danielmorozoff

The MACAW model is quite impressive -- it significantly outperforms GTP3 on Q&A/reasoning tasks (~10%) while requiring x10 less parameters. It is based on T5 , which is a well known model from Google. The novel innovation here is the training paradigm.

Also did I mention it's OSS: https://macaw.apps.allenai.org/

Here is the paper: https://arxiv.org/abs/2109.02593

pretty much everything. Connectome != functional understanding of the brain. We have had c elegans ( worm) and more recently fly connectomes for years. we are still struggling to understand basic logic encodings in those animal models. Imho we lack a foundational understanding of the logic encoding mechanisms in the brain. Many neuroscientists/ computer scientists are working on this problem, but to my knowledge we are still not there.

I personally like gpustat -- it's a nvidia-smi wrapper but it has colors...

They also i guess now have a web sever plugged into it which seems pretty cool

https://github.com/wookayin/gpustat https://github.com/wookayin/gpustat-web

If you’re looking to index/ process video - maybe we can help. Checkout Vidrovr (https://vidrovr.com)

Full disclosure im one of the founders.

Here's the paper: https://arxiv.org/abs/2101.03961

This was taken down because it was labeled as Show HN.

https://news.ycombinator.com/item?id=25825332

Multi post:

1st Post: https://danmorozoff.substack.com/p/ai-reversion-to-the-old-d...

2nd Post: https://danmorozoff.substack.com/p/ai-reversion-to-the-old-d...

How can I resubmit -- it links back to this?

Should be at the end of the 1st article.

Multi post

1st Post: https://danmorozoff.substack.com/p/ai-reversion-to-the-old-d...

2nd Post: https://danmorozoff.substack.com/p/ai-reversion-to-the-old-d...

This is a blog I began writing trying to distill my observations around running a business in ML/CV to learnings. Both technical and business related posts. Comments and thoughts are always welcome.

Article in the Lancet: https://www.thelancet.com/journals/lancet/article/PIIS0140-6...

Here is the actual paper: https://arxiv.org/abs/2008.06007

I am scribd user. Recently, I discovered Libby -- free app from the public library (SFPL and NYPL for me). Libby is x10 better experience than scribd, is free and has all books you may find in the library.

The benefits of scribd is if you're an avid reader you can consume more than 3 books/audiobooks at a time, which is a restriction on Libby- requires you to return them etc.

For those interested: https://www.overdrive.com/apps/libby/

Just figured I would post as I have been incredibly impressed with Libby.

In case people are interested in the paper https://d4mucfpksywv.cloudfront.net/emergent-tool-use/paper/...

Vidrovr | Computer Vision/ ML Engineer | Flatiron New York, NY; | ONSITE or REMOTE, FULL-TIME | $100,000-120,000 + Equity | https://angel.co/company/vidrovr/jobs/555397-machine-learnin...

Vidrovr build video understanding systems- effectively building gen 2 machine learning/visual systems. It is funded by VCs, NSF, and the DoD. If you are looking for difficult modeling problems in various areas (NLP, CV, graph fusion, GANs etc) drop us a line

We're looking for senior engineers for positions listed bellow:

  - Various ML areas
  - Python ML systems (Pytorch, Tensorflow)

For all employees, we offer:

  - We are ok with remote folks.
  - Unlimited vacation
  - Benefits (medical/dental)
  - Other perks (gym memberships)

If you want to learn more, reach out to us at recruiting@vidrovr.com or dan.morozoff@vidrovr.com

Vidrovr Inc. | Production ML Engineer | New York | Onsite/Remote

We currently have an opening for a ML engineer (Intermediate/Senior).

We work on video understanding - the intersection of state of the art computer-vision, ML and knowledge graph generations. We are funded by VCs, the NSF and the DoD.

If you are interested in working on hard problems with a tight knit group of researchers/engineers focused on video understanding. Please reach out to me with your CV: dan.morozoff@vidrovr.com

You can find out about us here: https://vidrovr.com

We are funded by the nsf sbir and I must say if it wasn’t for this mechanism our company would not exist. Effectively it funds reach science that can have translational applications. The value of the sbir is to get you to mvp ( phase I) and scale out to prove traction in phase II

Effectively it provides capital when vcs aren’t willing to either because you’re too early from rev numbers or they just lack the alignment on vision/market or some other reason

Can someone comment on performance -- speed vs training vs other frameworks. Most of our work is in TF/Pytorch and I have not used any JS implementations would love to hear some peoples insights into how it compares.

Omg yes ! I was confused having read the first paragraph and did not understand why it was titled this way.

Awesome did not know this thanks!

I think expertise is assessed not by just using, but building or fixing things that are broken. If a toy model/example are good enough for you then sure, use your LSTM / keras implementation. But if you're faced with your model not working in your usecase -- what do you do? To answer that question and create a plan to tackle that problem requires knowledge and experience

I think you misunderstood me.

I do not believe "you need to understand all these deep and hard concepts before you start to touch ML." That is a contortion of what I said.

First point: ML is not a young field- term was coined in 1959. Not to mention the ideas are much older. *

Second Point: ML/'AI' relies on a slew of various concepts in maths. Take any 1st year textbook -- i personally like Peter Norvig's. I find the breadth of the field quite astounding.

Third Point: Most PhDs are specialists-- aka, if I am getting a PhD in ML, i specialize in a concrete problem domain/subfield, so I can specialize in all subfields. For example, I work on event detection and action recognition in video models. Before being accepted into a PhD you must pass a Qual, which ensures you understand the foundations of the field. So comparing to this is a straw man argument.

If your definition of ML is taking a TF model and running it, then I believe we have diverging assumptions of what the point of a course in ML is. Imo the point of an undergraduate major is to become acquainted with the field and be able to perform reasonably well in it professionally.

The reason why so many companies (Google,FB,MS etc) are paying for this talent, is that it is not easy to learn and takes time to master. Most people who just touch ML have a surface level understanding.

I have seen people who excel at TF (applied to deep learning) without having an ML background, but even they have issues when it comes to understanding concepts in optimization, convergence, model capacity that have huge bearings on how their models perform.

https://en.wikipedia.org/wiki/Machine_learning *https://www.amazon.com/Artificial-Intelligence-Modern-Approa...

We use rocks in production and research and have used LMDB for research purposes. Here's a benchmark:

https://www.influxdata.com/blog/benchmarking-leveldb-vs-rock...

I think rocks gives better read+ write performance together whereas lmdb is heavily skewed to read performance in our experience -- which is mirrored by the benchmark.

I'm a bit concerned that this will be too superficial and not really cover ML or the mathematical foundations well. ML and 'AI' are very multifaceted fields that require a strong foundation in mathematics. As such, the lack of information theory, signal processing, fourier analysis, abstract algebra (from a mathematical standpoint) not to mention CS courses is disheartening. This seems like a major tailored to the strong need for ML talent / skyrocketing job prospects/salaries as opposed to fundamental understanding- basically trying to attract students. For example the whole concept of Tensorflow is the combination of high dimensional tensors with the ideas of abstract algebra -- hence the name.

ML/AI imho should be broached at the Masters level for those who are interested as they have strong foundational knowledge.

Is this actually true?

I have wondered for years... apologies for the poor grammar

I have been wondering why this hasn't existed for years! Thank you guys for making this. Long awaited.

I was just in Toronto. It could also be because your buying power is significantly more than in the valley. You make more but everything (house or rent) is also more...

This is a very interesting notion, could you elaborate why convexity would be a candidate? Convexity defining a class is clear, but as a proxy for interpretability I have not heard this before. I would assume this would need some measure of degree of convexity - different than strong > strict > convex