Daniel Bourke

Announcement:

ZTM’s free week is live!

Take my AI/Machine Learning courses for *free* for the next week!

More details:

• ZTM = Zero to Mastery, the most beginner friendly place on the internet to learn to code (including web, AI, machine learning and more)
• dates: April 14 to April 21 (now extended to April 28)
• cost: $0, no credit card required
• what’s on offer: take all of my courses (as well as every other course on ZTM for free to try things out)
• includes: 120+ courses (from JavaScript to PyTorch to Prompting), 15,000+ videos, access to private Discord for help & guidance

Any further questions, let me know in the comments and I’ll answer them.

Sign up for free here to try things out: dbourke.link/ZTM-Free-Week-0425

1 week ago | [YT] | 88

Daniel Bourke

Machine learning tip:

SigLIP is an absolute goated vision model 🐐

I use it almost every day for zero-shot text-to-image matching.

For example, having a large folder of images and finding out which ones contain “food” using just the search term “a photo of food”.

How?

SigLIP stands for “Sigmoid Language Image Pretraining”.

In a nutshell, it’s a model that learns a joint representation of vision and language so the two become interchangeable.

The best part is that it’s open-source!

That’s why it’s also the vision backbone of many of the open-source VLMs coming out (Idefics3, PaliGemma, HPT-Edge).

For more on SigLIP and how it was trained, I’d highly recommend watching this talk from Lucas Beyer, one of the co-creators of the model (where the slide is from).

(of course read the paper too but I find talks often contain little bits of information you may not pick up on in papers)

One of my favourite quotes from the talk was “the more precise your text, the higher the score”.

This pays big dividends when you’re sorting through a large dataset or doing precise image labelling.

As seen with the goat emojis demo, the more specific the text, the higher the score.

Customer computer vision project workflow example*:

1. Start with large image dataset (e.g. COYO700M/DataComp-1B or larger if you have resources)
2. Embed images with SigLIP -> index with FAISS
3. Define text ontology for precise extraction
4. Filter samples with zero-shot image/text matching
5. Use samples for downstream specific vision task improvement (e.g. to fine-tune a smaller vision model)
6. ????
7. Profit
8. Bonus: Use filtered samples with SigLIP + image caption model + image generation model to generate even more task-specific samples

*should work quite well with any vision task you could reasonably expect to exist on the internet (e.g. if your task requires super specific custom vision data, this may not work as well)

--

Link to talk: https://youtu.be/Nk9YnMHB6hU?si=j-lQ2...
Link to SigLIP model on Hugging Face: huggingface.co/google/siglip-so400m-patch14-384

Note: "goated" = slang for "Greatest Of All Time"

Skip the Carousel

1 2

7 months ago | [YT] | 172

Daniel Bourke

Apple Intelligence: What are your favourite pros and cons?

Here are mine:

Pros:
- On-device machine learning is very cool
- System integration very very cool
- Siri's semantic index sounds like it could be improved over time too (e.g. teach Siri more of your preferences)
- Math notes = 🤯

Cons:
- Image generation on device may be a bit of a toy (e.g. I feel I'd get bored after 3-5 images, genmoji seems cooler)
- How long do these things actually take? Demos seemed quick but I'm always skeptical of demos
- Unavailable on older devices (this kind of makes sense though due to processing requirements)

10 months ago | [YT] | 38

Daniel Bourke

My brother and I have been working on a food tracking and education app called Nutrify.

It uses computer vision to identify foods and track/save their nutrition information.

We focus on whole foods rather than barcodes/foods in packaging.

And version 1.2 just went live on the App Store!

This update includes:
• Calorie and macronutrient goals 🏆
• Calorie and macronutrient breakdowns per category/food 📊
• 57 new foods in the Nutridex/FoodVision AI model 📸 -> 🍍

If you or your friends are looking to learn more about whole foods, be sure to check Nutrify out/share it with them.

Also, if you have any feedback or foods you'd like to see in Nutrify, please let us know!

This is a project I've been wanting to work on for a while which combines health, fitness & machine learning.

You can download Nutrify and start using it straight away on the iOS App Store.

App Store link: apple.co/4ahM7Wc

Skip the Carousel

1 2 3 4 5

10 months ago | [YT] | 111

Daniel Bourke

2024 Eating Simply 🫐

Lunchtime meal:

• toasted chickpeas (excellent source of fibre = good for gut) + drizzle of olive oil (antioxidants + anti-inflammatory, hat tip ‪@BryanJohnson‬ )
• plain biltong (good for protein)
• peach & blueberries for dessert (fibre + antioxidants)

About a handful of each.

4-5 minutes to prepare, fuelled for hours.

My number 1 nutrition tip is to just eat simple whole foods.

Things you could grow/harvest yourself.

Eating well = better performance all round.

Simple foods = delicious + easy to make + better for the environment.

Win, win, win!

--

Photos taken with Nutrify: The Food App

Skip the Carousel

1 2 3 4 5

1 year ago | [YT] | 47

Daniel Bourke

Yo everyone!

If you've enjoyed my YouTube courses on ML, TensorFlow and PyTorch and looking to learn more/show support, the best way is by purchasing a membership on Zero to Mastery and watching the full courses there.

And right now we're offering the best deal of the year: 27% off annual memberships.

Membership not only gets you access to my hands-on AI and ML courses, it gives you access to all of ZTMs courses (including web development + design + cloud & more).

To get the deal, use the code "BFCM23" at any of the following links:

1. Data Science and Machine Learning Bootcamp - dbourke.link/ZTMmlcourse
2. TensorFlow for Deep Learning - dbourke.link/ZTMTFcourse
3. PyTorch for Deep Learning - dbourke.link/ZTMPyTorch

See you inside!

PS if you're in the ZTM Discord, feel free to message me @mrdbourke

1 year ago | [YT] | 47

Daniel Bourke

​‪@AIJasonZ‬ 's channel is an absolute treasure trail of creative uses for LLMs (Large Language Models).

Code & practical examples to go along with!

Highly highly recommend checking out his videos.

1 year ago | [YT] | 9

Daniel Bourke

Epic talk by ‪@AndrejKarpathy‬ at Microsoft Build on GPT and other similar models.

The talk could really be titled "How to train and use ChatGPT" because of the fantastic overview and tips.

Especially the take on "ranking is easier than creating new samples" for updating the models to steer towards human preferences.

I took plenty of screenshots throughout.

Highly recommend for those wanting to know the behind the scenes of ChatGPT-like models.

1 year ago | [YT] | 26

Daniel Bourke

Epic video by ‪@robmulla‬ on how to store data with Pandas for speed/efficiency

I just found out about the difference between:
• CSV
• Parquet
• Feather

Turns out Parquet and Feather can be much faster and have a smaller footprint than CSV files.

If you're saving dataframes with Pandas, check this out:

2 years ago | [YT] | 33

Daniel Bourke

5 years since writing the first line and my first novel is now ready to ship worldwide!!!

The quality is outstanding!!

If you like a story with a little love, a little romance, a little machine learning, a little philosophy and a whole lot of walking, you'll love this book.

It's best thing I've ever made.

Check out the launch video, it's a bunch of fun!

Note: No AI was used in the writing of this book, all hand written, hand typed, hand edited, hand packed.

2 years ago | [YT] | 21