razcle

9 ms·

31.
There's a better way to calculate your ML test metrics(humanloop.com)
3 points by razcle 4y ago|0 comments
32.
by razcle 4y ago
Reading this atm. About half way through and already it's one of my favourite books. Would love to contribute to the notes if you're accepting PRs
33.
Insight to Tesla Engineering Culture(twitter.com)
2 points by razcle 4y ago|1 comments
34.
Some Intuition for Jensen's Inequality(razcle.github.io)
1 points by razcle 4y ago|0 comments
35.
Annotated Data for Social Good(lacunafund.org)
1 points by razcle 4y ago|0 comments
36.
by razcle 4y ago
Hi Raza here, one of the other co-founders. I know that HN likes to nerd out over technical details so thought I’d share a bit more on how we aggregate the noisy labels to clean them up. At the moment we use the great Skweak [1] open source
37.
Huawei's chief scientist on machine translation(chinai.substack.com)
2 points by razcle 5y ago|0 comments
38.
by razcle 5y ago
Humanloop | Infrastructure for AI | Backed by YC and Index | London + Remote Hiring - Software Engineers: front-end specialist - Machine Learning Engineer - Interaction designer (see jobs.humanloop.com for full details) We're a team of
39.
AI Doesn’t Have to Be Too Complicated or Expensive for Your Business(hbr.org)
1 points by razcle 5y ago|0 comments
40.
by razcle 5y ago
Hi all, I wrote this piece and will be around for the next hour or two if anyone fancies a chat about GPT-3 and large scale language models!
41.
Modeling libraries don’t matter (2020)(shreya-shankar.com)
31 points by razcle 5y ago|3 comments
42.
by razcle 5y ago
No, not 100% accuracy. I've left out details for the sake of brevity but with a precision and recall high enough for the team to be able to answer the questions they cared about.
43.
by razcle 5y ago
Maybe so but most data science workflows still don't acknowledge this "obvious" truth.
44.
by razcle 5y ago
I very nearly said this myself! I think the mistake of this quote is in the application of the expertise. The bitter lesson is that data + compute can outperform inductive biases but that doesn't mean you don't need domain experti
45.
by razcle 5y ago
I think this is one of those points that is obvious in retrospect but almost universally under appreciated. Almost all data science workflows treat the annotators or subject matter experts as secondary. The tooling isn't set up to put
46.
by razcle 5y ago
We see ourselves as quite different to Scale really as we don't provide annotation services, mainly the software. One of the main differences is that we've pretty exclusively focussed on language rather than vision which has quite
47.
by razcle 5y ago
Hi Andy, thanks for the feedback on the site! We're actually redesigning at the moment so it should hopefully be fresher soon :P. Also great pointer to Rob Munroe's book. He actually used to be CTO at figure 8 before they were acq
48.
by razcle 6y ago
One of the makers of Humanloop here. We have recently started trialling our active learning tech with real customers and are excited to share the results. Happy to answer any questions!
49.
Measuring Active Learning Performance in the Real World(humanloop.com)
6 points by razcle 6y ago|1 comments
50.
by razcle 6y ago
Raza here (one of the founders of Humanloop). Just wanted to share that you can now train any Huggingface model on your own data just by labelling and curating a small portion of that data, with Humanloop. Humanloop provides a labelling int
51.
Humanloop and Hugging Face=the fastest way to build an NLP API on your own data(humanloop.com)
3 points by razcle 6y ago|1 comments
52.
by razcle 6y ago
Humanloop | Infrastructure for AI | Backed by YC and Index | London + Remote Hiring - Software Engineers: full stack and front-end - Machine Learning Engineer (see jobs.humanloop.com for full details) We're a team of ML researchers and
53.
by razcle 6y ago
Raza here (one of the founders of Humanloop). Just wanted to share that you can now train any Huggingface model on your own data just by labelling and curating a small portion of that data, with Humanloop. Humanloop provides a labelling int
54.
Humanloop and Hugging Face=the fastest way to build an NLP API on your own data(blog.humanloop.com)
5 points by razcle 6y ago|1 comments
55.
by razcle 6y ago
Yeah I think is a good point. I'm actually planning to do a follow up post that is a case study with some real world data and the plots in that are much more like what you describe. I may update the post. thanks!
56.
by razcle 6y ago
TL;DR I think its complementary. Synthetic data is particularly valuable when even the unlabelled data is expensive to obtain. For example if you want to train a driverless car, you may never see an ambulance driving at night in the rain ev
57.
by razcle 6y ago
So entropy based active learning methods are an example of pool based sampling. Even within pool based sampling there a few different techniques. Entropy selection for pool based methods looks at the output probability for each prediction o
58.
by razcle 6y ago
Great point Andrew. I was shooting for an easily digestible example rather than a realistic one. Some examples that we've actually worked on/are working on: * Contract classification * Content moderation * NER * Customer review un
59.
by razcle 6y ago
Hi lwhsiao, Raza here, author of the post. My high-level answer is weak-labelling overcomes cold starts and active learning helps with the last mile. More detail: We see weak learning as very complementary to active learning. By using label
60.
by razcle 6y ago
I'd also point out that people always focus on just the labelling savings from active learning but there are other benefits in practice too: 1) faster feedback on model performance during the annotation process and 2) Better engagement

More ›