Share this page!

Enter URL or ID to Unroll

You can paste full URL like: https://x.com/threadreaderapp/status/1644127596119195649
or just the ID like: 1644127596119195649

How to get URL link on X (Twitter) App

  1. On the Twitter thread, click on or icon on the bottom
  2. Click again on or Share Via icon
  3. Click on Copy Link to Tweet
  4. Paste it above and click "Unroll Thread"!
  5. More info at Twitter Help
Daniel Bourke Profile picture
@mrdbourke
Jun 8, 2021 • 10 tweets • 6 min read • Read on X
Scrolly
Machine Learning Monthly for May 2021 is live (video & audio)!

The latest and greatest (but not always the latest) from the machine learning world in the past month + plenty of dancing.

This month we've got...
Huuuuuuge updates to @TensorFlow:

• TensorFlow Lite models now work with TensorFlow.js (train once, deploy twice)

• Google's on-device machine learning page tailors ML guides for your smaller device needs

• TF Lite model maker library helps you train on-device models faster
• TensorFlow Hub gets a facelift, plus, now you can try pretrained models before you buy them (jk the models are free)

• TensorFlow Cloud library helps you scale up your smaller experiments to cloud-scale in a few lines of code (e.g. Google Colab -> 8 GPUs) Use TensorFlow Cloud to scale up your machine learning model
• Google Cloud's AI Platform gets renamed to Vertex AI and now Google Cloud's one-stop-shop for your ML needs (think data storage, feature storage, model training, model deployment etc)

• To go along with Vertex AI is a new MLOps White Paper piecing together everything ML MLOps process
• The new TensorFlow forum! Now there's a town square to meet and talk with TensorFlow developers from around the world

• The People and AI Guidebook 2.0 helps you design ML-powered applications by thinking about things like: "explain the benefit, not the technology" TensorFlow Forum, the new town square for TensorFlow develop
And from the rest of the internet:

• Next-generation pose detection with MoveNet and TensorFlow.js (17 body keypoints @ up to 51 FPS in the browser of an iPhone 12!!!)

• Datasets & code are on arXiv ala @paperswithcode (find the data and code associated with ML papers) Code & data on tab on arXiv
• @facebookai's wav2vec-U (unsupervised) speech recognition model performs equivalent to state of the art 2 years ago without *any* labelled data (previous model used ~1000 hours)

• What is active learning? by @roboflow - doing practical ML? You'll want active learning
• Reproducible Deep Learning by @s_scardapane - Ever tried to build a reproducible deep learning model? It's harder than you think. Not to worry, Simone's course goes through steps to help you do so.

See more: sscardapane.it/teaching/repro… Reproducible Deep Learning workflow
• The Rise of @huggingface by @marksaroufim - an outstanding take on how ML companies like HuggingFace and @weights_biases have built incredible value by creating community around their product offerings. The rise of HuggingFace by Mark Saroufim
Far out...

As usual a massive month on tour for the world of ML!

See the full write up: zerotomastery.io/blog/machine-l…
See the video version:

• • •

Missing some Tweet in this thread? You can try to force a refresh
　

Keep Current with Daniel Bourke

Daniel Bourke Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @mrdbourke

Daniel Bourke Profile picture
Dec 3, 2024
New video: Tracking every item in my house with video using Google Gemini 🎥 -> 🛋️

I call it "KeepTrack" 😎

Input: 10-minute casual walk around video.
Output: Structured database w/ 70+ items.
Cost: ~$0.07 w/ caching, ~$0.10 w/o caching.

Full details... 🧵
TL;DR

Gemini features which make this possible:

1. Video processing.
2. Long context windows (video data = ~300 tokens per second, 10 minute video = 165,000 tokens).
3. Context caching (process inputs once, inference for 4x cheaper).

Prices are with Gemini 1.5 Flash. Bar chart comparing the total costs of using cached versus non-cached input tokens for processing a large video file three times. The cached method costs $0.0741937875, which is 17.46% cheaper than the non-cached method costing $0.08989155
1. Video Processing/Prompting

Intuition: Show *and* tell.
Technical: Video = 4 modalities in one: vision, audio, time, text (Gemini can read text in video frames).

