chetan_v

"The contract of sale contains a confidentiality provision that protects information relating to the buyer and price." Wow that sucks, would have been quite interesting to know how the government found a buyer and how much it was sold for.

Thanks for the question, unlike sci-kit learn we have focused on building better hyper parameter tuning and continuous streaming unsupervised learning rather than one time batch unsupervised learning, this allows us to much better track patterns over time. We also have several models not on sklearn and even for the ones that are there, we have reimplemented them with better GPU support. You can also find some more info on that here : https://www.all.vision/technical-whitepaper

Hi HN! We are Joe and Chetan, co-founders of All Vision Technologies (https://all.vision/). We met when working at Capital One and noticed tooling around unsupervised learning for building and deploying models left a lot to be desired. As a result unsupervised learning seems to be very underutilized in industry.

We left Capital One to work on this problem full time and have been working with great customers like the US Space Force. We just launched our Unsupervised Model Zoo, which runs and compares 12 clustering and anomaly detection algorithms. Also, we implemented cluster tracking over time to show things like conceptual drift or changing populations in your dataset.

This tool with one click lets data scientists clean data (automatically!), set up cloud and processing around their models, explore differences between clusters, and work faster.

What I'm proudest of: We rebuilt many of these algorithms from scratch, achieving up to 10x speedups in comparison with open-source implementations like Nvidia Rapids, you can read more about how we did this in our technical paper here: (https://www.all.vision/technical-whitepaper).

Welcome to our first milestone in our journey of building powerful and easy to use unsupervised learning tools.

We have a free trial for anyone who signs up from this post. To get started, make an account at (https://all.vision/) and upload any unlabeled dataset to find patterns inside your data set.

Seems you can see some of them on their company linkedin page : https://www.linkedin.com/company/anthropicresearch/about/

Looking at the team seems to be all ex-openai employees and one of the cofounders worked on building gpt3. Will be exciting to see what they are working on and if it will be similar work to openai but more commercialized.

Seems like a cool product, congrats on the launch!

Thanks for asking : So, first with scikit learn, it doesn't work on spark and other distributed frameworks. It's also a bit slower, doesn't provide gpu support for fast inference, can't support larger datasets easily, and runs fewer models than our tool.

The AllVision platform and dev tools tests and runs 10+ unsupervised models at once, so we automate the work needed to find the best model for your dataset. Under the hood, we redesigned many of those unsupervised models so they run significantly faster, so teams can optimize hyperparameters very quickly (hours instead of days), and we're building it on a series of API's so it's more easily deployed within a large scale system.

All Vision Technologies | New York City | Full Time Onsite (Remote during covid)| https://all.vision/ | Frontend and Backend Engineers

All Vision Technologies provides infrastructure to optimize the quality of data for training machine learning models. We enable data scientists and AI/ML software developers to build models that account for hidden causal features, that incorporate native mechanisms for improving data labels, and that exhibit peak accuracy and performance.

We currently have openings for both Front end and back end engineers. We use React on the front end and flask/fastapi on the backend, pytorch for machine learning and docker for deployment.

If you are interested apply at https://angel.co/company/all-vision-technologies/jobs. Can also send any questions about the position or company to chetan@all.vision

First Nobel prize for AI from this?

Yeah it seemed kind of strange to me that Data Science and R&D are both not considered highly independent, I guess on some teams there is a more collaborative atmosphere but most of the data science work I have done seemed pretty independent. The people skills I agree with also, can't see how any data scientist could get by without strong people skills to communicate their results with leaders and to collaborate with subject matter experts.