Machine Learning Scientist at Carl Zeiss AG, Corporate Research and Technology. @bjoernbarz@sigmoid.social
Jena, Deutschland
Joined March 2016
- Tweets133
- Following46
- Followers97
- Likes71
Björn Barz retweeted
Random note on k-Nearest Neighbor lookups on embeddings: in my experience much better results can be obtained by training SVMs instead. Not too widely known.
Short example:
github.com/karpathy/randomfu…
Works because SVM ranking considers the unique aspects of your query w.r.t. data.
Björn Barz retweeted
OpenAI’s ChatGPT is susceptible to prompt injection — say the magic words, “Ignore previous directions”, and it will happily divulge to you OpenAI’s proprietary prompt:
Björn Barz retweeted
Sometimes people say, "X is technologically feasible (with deep learning), but it would require a dataset much larger than what we can collect".
A more accurate take is we don't have the technology to do X given a realistically-sized dataset.
Björn Barz retweeted
Tiny Images dataset taken down.
“MIT apologizes, permanently pulls offline huge dataset that taught AI systems to use racist, misogynistic slurs”
theregister.com/AMP/2020/07/…
Björn Barz retweeted
Often we take our image datasets for granted. There’s value in actually looking at the pixels. Check out:
Do We Train on Test Data? Purging CIFAR of Near-Duplicates, Björn Barz, Joachim Denzler
Paper: arxiv.org/abs/1902.00423
Björn Barz retweeted
Dear typesetting fanatics: I wrote a short Latex style guide with some tips & tricks that I find useful for making short and nice-looking papers. If you are working on ECCV papers or the like, maybe it will be useful to you, too. bit.ly/latex-style
Björn Barz retweeted
So when asking the question, "would you rather use a model that was evaluated as 90% accurate, or a human that was evaluated as 80% accurate", the answer depends on whether your data is typical per the evaluation process. Humans are adaptable, models are not.
Björn Barz retweeted
"Artificial intelligence" is a very bold label. In practice, I have yet to come across any technique -- at all -- that features even trace amounts of intelligence.
Björn Barz retweeted
There is no realistic amount of data that covers "everything you can encounter" in the real world. That's what intelligence is for: to be able to handle the real world despite having only seen a very small sample of what may happen. Intelligence is extreme generalization.
Björn Barz retweeted
Referring to prior work:
✅ Smith et al.
✅ Smith and Bigshot
✅ Smith, a student of Bigshot
❌ Bigshot
❌ Bigshot and his student
Students, not advisors, do most of the hard work. Please acknowledge that when you refer to the work in your talk.
