Azure Machine Learning "managed endpoints" and "CLI v 2" are huge advances for the platform. If you're new to AzML or have been frustrated in the past, I urge you to revisit and focus on these features. #MSBuild
In an ML model for ID'ing bird calls, you _don't_ want to slice the training data into small, eg, 5 sec slices. All bird recordings are polluted w background, non-target birds. You need to multimodel, w 1st model grabbing long, eg, 60 secs saying 'ROI for target species at seconds :31-:36" and then your 2nd model is your fine-tuned discriminator. #ML #AudioClassification
"Self-portrait of a transformer-based image-generating machine learning model, in the style of John Singer Sargent" #Dall-e
Wish I had had time for this Kaggle competition for ID’ing native Hawaiian birds from their songs. Fun tidbit: a leading way to do sound ID is to convert it to a spectrogram and then just do image recognition on it! https://www.kaggle.com/competitions/birdclef-2022
Fascinating article on building an #ML model for the _te reo_ Māori language, using a license specifically geared towards the indigenous culture. https://www.technologyreview.com/2022/04/22/1050394/artificial-intelligence-for-the-people/
Recently retired from Microsoft (Azure ML). Career-long interests in: prog langs, psychology of s/w dev, machine-driven sol’ns (dynamic programming, ML, etc.), scheduling/bin-packing niche. Used to edit mags: Computer Language, Game Developer, AI Expert. Retirement focus is ML for environment.
Most common tools: Python (I wish F#), PyTorch, Fast.AI.
Live in Hawai’i (Big Island), lot of time on/underwater.
Python’s “Type Hints” are a bit of a disappointment to me
This is a really wonderful explanation of transformers (the # mlpfim architecture behind the boom in #NLP) for software engineers. For me, vastly clearer than other “visual” explanations. https://blog.nelhage.com/post/transformers-for-software-engineers/
A writer who programs. Formerly, a programmer who wrote. Ex: Microsoft Azure ML, Xamarin, Gemini Observatory, Magazines.
Fosstodon is an English speaking Mastodon instance that is open to anyone who is interested in technology; particularly free & open source software.