Excited to announce we've raised a Series C at $1 Billion valuation to continue building our developer-first MLOps platform.
Checkout this post by
@l2k
for details:
I want to take a moment to thank the people who've made this possible.
We're thrilled to launch Paper Projects – an opportunity to implement, build on top of & apply interesting ML papers.
We've picked 9 ML papers for you to creatively explore. You can work in teams or alone.
#machinelearning
#deeplearning
#100daysofmlcode
My favorite linear algebra for ML resource is this blog. Mainly because it comes with code, which for me makes the math so much more intuitive! 🙂
The intro to SVD is *chefs kiss*
Excellent set of class notes for Stanford's CS224 (Machine Learning with Graphs) course by Anil Karaka!
If you're looking to get started with Graph Nets, this is a great place to start!
We strongly believe that ML research should be reproducible & accessible to all.
So we’re excited to offer support to participants of the Reproducibility Challenge in form of $500 to help with computation costs, & a slack community where you can collaboratively reproduce papers.
My team is hiring!
👩🔬 ML Engineers
👨🔬 Project Managers
Perks include: working on insanely exciting/challenging problems, on a product ML engineers love, with the smartest/kindest/most fun folks in the world
DM me!
#deeplearning
Something that came up today in a convo with friends:
It's very difficult to compete with someone who is having fun with their work.
You can try to emulate them, but it's hard to do for 12 hour days/weekends. The difference is: to them they are playing & to you you're working.
Introducing one of our most highly requested features – the Confusion Matrix!
You can now log a multi-class CM in one line of code, & compare model predictions across classes.
🪄 Log a confusion matrix in 2 mins
#deeplearning
#machinelearning
Totally forgot about this – I have some notes on Machine Learning, linear algebra, stats, data wrangling et all. They might turn out to be useful to some people here.
#MachineLearning
#DeepLearning
#100DaysofMLCode
Every day when I go to Fully Connected, I'm blown away by the quality of intermediate/advanced deep learning tutorials we're publishing!
If you haven't bookmarked it yet, you definitely should! 😊
🐝
Building a neural network can be confusing! I wrote a guide to help you navigate the treacherous NN waters using
@weights_biases
🧜♀️
Highly rec forking the kernel & playing with the code!
Post:
Code:
#MachineLearning
#DataScience
Deep learning gets all the fancy tools.
So we made our friends who scikit a tool to measure & compare your model performance, with one line of code.
Try it here:
It was such a blast working on this! 😊
#MachineLearning
#DataScience
#100DaysofMLCode
I really hope this Siraj revelation incites a collective return to what matters most — technical skill and integrity.
We’re here because we’re trying to push the boundaries of machine learning, and build wicked cool things.
Less hype around ML is a great thing in my opinion! 🙌🏼
In my post on Siraj and how he got to this point on Friday morning, I expressed hope that his livestream was a chance for him to come clean and start over. Welp.
I've written a couple of Kaggle kernels over the last year. I think folks here might find these helpful 🐦
Short thread –
1. Getting Started with a competition
Understand your dataset inside out + feature engineering
#kaggle
#machinelearning
#datascience
This is a great collection of resources on deep learning, and other subfields including graph neural networks, bayesian deep learning, medical imaging etc.
Today, we’re so excited to launch V2 of our wonderful community, Fully Connected!
FC's mission is simple – to build a shared knowledge base for machine learning in the real world. 🙂
🪄
We are so excited to launch 2 highly requested features today – Production Monitoring, and the ability to build powerful LLM/ML applications in visual, interactive, and composable way.
Check them out here:
🪄 Production Monitoring:
🪄 Weave:
I'm working on a bunch of Computer Vision blog posts, taking requests for topics! 👩🎤
Lineup so far:
• Object Tracking using SLAM
• Image Captioning with CNN+RNNs
• Style Transfer
• Face Detection
• YOLO detection model
• Attention Mechanism
#MachineLearning
#100DaysOfMLCode
If you're looking to play with Generative Modelling, David Foster's book is 🔑.
