water of leith

Bluesky - @srenals.bsky.social
Joined April 2008
I forgot the URL! vacancies.ed.ac.uk/pls/coreh… we are looking for a researcher to join us at CSTR, on a project to learn speech recognition models that learn directly from the waveform
1
4
we are looking for a researcher to join us at CSTR, on a project to learn speech recognition models that learn directly from the waveform
2
9
This beautiful video is something you’d hope to see from @UKLabour. But it’s not Labour. It’s @MagicMagid standing for @TheGreenParty in the European elections. A lot of red votes could go green this month. xitter.cf/MagicMagid/status/1125…
187
1,690
220
5,102
Replying to @loweimi
@loweimi from @InfAtEd talking about DNN statistical interpretation and normalization for ASR
2
4
Very nice short paper with Ben Krause benkrause.github.io, @MannyKayy @driainmurray showing dynamic evaluation can improve Transformer-XL language models arxiv.org/abs/1904.08378
Research note: Dynamic evaluation also works for Transformer models. It improves Transformer-XL language models, achieving new SOTA on enwik8 (0.99 to 0.94 bpc), text8 (1.08 to 1.04 bpc) and WT103 (18.3 to 16.4 ppl). Paper: arxiv.org/abs/1904.08378 Code: github.com/benkrause/dynamic…
1
11
Steve Renals retweeted
"This is not democracy. Spreading lies in darkness, paid for by illegal cash." "Is this how you want history to remember you? As the handmaidens to authoritarianism?" Brilliant and brave @carolecadwalla calls out the "gods of Silicon Valley".
474
6,809
745
14,274
The image analysis #algorithm behind the magnificent #blackHole image is mainly based on a #GaussianMixtureModel #GMM designed for high-dimensional data, namely a mixture of probabilistic PCA! people.csail.mit.edu/klbouma…
2
43
5
111
Steve Renals retweeted
As well as censoring our content over the past few weeks, Twitter have now deleted all the people we were following, which in turn means we have lost a ton of followers Please RT and follow if you’re still right behind us - we have no idea why Twitter is doing this
239
4,088
154
2,199
2018 Writing Reflections: one of the things I enjoyed most was preparing for the @royalsociety diversity conference. I wrote that 'the work of transformation lives in the mundane' and next year hope to expand this consciousness more. (Text and video): blog.shakirm.com/2018/11/con…
3
2
14
we are recruiting academic staff at Edinburgh in a number of areas, including speech technology, computational social science, computational cognitive science, and machinel learning ed.ac.uk/informatics/about/w…
15
9
I should have said the experiment and blog post was done by @siderealed
iPhone vs intel i7 desktop for an SMT solver - homes.cs.washington.edu/~bor… - iPhone has amazing performance! (Via @charlesarthur ). Graph of year on year iPhone improvements is very impressive.
iPhone vs intel i7 desktop for an SMT solver - homes.cs.washington.edu/~bor… - iPhone has amazing performance! (Via @charlesarthur ). Graph of year on year iPhone improvements is very impressive.
1
1
1
3
results of mid-term reviews of EPSRC CDTs obtained and released under FoI by @UKRI_info at ukri.info/part_1_cdt_investi… - very interesting, and some problematic issues raised (although equally the CDT review process wasn't great)
1
1
Everybody In - Crisis campaign to end homelessness in Scotland, England, and Wales. crisis.org.uk/get-involved/e…
4
Thoughtful op-ed by @kaifulee contrasting American and Chinese approaches to AI. He raises imp. questions, e.g., to what extent is discovery science "done"?, and makes interesting observations, e.g. Chinese co. don't feel compelled to "think different". nytimes.com/2018/09/22/opini…
1
16
33