Research

Deep Learning Research

The Theoretical Aspects of Gradient Methods in Deep Learning

In this talk from Microsoft Research Asia, Jian Li of Tsinghua University explains Theoretical Aspects of Gradient Methods in Deep Learning by

Read More
Computer Vision Research

Fourier Feature Networks and Neural Volume Rendering

Fourier Feature Networks are an exciting new development in Computer Vision, and their use for modeling radiance fields has produced a range of impressive results at the meeting point of Computer Vision and Computer Graphics. In this lecture, Matthew covers the motivation behind using Fourier features in neural network training, introduces the fundamentals of volumetric […]

Read More
Research Science

2021’s Biggest Breakthroughs in Math and Computer Science

2021 was a big year. Researchers found a way to idealize deep neural networks using kernel machines—an important step toward opening these black boxes. There were major developments toward an answer about the nature of infinity. And a mathematician finally managed to model quantum gravity. Read the articles in full at Quanta Magazine.

Read More
AI Computer Vision Facial Recognition Research

Microsoft’s AI Understands Humans Despite Never Seeing One

It looks like synthetic data has its day and how! Two Minute Papers explores the paper “Fake It Till You Make It – Face analysis in the wild using synthetic data alone” in the video below.

Read More
Neural Networks Research

DIABLo: a Deep Individual

Here’s an interesting video from Microsoft Research In this project, we have developed and studied a deep neural network-based individual-agnostic general-purpose binaural localizer (BL) for sound sources located at arbitrary directions on the $4\pi$ sphere. Unlike binaural localization models trained with an HRIR catalog associated with a specific head and ear shape, an individual-agnostic model […]

Read More
AI Research

Implicit MLE: Backpropagating Through Discrete Exponential Family Distributions

Backpropagation is the workhorse of deep learning, but unfortunately, it only works for continuous functions that are amenable to the chain rule of differentiation. Since discrete algorithms have no continuous derivative, deep networks with such algorithms as part of them cannot be effectively trained using backpropagation. This paper presents a method to incorporate a large […]

Read More
AI Research

Three Explorations on Pre-Training: an Analysis, an Approach, and an Architecture

In this talk from Xinlei Chen, Facebook AI Research, covers three of their recent explorations on pre-training. First is an analysis on object/attribute detection pre-training, which produces bottom-attention features extensively used in vision and language research. The main finding is that plain grid features can work equally well without object proposals, while being significantly faster. […]

Read More
Computer Vision Natural Language Processing Research

Tightly Connecting Vision and Language

Remarkable progress has been made at the intersection of vision and language. While showing great promise, current vision and language models may only weakly “connect” the two modalities and often fail in the wild. In this talk, Goggle’s Soravit Changpinyo will present recent efforts aiming to bridge this gap along two dimensions: informativeness and controllability. […]

Read More
AI Research

DeepMind’s AI Plays Catch and So Much More!

Two Minute Papers examines the paper “Open-Ended Learning Leads to Generally Capable Agents.”

Read More
AI Interesting Research

PonderNet: Learning to Ponder

Humans don’t spend the same amount of mental effort on all problems equally. Instead, we respond quickly to easy tasks, and we take our time to deliberate hard tasks. DeepMind’s PonderNet attempts to achieve the same by dynamically deciding how many computation steps to allocate to any single input sample. This is done via a […]

Read More