You'll learn to build state-of-the-art deep learning models that can paint, write beautiful prose, compose music and play games.
#MachineLearning
#DataScience
#100DaysOfMLCode
We're thrilled to announce that we just raised $45 million!
So proud of the whole W&B team & of
@l2k
for making this happen. 🥂
We're building a really special product, with really special people and having an insane amount of fun doing it!
Great intro to Graph Neural Networks by Yash Kotadia that summarizes the need for GNNs, explains why they're special & analyzes the Gated Graph Convolutional Network architecture which is performant while being scalable.
#machinelearning
#100daysofmlcode
I get a ton of messages from people asking what the best way to learn Machine Learning is.
Get
@aureliengeron
's book. Read it. Play with the notebooks till you can recreate all dat code. Carry it around. Become best friends with it. 💑
#MachineLearning
In today's episode
@l2k
interviews
@chipro
about the gnarliest challenges in moving machine learning pipelines from research into production.
Chip draws on lessons from her ML research at Snorkel, NVIDIA, Netflix & Primer.
#machinelearning
#deeplearning
GitHub Actions provide a way to incorporate software best practices into the world of
#machinelearning
.
.
@HamelHusain
is showing you how to use em to automate your ML workflow.
PS: yes there's an action to automate model tracking with W&B🤗
Watch here–
“Hands down one of the best AI & ML podcast episodes I’ve watched recently.”
“Love who you bring on to these interviews. Very insightful questions.”
“This channel is impressive as whole team & guests.”
Thanks for all the love! 😇
Checkout more episodes:
Hi folks! If you watch ML podcasts or engage in ML communities avidly, I'd love to get your feedback.
Reply to this tweet if you're down for a 20min call! 🙂
As a token of our gratitude, we’ll offer you a W&B shirt/Starbucks giftcard.
#machinelearning
#deeplearning
#datascience
💯 advice from dad: "Imagine you've been validated, in every way you want to be validated - by your peers, heroes, society. What do you work on now? What's the one thing you're drawn to even after you've received complete societal validation? Start working on that thing now!"
We're thrilled to announce the new YOLOv5 x W&B integration!
You can now visualize your datasets, log training progress in real-time, resume crashed runs across any device, & make your model training more robust!
Find out more:
#deeplearning
Congrats
@jeremyphoward
& team on releasing Fastai v2! We're thrilled about the launch.
Take Fastai for a spin:
◦ Visualize, track & compare Fastai models
◦ Semantic segmentation with Fastai
#deeplearning
#machinelearning
Pretty deep into computer vision & I've never been moar excited in my life! 😍
Finding machine learning was like finding my work soulmate. It's exhilarating & peaceful & challenging & all-consuming in the best possible way & I wanna do it for the rest of my life!
#100DaysOfMLCode
Can Apple's M1 help you train models faster & cheaper than NVidia's V100?
In this post, our in-house investigative reporter extraordinaire
@vanpelt
analyzes runtimes, energy usage, & performance of a Tensorflow model on the two to find the answer. 🧐
Let's get more women on the stage! ❤️ I just built an app to find + book women in tech speakers.
Ladies, would love for you to sign up!
Conference organizers, we're adding more women to this list constantly! Book em!
#WomeninTech
#Women
#Tech
See how a neural network learns to render new views from a learned neural representation of a single scene.
We also visualize the effects of tweaking learning rates, embedding sizes etc 🛠️
#machinelearning
#deeplearning
#100daysofmlcode
This was such a delight to read! 😍
Cannot recommend enough if you're curious about interpretability in neural networks! It even comes with tools that let you look under the hood of the models.
Today on Gradient Dissent:
@josh_tobin_
joins
@l2k
to talk about his work with
@OpenAI
's robotics team translating models from physics simulators to the real world, writing good unit tests for ML, & his incredible
@full_stack_dl
classes.