Instead of writing down every item in my house, I just walked through pointing at things and talking about them.

Gemini tracked everything I said/saw almost flawlessly (it missed a few things due to 1 FPS sampling but this will get better).

Doing this via text/photos alone would've taken much longer.

There are many more fields/problems where video input unlocks a whole new range of possibilities.Screenshot of a documentation page for Google AI's Gemini API, focusing on prompting with video. It includes technical details for supported video formats, such as MP4, MPEG, MOV, AVI, and others, and mentions that the File API is required for uploading large video files.
Read 12 tweets
Daniel Bourke Profile picture
Dec 10, 2021
If machine learning projects were a relationship...

Data collecting and processing is the dating phase, fun, chaotic, up and down, tormenting and carefree, seeing if you're a good fit.

Modelling is the wedding day, takes forever to plan, over before you know it.
People using your model is the honeymoon.

Then comes the data drift.

Your data changes like the person you thought you married, maybe they're getting fat (distribution changes) or they're finding it hard to love you (your data features are no longer ideal).
So you bring in data monitoring, model evaluation (marriage counselling) and pull all the tricks.

Your marriage counsellor tells you to go back to what got you started.

The fun dates (collecting data), talking for hours learning about each other (processing data).
Read 7 tweets
Daniel Bourke Profile picture
Jul 6, 2021
Friends, the latest issue(s) of Machine Learning Monthly (June 2021) have been delivered beautifully to your inbox/YouTube subscription page.

Quick takes below.
1. Text-based Video Editing

This is wild.

Not too often things make me go "Woah".

This did.

I'd never even thought of this kind of editing.

Chop and change videos using pure text transcriptions.

• Project page: davidyao.me/projects/text2… Text-based video editing
2. Cracking the Machine Learning Interview

The wonderful @chipro has done it again with an outstanding book on the machine learning interview process.

Getting a job is a skill in itself.

Lucky you've now got Chip's book to help you.

• Read it free: huyenchip.com/ml-interviews-… Chip Huyen's online machine learning interviews book
Read 5 tweets
Daniel Bourke Profile picture
Apr 27, 2021
Outstanding post by @marksaroufim

He put into words something I’ve been thinking but didn’t quite know why.

@huggingface & @weights_biases are two of my favourite ML companies.

Why?

Because like @fastdotai...

They create community.
I’d also add @roboflow into the mix of my favourite up and coming ML companies.

People like people.

Roboflow are making things and sharing ideas directly from the engineers/founders.

It’s good to relate to the people behind the product.
Not to mention the memes throughout this post are worth their weight in gold.

This one describes perfectly describes my last 3 years online (except replace Twitter w/ YouTube).

Note for myself going forward: leverage product off media/community base. Gaussian plot with disturbed ML product manager at the mean
Read 5 tweets
Daniel Bourke Profile picture
Jul 31, 2020
1/ Introducing the 2020 #machinelearning roadmap:

An interactive mindmap which connects many (not all) of the most important concepts in machine learning.

Map: dbourke.link/mlmap
Video walkthrough:
Accompanying slides: github.com/mrdbourke/mach… machine learning mindmap th...
2/ In the map you'll find 5 branches:

1. 🤔 Problems - some of the main use cases for ML.
2. ♻️ Process - what does a solution look like?
3. 🛠 Tools - how can you build your solution?
4. 🧮 Math - ML is applied mathematics, what kind?
5. 📚 Resources - where to learn the above.
3/ Although very colorful, at first glance, the map can be very intimidating.

So there's a video walkthrough to go along with it:

We start with a high level overview which answers questions like "what is machine learning good for?" what is machine learning go...
Read 13 tweets

Did Thread Reader help you today?

Support us! We are indie developers!

This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Don't want to be a Premium member but still want to support us?

Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal

Or Donate anonymously using crypto!

Ethereum

0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E copy

Bitcoin

3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi copy

Thank you for your support!

Follow Us!

Send Email!

:(