​

Top ML Papers of the Week

Jurassic World Remake: Bringing Ancient Fossils Back to Life via Zero-Shot Long Image-to-Image Translation

A zero-shot image-to-image translation method that transforms given fossils images into images of living animals of specified species based on input prompts, such as Dinosaurs

Paper: https://arxiv.org/pdf/2308.07316.pdf​

​

Group Pose: A Simple Baseline for End-to-End Multi-person Pose Estimation

Paper: https://arxiv.org/pdf/2308.07313.pdf​

​

Accurate Eye Tracking from Dense 3D Surface Reconstructions using Single-Shot Deflectometry

Rethinking the way how specular reflections can be used for eye tracking.

Paper: https://arxiv.org/pdf/2308.07298.pdf​

​

UniWorld: Autonomous Driving Pre-training via World Models

Embedding a spatial-temporal world model into robots, termed UniWorld, to help them perceive their surroundings and predict the future behavior of other participants.

Paper: https://arxiv.org/pdf/2308.07234.pdf​

​

Towards Robust Real-Time Scene Text Detection: From Semantic to Instance Representation Learning

Paper: https://arxiv.org/pdf/2308.07202.pdf​

​

ML Deep Dive: Federated Learning with Substra

There is a specific type of machine learning topics that I never use in my work and I don't fully understand it, but it still fascinates me due to its clear and important potential.
​
One of these topics is Federated Learning.
​
I have covered in other editions of the newsletter what federated learning is and how it works from a high level.
​
Here's a simple definition as a reminder.
​
Federated learning (FL) is a decentralized machine learning procedure to train models using multiple data providers. Instead of gathering data on a single server, the data remains locked on local servers as only the algorithms and models are shared between the data holders.
​
I have covered before 2 frameworks that can be used for Federated Learning.
​
One is MONAI, which is mainly focused on the healthcare sector.
​
The other is Nvidia FLARE, which is a more general purpose FL framework.
​
But another framework for FL which recently has been added to the Pytorch ecosystem is called Substra.
​
Substra is an open source federated learning framework that provides a flexible python interface and a web app to run FL training at scale.
​
Below is a simple diagram on how Substra works. Basically it's a diagram for how FL works.

Easily installed through: pip install substrafl

Why should you care?

If you work in an industry where privacy is the number one priority but at the same time this priority is hindering you and your team from building powerful machine learning models because you just don't have enough data, then federated learning could be a potential solution for you.

Because federated learning will allow you to build powerful ML models without the need for direct access to other companies data. The only thing that will be shared between your company and other companies are the model weights.

​

Tweet of the Day

​

PyTorch @PyTorch New LIVE PyTorch training course! 🚨 Designed for machine learning practitioners who want to add deep learning models in PyTorch, this course provides hands-on experience to train and fine-tune deep learning models. Enroll today: https://hubs.la/Q01_5YLB0 10:4 PM • Aug 15, 2023 9 Retweets 62 Likes Read 1 replies

​

​

Meme of the Day 😂

​