#machinelearning
Spreading the word about good experiment tracking hygiene in
@chipro
's Stanford CS 329S class today.
My favorite part is always the 🤯 moment when people see the move from log files/google docs to W&B.
Track your models, kids.
In 2020, I trained 3,403 models over 450 hours, with 11 hours of GPU training time.
I tuned 4 hyperparameters per model on average.
Get insights on your model training with
@weights_biases
!
#myyearwrapped
.
@l2k
is interviewing Jensen, CEO of
@nvidia
, next week on our podcast, Gradient Dissent.
What questions would you guys like him to ask Jensen about NVIDIA, GPUs, what he's cooking up next? Reply in this thread! 🙂
#deeplearning
#machinelearning
Recommend this 🔑 article on being a 10x data scientist.
Includes 💯 nuggets like:
"The only way to build trust with the scientific community is to commit to radical transparency ... post all your training runs to a public dashboard via
@weights_biases
."
I got the "but you don't look like a programmer" from an 8 yr old girl today and it broke my heart that we're teaching arbitrary gender stereotypes to kids that young. 💔
I can't wait to have a daughter someday and raise her to be a proud badass nerd and embrace her femininity!
What happens when a transformer meets a twitter account? Find out for yourself! 🤔
In 5 minutes, fine-tune a pre-trained transformer on anyone's tweets, just by clicking a couple of buttons.
🛸 Try it here
#machinelearning
#deeplearning
#100daysofmlcode
Found this while doing research for another project!
State of the art models/papers (with code!) for ImageNet, in one place! Excellent starting point if you wanna use transfer learning for your image classification project!
#MachineLearning
#DataScience
We're excited to bring you our first ever meetup in London on April 20th!
Join our CEO
@l2k
and
@EMostaque
, CEO of
@StabilityAI
for a night of talks on LLMs, Q&As, drinks and hanging with like-minded ML engineers!
👩💻 We have limited seats, register here:
So excited to announce our new
@OpenAI
integration!
You can now add one line of code to fine-tune your GPT-3 model with OpenAI's API and sync all training metrics, datasets and models to your W&B dashboard. Try it!
💻 openai wandb sync
Reading this post from
@dalmiaman
where gives a play by play of how he won a
#Kaggle
silver medal. 🏅
Fascinating read through his initial struggles, journey to finding the final solution & everything he learned in between.
#deeplearning
#machinelearning
Did you know you can massively accelerate model training time w/ a Keras utility wrapper function?
My wonderful colleague Stacey shows you the magic trick of data-parallel distributed training in this post!
#machinelearning
#deeplearning
#100daysofmlcode
If you're doing ML & need a community of people to nerd out with – W&B Forum is it.
We have folks entering & winning hackathons together, getting help on weekend projects, talking shop.
It's a great time🥰
#MachineLearning
#DeepLearning
#100DaysofMLCode
I made a couple resources to get you started with Weights & Biases:
+ Visualize model predictions (images, videos, audio, point clouds)
+ Track model performance
Lemme know what you think🤗
#Kaggle
#DeepLearning
#MachineLearning
500 people have registered for our London meetup with
@l2k
and
@StabilityAI
this month.
So we've booked a larger, sicker venue.
We still have a couple spots left. Register here:
See you April 20th!
I’ll be in London Jan-March. Would love to meet with ML engineers based in the area.
What are some local ML communities I should check out? Who should I meet?
MNIST is the secret handshake we use to introduce people into the world of machine learning. 🤙🏼
In this
@weights_biases
dashboard, I dive deep into what a state of the art model (99.4% acc) gets wrong about MNIST!
#MachineLearning
#DataScience
The 2nd episode of Gradient Dissent is live!
This week
@l2k
talks to
@nkoumchatzky
about how NVIDIA uses machine learning in production for self driving.
This was a fascinating conversation. We really hope you enjoy it!
