Deep Learning Weekly Issue #165

AI democratization in era of GPT-3, Amazon's new edge processor, a look inside the neural network black box, & more

Hey folks,

This week in deep learning we bring you AI democratization in the era of GPT-3, how to trick a neural network, Amazon's new products that provide surveillance-as-a-service, and the AI upgrade for the National Guard’s fire-mapping drones.

You may also enjoy Amazon’s AZ1 Neural Edge processor, Snapdragon's 750G chip with mmWave 5G support and AI noise suppression, this paper on 3D dense face alignment and more!

As always, happy reading and hacking. If you have something you think should be in next week's issue, find us on Twitter: @dl_weekly.

Until next week!

Industry

AI Democratization in the Era of GPT-3

In this article, the author explains concerns around the democratization of AI in light of Microsoft and OpenAI’s recent exclusivity agreement.

Amazon is embracing surveillance-as-a-service

Amazon rolled out over a dozen new devices and services that show that they are in the business of selling surveillance-as-a-service.

Towards ML Engineering: A Brief History Of TensorFlow Extended (TFX)

This article gives a tour of Sibyl and TensorFlow Extended (TFX), two successive end-to-end (E2E) ML platforms at Alphabet.

The National Guard’s Fire-Mapping Drones Get an AI Upgrade

Algorithms that quickly track the movement of wildfires could help firefighters‚ but the tech could also be put to non-humanitarian use.

Google makes AI Platform Prediction generally available with expanded features

Google announced the general availability of AI Platform Prediction, a service that allows companies to host machine learning models on its public cloud without having to worry about infrastructure management.

Mobile + Edge

TensorFlow Lite GPU delegate

New in TensorFlow Lite 2.3.0 is the Experimental GPU Compatibility list that allows one to safely enable GPU acceleration for up to 10x speed increase by taking into account OpenGL versions, driver features, and device resources.

Snapdragon 750G unveiled with mmWave 5G support, AI noise suppression

Qualcomm introduced the Snapdragon 750G – a global 5G chipset with AI-powered noise suppression and support for 120fps gaming.

Amazon’s AZ1 Neural Edge processor will make Alexa voice commands even faster

It made the silicon module for this processor with MediaTek.

Adobe’s Liquid Mode leverages AI to reformat PDFs for mobile devices

Adobe announced an update for Acrobat Reader that leverages Sensei, the company’s machine learning platform, to make it easier to read whole documents on smartphones and tablets.

Learning

Advancing Instance-Level Recognition Research

In this post, Google AI highlights some results from the Instance-Level Recognition Workshop at ECCV ’20.

Introducing TensorFlow Recommenders

Introducing TensorFlow Recommenders, an open-source package that makes building, evaluating, and serving recommender models easy. Find recommendations for movies, restaurants, and much more!

Retrieval Augmented Generation: Streamlining the creation of intelligent natural language processing models

Building an NLP model that researches and contextualizes is more challenging, but it's essential for future advancements. Facebook AI recently made substantial progress in this realm with our Retrieval Augmented Generation (RAG) architecture, an end-to-end differentiable model that combines an information retrieval component with a seq2seq generator.

Peering Inside The Blackbox — How To Trick A Neural Network

In this tutorial, the author shows you how to use gradient ascent to figure out how to misclassify an input.

Libraries & Code

[GitHub] PaddlePaddle/PaddleOCR

Awesome OCR toolkits based on PaddlePaddle (3.5M practical ultra lightweight OCR system, support training and deployment among server, mobile, embedded and IoT devices).

[GitHub] microsoft/Bringing-Old-Photos-Back-to-Life

Bringing Old Photo Back to Life (CVPR 2020 oral).

[GitHub] cleardusk/3DDFA_V2

The official PyTorch implementation of Towards Fast, Accurate and Stable 3D Dense Face Alignment, ECCV 2020.

Papers & Publications

SSMBA: Self-Supervised Manifold Based Data Augmentation for Improving Out-of-Domain Robustness

Abstract: Models that perform well on a training domain often fail to generalize to out-of-domain (OOD) examples. Data augmentation is a common method used to prevent overfitting and improve OOD generalization. However, in natural language, it is difficult to generate new examples that stay on the underlying data manifold. We introduce SSMBA, a data augmentation method for generating synthetic training examples by using a pair of corruption and reconstruction functions to move randomly on a data manifold. We investigate the use of SSMBA in the natural language domain, leveraging the manifold assumption to reconstruct corrupted text with masked language models. In experiments on robustness benchmarks across 3 tasks and 9 datasets, SSMBA consistently outperforms existing data augmentation methods and baseline models on both in-domain and OOD data, achieving gains of 0.8% accuracy on OOD Amazon reviews, 1.8% accuracy on OOD MNLI, and 1.4 BLEU on in-domain IWSLT14 German-English.

Towards Fast, Accurate and Stable 3D Dense Face Alignment

Abstract: Existing methods of 3D dense face alignment mainly concentrate on accuracy, thus limiting the scope of their practical applications. In this paper, we propose a novel regression framework which makes a balance among speed, accuracy and stability. Firstly, on the basis of a lightweight backbone, we propose a meta-joint optimization strategy to dynamically regress a small set of 3DMM parameters, which greatly enhances speed and accuracy simultaneously. To further improve the stability on videos, we present a virtual synthesis method to transform one still image to a short-video which incorporates in-plane and out-of-plane face moving. On the premise of high accuracy and stability, our model runs at over 50fps on a single CPU core and outperforms other state-of-the-art heavy models simultaneously. Experiments on several challenging datasets validate the efficiency of our method. Pre-trained models and code are available at this https URL.