andy99

I think the most common HN use case (mine anyway) would be to filter out topics that I don't want to see.

If you couldn't guess by the headline > Walmart said the primary use of the phone is for its new employee app called Me@Walmart, which the company explained in a release is meant to "simp…

I like the idea of tags, but it introduces extra work in terms of monitoring for abuse, arbitrating disagreement, and depending on the tags could even introduce additional editorial bias. I'm cur…

The notebook is to compare the count of terms found in the hiring and seeking posts. I pre-populated the terms with a list of mostly front end technologies.

I think knowing what you don't know is still a pretty big win. It can help people trust the models in cases they do work, and it can serve as a diagnostic for why it fails in certain circumstance…

You've probably seen it, but a more recent, related paper (that I think has some of the same authors) about inherent features of modern ML that make models so fragile, even if they test OK: Under…

For common neural network models, the output probability has no meaning for out of distribution inputs, so you need to do an ensemble or some other method to get at the actual model confidence. I don&…

The point of the article is more that even if all of your testing and validation is rigorous and the performance looks great, trivial changes in the production data can break your model anyway. My vie…

SEEKING WORK | Canada (Eastern Time) | Remote first, potential for some time on premise subject to current restrictions Specialized in turnkey deep learning R&D, focusing on uncertainty quantifica…

Your example is good, but I would argue it points to a good confidence (uncertainty) estimate as the most important element of trusting a model, rather than a human-interpretable explanation. Like you…

The concept of calibration - actually being right at the rate that matches your estimated percent confidence - is important for machine learning. It's important to know how well you can trust the…

Uncertainty quantification and OOD detection in machine learning. It's on some people's radar, but has the potential to get ML adopted much more widely as people understand what it is actual…

SEEKING WORK | Canada (Eastern Time) | Remote first, potential for some time on premise subject to current restrictions Specialized in turnkey deep learning R&D, focusing on uncertainty quantifica…

Cool! I'll have to take the opportunity to mention my own charts.css[0]. It got some interest on HN last year [1] and I unfortunately dropped the ball and didn't make any revisions based on …

Thanks for the reply! I'm actually working in a different domain but it seems to have a lot in common with yours - lots of unlabelled data, images that have nothing in common with Imagenet, in th…

Very cool. Do you have lots of unlabelled data, and if so, do you do any self-supervised pre-training? Have you ever considered releasing the backbone weights for the pre-trained models you have? No i…

I made a comment on this too, they are not presenting it in the best light because their example has a finite pool of data so using all of it will give the best performance. You can see the active lea…

I have a suggestion about the first plot you show in the writeup. From what I can see, it is based on a finite pool of data and so it undersells active learning: performance shoots up as AL finds the …

I'd add that there is a deep connection between active learning and understanding the "domain of expertise" of a model, for example what inputs are ambiguous or low confidence, and whic…

I have a playbook - there is so much to like about it, in particular that it runs QNX and with a terminal emulator (I'd have to look up the name) you have a very cool terminal based tablet, even …

I'm curious to know the answer to this question as well. I have done a fair bit of work working with organizations to identify ML use cases. When we looked at it from a business process perspecti…

This is a point worth emphasizing. So many proposed "AI" use cases insist on trying to make some kind of end-to-end prediction, rather than let ML do something it is well positioned to do (e…

Here is one of mine: https://github.com/rbitr/ivc Version control for ipython notebooks, the goal is to be able to use a Jupyter notebook in a hacky, iterative way, but track you…