tbenst

10 ms·

91.
by tbenst 6y ago
Karl has said publicly a few times that he sees optogenetics as a tool for basic scientific discovery, that then inform treatment of human patients, ie where to use transcranial magnetic stimulation. Any method that requires genetic manipul
92.
by tbenst 6y ago
You need an actuator to induce current into the neuron. In this case, a light-sensitive ion channel, which have proved highly effective.
93.
by tbenst 6y ago
Great question! See the supplementary note on tissue heating: https://web.stanford.edu/group/dlab/media/papers/chenNBT2020...
94.
by tbenst 6y ago
I sit next to Ritchie Chen, the first author on this paper. Let me know if any questions and I’ll try to get them answered :) Edit: for anyone that wants to read the paper, here’s a non-paywall link: https://web.stanford.edu&#x2F
95.
by tbenst 6y ago
I’m reminded by the stories in 2016 on how the Pope has endorsed Donald Trump. Was it true? Not in the slightest. And yet these stories got millions of impressions. Enabling the spread of harmful, false, foreign-planted propaganda to spread
96.
by tbenst 6y ago
This equivocation of news sources is how misinformation spreads. Look at how WSJ handled same story vs NYP. WSJ is indeed more reputable as is NYTimes
97.
by tbenst 6y ago
I mean, Microsoft claimed in 2016 to surpass human performance and google subsequently claimed in 2017, and yet anyone that regularly uses voice dictation knows that <5% WER is far from realistic due to lack of generalization to new audi
98.
by tbenst 6y ago
How about 80% accuracy on CIFAR-10 with unsupervised training? => logistic regression + Kmeans http://ai.stanford.edu/~acoates/papers/coatesleeng_aistats_2...
99.
by tbenst 6y ago
To be fair, most of the machine learning literature had no or a poor excuse for peer review. And many deep learning layers can be described by dense layers. For example, convolution. You almost certainly can frame Go playing as polynomial r
100.
by tbenst 6y ago
That's a strong unsubstantiated claim. On the other hand, there's been some nice theoretical work arguing that deep learning is a form of polynomial regression. https://arxiv.org/abs/1806.06850
101.
by tbenst 6y ago
Maybe it does something, but tinnitus tends to simple improve over time, too: [1]. Again not a great study as unclear what treatment options people sought, but they have zero evidence of efficacy as I understand it. It’s inaccurate to call
102.
by tbenst 6y ago
I’m a neuroscientist and just read their “phase 2b protocol” paper. As far as I can tell, there are no actual control groups. Registering a trial with clinicaltrials.gov is a nice step, but they don’t compare at all to patient improvements
103.
by tbenst 6y ago
Out of curiosity what’s the alternative? I thought it’s amazing that you can customize the font at all
104.
by tbenst 6y ago
Dvorak is 25% more efficient than Qwerty using some very reasonable calculations: http://mkweb.bcgsc.ca/carpalx/?dvorak
105.
by tbenst 6y ago
Every 3-6 months I try a new project in Flux and it’s definitely catching up quickly! The neural ODE integration is super cool for example. There are also other applications where you might need to calculate higher order gradients or even d
106.
by tbenst 6y ago
The biggest change by far is to process not code: https://www.python.org/dev/peps/pep-0602/ I have mixed feelings about a 12 month development cycle. From the perspective of a library maintainer but not a cor
107.
by tbenst 6y ago
As always, benchmarks depends on your application! It appears that for my workflow, this drive is a substantial _regression_ from the 970 Pro. I have a firehose of data that needs to be stored without dropping samples for a scientific appli
108.
by tbenst 6y ago
It’s also possible that tests are simply a very poor way to assess knowledge/aptitude. I certainly felt this way when taking the SAT/GRE in the US at least.
109.
by tbenst 6y ago
I love Altair but the practice of encoding data alongside the plots makes it unwieldy for sharing jupyter notebooks (by default a 5000 row limit). Github also fails to preview Altair plots in .ipynb right now
110.
by tbenst 6y ago
> NVIDIA GPUs offer up to 8x more half precision arithmetic throughput when compared to single-precision, thus speeding up math-limited layers Right but I’ve benchmarked the best case scenario, ie a large GEMM call in C++, and still not
111.
by tbenst 6y ago
Thank you for finding that! I’m glad to see the situation has improved since I last looked. 3x improvement over fp32 is impressive for sure. Their marketing claims of 8x still bother me though.
112.
by tbenst 6y ago
Thank you for cross-referencing that, the data does look accurate and my statement now seems exaggerated. I do think we need skepticism on the A100 charts though until third party benchmarks. > In my experience NVidia benchmark numbers i
113.
by tbenst 6y ago
Thank you for clarifying! I’m still skeptical of the chart’s A100 values but appreciated your reasonable attempt to de-bias. It’s always easier to critique then create so I also want to make sure I complement you on an excellent article :).
114.
by tbenst 6y ago
I appreciate the time and care that went into this post, and there’s a nice discussion of various features. Unfortunately the performance charts are completely devoid from reality, and in particular the discussion on tensorcores may be true
115.
by tbenst 6y ago
Agreed. The question I meant to raise was nuclear power versus other alternative energy sources, eg wind/solar/hydropower/geothermal. Comparisons of nuclear power to these other sources are typically dishonest in that they do
116.
by tbenst 6y ago
I’m continuously baffled by those that claim that nuclear power is safe. Look at Fukushima for example. It seems hard to imagine a nuclear power plant design, even Gen III or IV that would be immune to any possible natural disaster. Analy
117.
by tbenst 6y ago
Could someone explain key differences between Parity and GPL? I haven’t encountered this before
118.
by tbenst 6y ago
Super neat! It’s like a mash up of Excel and Jupyter, using something that looks similar to functional reactive programming
119.
by tbenst 6y ago
I’ve been very pleased with VegaLite. My one complaint is the use of column for the grouped bar chart. Unfortunately as a result it’s difficult to have such a chart in a FacetGrid.
120.
by tbenst 6y ago
I was a purescript developer and Haskell dabbler way before I was an investor, for what it’s worth, and happy to engage on the substance of my points if you would care to have a collegial discourse befitting of academics!

More ›