Doug Blank

@dsblank
319 Followers
624 Following
166 Posts
Professor Emeritus, Bryn Mawr College. Now Head of Research at Comet ML. Interested in open social web, machine learning, artificial intelligence, analogies, and robotics.

The history of "Visualizations of Embeddings" in "Archaeology of AI"

#MachineLearning #ArtificialIntelligence #Visualizations #history

https://link.medium.com/5KsP8acB9zb

Compare and contrast #MachineLearning visualizations of multiple models' "attention" using Kangas and its new metric mask.

See the live demo: https://kangas.comet.com/?datagrid=/data/attention.datagrid&select=ID,Image,Model+Name

Try "Group by" ID to compare images; try "Group by" Model Name to compare models.

Open source from the folks at https://comet.com/

Kangas - Data and Model Analysis

Me: Can you give me code that implements a multi-layer neural network?

ChatGPT: Sure...

Me: Did you forget the backward pass?

ChatGPT: Yes, I forgot to include the code for the backward pass in the previous example. Here is an updated version...

I wonder if anyone else in the world still has this kind of merch from the deep dark past of #machinelearning #deeplearning ?

@MichaelGasser were you there? And nice to see you here!

In 1991 I went to my first NeurIPS conference (called something else back then). Who would have thought that 31 years later I'd still be working on my true love. Thanks, https://comet.com/ !

#NeurIps conference starts tomorrow.

#deeplearning #machinelearning #datascience

Home

Comet provides an end-to-end model evaluation platform for AI developers, with best-in-class LLM evaluations, experiment tracking, and production monitoring.

Comet

Something I've been working on, released today, Kangas DataGrid: a #Python tool for exploring multimedia datasets in a tabular format.

https://github.com/comet-ml/kangas

Some interesting technology in here I'll be posting about.

#foss #openscience #opensource #datascience

GitHub - comet-ml/kangas: 🦘 Explore multimedia datasets at scale

🦘 Explore multimedia datasets at scale. Contribute to comet-ml/kangas development by creating an account on GitHub.

GitHub

A real Scheme implementation (call/cc, choose/fail, etc) that can use Python libraries in @JupyterProject --- but also has stack traces or infinite loops!

https://github.com/Calysto/calysto_scheme

GitHub - Calysto/calysto_scheme: A Scheme kernel for Jupyter that can use Python libraries

A Scheme kernel for Jupyter that can use Python libraries - GitHub - Calysto/calysto_scheme: A Scheme kernel for Jupyter that can use Python libraries

GitHub

tensorflow: Assembly Language for #DeepLearning

keras: C++ for #DeepLearning

conx: Python for #DeepLearning

conx.readthedocs.io/en/latest/

I now have a basic stub of an #activitypub microblog.pub Python library working with choice of database (mongo, sql, redis, etc) and choice of framework (flask or tornado):

https://github.com/dsblank/activitypub/tree/master/apps/blog

How could a generic AP library be useful?

dsblank/activitypub

activitypub - A general Python ActivityPub library

I have an #activitypub library that now supports many SQL databases (sqlite (in-memory and on file), MySQL, PostGreSQL, etc), MongoDB, and Redis. Now I'm starting to work on a multi-framework web framework. Targeting Flask and Tornado to start...