capacity quoted by GP.
593.
by rsfern 6y ago
They are estimating infection rates by combining mortality stats specifically for the younger part of the population and seroprevalence studies. So if you accept the known biases in the seroprevalence studies, and make the (strong, IMO) ass
594.
by rsfern 6y ago
Nice! Is there a way to get a link to a specific entry rather than just the full list of query results?
595.
by rsfern 6y ago
Counter counter point: I pretty regularly read pdf docs on my phone, and it’s a horrible experience. I went so far as to write a script that downloads the Tex source of a paper from arxiv, changes up the paper size to match my phone screen
596.
by rsfern 6y ago
I agree that subjective and emotional reasoning are perfectly valid in choosing a tool, but disagree a bit with your take on efficiency. Efficiency isn’t a measure of how long a task takes, it’s about how much additional work is performed r
597.
by rsfern 6y ago
I think it’s more misallocation of training than lack of training per se. Not enough emphasis on deescalation training.
598.
by rsfern 6y ago
I’m not a protein crystallographer, but here’s my generalist take. We understand the physics of e.g. X-ray diffraction pretty well, so we can fit pretty decent forward models for the x-ray data given a proposed structure. The hardest task h
599.
by rsfern 6y ago
Hydrogen storage is definitely not a solved problem. It was a hot funding topic for the US DOE a few years ago, but sufficient energy density and operating pressures are super challenging targets to hit. Efficiency depends on fuel cell type
600.
by rsfern 6y ago
Stockpile stewardship is explicitly one of DOEs purposes, and has been since the Manhattan project really. Weapons research is one of the top level missions of LLNL [0], where the NIF is located. I guess it’s controversial to some, but it’

More ›