#machinelearning
#deeplearning
Celebrate the future of ML with us in San Francisco.
We’re hosting a one-day user conference on how GenAI is transforming the world, and tickets are going fast!
You’ll hear from the creators of the most exciting ML tools like
@LangChainAI
,
@OpenAI
,
We’re excited to launch Fully Connected, v2 of our ML community today!
FC brings together ML engineers to
⭐️share research ideas
⭐️learn from industry leaders
⭐️discover the latest tools & papers
⭐️collaborate on projects together
Everyone is welcome🙂
I made a quick video to walk y'all through picking the optimal neural net architecture by tracking your metrics, using
@weights_biases
in your Kaggle kernels. 👩🎨
• Kernel:
• Demo:
#MachineLearning
#DataScience
Remote work is hard.
Remote machine learning is even harder.
@l2k
talks to leaders at different sized ML teams on how they track experiments, debug models, keep teams aligned, & manage distributed training.
#machinelearning
#deeplearning
#100daysofmlcode
I am blown away by the speed at which the W&B team is reporting on the latest ML news – 3-4 pieces of what to pay attention to everyday! 🙂
If y'all haven't seen it, you need to –
This team built a deep learning sea ice forecasting system, to help reduce risks from rapid sea ice loss and they used
@weights_biases
to train their models!
Saving polar bears and helping the global climate – def one of my fav applications of W&B! 😍
Highly recommend this video/paper on
• adversarial attacks on neural nets
• how changing just *one* pixel lets us fool almost any neural net
• most datasets contain features that are predictive, but can be flipped by adversarial attacks
(contd…)
#MachineLearning
#DataScience
Reproducing papers alone is a major undertaking – join the slack channel to get guidance, work on paper reproductions together, & get your questions answered.
👩💻 Community: , ml-reproducibility channel
👨💻 Learn more:
#deeplearning
PyTorch Lightning lets you decouple your science code from engineering code.
Ayush shows you how to visualize Lightning models and optimize model hyperparameters with an easy W&B integration.
#machinelearning
#deeplearning
#100daysofmlcode
If you are excited about building the next generation of ML tools and joining the W&B 🚀, we're hiring on the growth, eng, product, sales, & data science teams!
Super excited to announce a new investment in Weights & Biases by NVIDIA!
To celebrate, check out this fabulous interview
@l2k
did with NVIDIA CEO, Jensen:
Today, we’re proud to announce an investment from the team at
@nvidia
.
This means a lot to us here at W&B as we continue to build out the best platform for ML developers.
You can read all about our growing partnership here:
In this episode of Gradient Dissent,
@zacharylipton
talks to
@l2k
about his fascinating journey from musician to Professor, applications of ML in medicine, counterfacutally augmented data and algorithmic fairness.
It's a great listen!
#machinelearning
We're excited to announce our newest tool, Artifacts.
With Artifacts you can store & version your datasets, models & results.
👨💻 Read the behind-the-scenes story of how Artifacts came to life by
@shawnup
.
#machinelearning
#deeplearning
#100daysofmlcode
Lavanya🦋, Head of Growth at Weights & Biases will be talking about "Reproducible machine learning at scale" on my YouTube channel this Friday at 6PM CET / 9.30 PM IST / 9 AM PDT. Check out the event here: and click on the🔔icon to get reminded!
I'm writing a series posts showing anyone how to build & productionize an LLM powered app.
Here's the first one where I go from 17% to 91% accuracy through Prompt Engineering on a real world use case!
👩🏼💻 Notebook:
✍🏼 Blog post:
#ComputerVision
twitter – I'm very excited to dive deeper into explainability of vision models this long weekend.
If you have any papers you think I should read, please send them my way!
Thank you 🙂
Check out
@kevinkwshen
's amazing
#kaggle
kernel, where he shows you how to visualize a FasterRCNN's predictions and find the best set of hyperparameters.
In v2, he'll dig deeper into FasterRCNN's hyperparameters.