News

PyTorch – Internal Architecture Tour

This post is a tour around the PyTorch codebase, it is meant to be a guide for the architectural design of PyTorch and its internals. My main goal is to provide something useful for those who are interested in understanding what happens beyond the user-facing API and show something new beyond what was already covered in other tutorials.

Building Windows: 4 million commits, 10 million work items

Microsoft’s switch to using Git as the version control system for Windows’ development has resulted in many challenges. Git wasn’t really built for a 300GB repository with 3.5 million files, and the engineering effort to make Git scale in this way continues.

Semantic Image Segmentation with DeepLab in Tensorflow

Today, we are excited to announce the open source release of our latest and best performing semantic image segmentation model, DeepLab-v3+ [1], implemented in Tensorflow. This release includes DeepLab-v3+ models built on top of a powerful convolutional neural network (CNN) backbone architecture [2, 3] for the most accurate results, intended for server-side deployment. As part of this release, we are additionally sharing our Tensorflow model training and evaluation code, as well as models already pre-trained on the Pascal VOC 2012 and Cityscapes benchmark semantic segmentation tasks.

Why the Tiny Weight of Empty Space Is Such a Mystery

It’s Time to Make Human-Chimp Hybrids

It is a bit of a stretch, but by no means impossible or even unlikely that a hybrid or a chimera combining a human being and a chimpanzee could be produced in a laboratory. After all, human and chimp (or bonobo) share, by most estimates, roughly 99 percent of their nuclear DNA. Granted this 1 percent difference presumably involves some key alleles, the newgene-editing tool CRISPR offers the prospect (for some, the nightmare) of adding and deleting targeted genes as desired.

Facebook’s Bet on an Augmented Reality Future

Today, Facebook is fighting its fellow technology powerhouses, Apple AAPL +1.72% and Google — and still to some extent, Snap — in a high-stakes battle to rule as the platform of choice for AR developers. The technology itself, while still in its infancy, has exploded in popularity, confirming Zuckerberg’s more recent intuition that AR could sprint toward mass adoption even while VR remained an awkward technology whose appeal is largely limited to hardcore gamers. AR’s key advantage is that it doesn’t depend on a pricey, bulky headset that isolates its users.

Studying how the brain relinquishes childhood memories

We called them fairy rocks. They were just colorful specks of gravel—the kind you might buy for a fish tank—mixed into my preschool’s playground sand pit. But my classmates and I endowed them with magical properties, hunted them like treasure, and carefully sorted them into piles of sapphire, emerald, and ruby.

Diving into Deep Convolutional Semantic Segmentation Networks and Deeplab_V3

Deep Convolutional Neural Networks (DCNNs) have achieved remarkable success in various Computer Vision applications. Like others, the task of semantic segmentation is not an exception to this trend.

Reptile: A Scalable Meta-Learning Algorithm

We’ve developed a simple meta-learning algorithm called Reptile which works by repeatedly sampling a task, performing stochastic gradient descent on it, and updating the initial parameters towards the final parameters learned on that task. This method performs as well as MAML, a broadly applicable meta-learning algorithm, while being simpler to implement and more computationally efficient.

Will the Quantum Nature of Gravity Finally Be Measured?

In 1935, when both quantum mechanics and Albert Einstein’s general theory of relativity were young, a little-known Soviet physicist named Matvei Bronstein, just 28 himself, made the first detailed study of the problem of reconciling the two in a quantum theory of gravity. This “possible theory of the world as a whole,” as Bronstein called it, would supplant Einstein’s classical description of gravity, which casts it as curves in the space-time continuum, and rewrite it in the same quantum language as the rest of physics.