Inspired by progress in unsupervised representation learning for natural language, we examine whether similar models can learn useful representations for images. We train a sequence Trans-former to autoregressively predict pixels, without incorporating knowledge of the 2D input structure.Despite training on low-resolution ImageNet without labels, we find that a GPT-2 scale model learns strong image representations as measured by linear probing, fine-tuning, and low-data classification. On CIFAR-10, we achieve 96.3% accuracy with a linear probe, outperforming a supervised Wide ResNet, and 99.0% accuracy with full fine-tuning, matching the top supervised pretrained models. An even larger model trained on a mixture of ImageNet and web images is competitive with self-supervised benchmarks on ImageNet,achieving 72.0% top-1 accuracy on a linear probe of our features. Read More
#image-recognitionMonthly Archives: August 2020
Why are so many AI systems named after Muppets?
An inside joke that says a lot about AI development
One of the biggest trends in AI recently has been the creation of machine learning models that can generate the written word with unprecedented fluidity. These programs are game-changers, potentially supercharging computers’ ability to parse and produce language.
But something that’s gone largely unnoticed is a secondary trend — a shadow to the first — and that is: a surprising number of these tools are named after Muppets. Read More
Deepfakes ranked as most serious AI crime threat
Fake audio or video content has been ranked by experts as the most worrying use of artificial intelligence in terms of its potential applications for crime or terrorism, according to a new UCL report.
The study, published in Crime Science and funded by the Dawes Centre for Future Crime at UCL (and available as a policy briefing), identified 20 ways AI could be used to facilitate crime over the next 15 years. Read More
The quest for quantum-proof encryption just made a leap forward
Quantum computers could make encryption a thing of the past, but 15 contenders are trying to prove they have what it takes to safeguard your data.
… Five of the shortlisted candidates announced last week use lattice approaches that have no known quantum solution, and NIST’s new status report says they are “the most promising general-purpose algorithms” in the list. Read More
CAI Achieves Milestone: White Paper Sets the Standard for Content Attribution
Today marks a significant milestone for the Content Authenticity Initiative (“CAI”) as we publish our white paper, “Setting the Standard for Content Attribution”. It addresses the mounting challenges of inauthentic media and our proposal for an industry-standard content attribution solution that will enable creators to securely attach their identity and other information to their work before they share it with the world. Read More
Sign language recognition using deep learning
TL;DR It is presented a dual-cam first-vision translation system using convolutional neural networks. A prototype was developed to recognize 24 gestures. The vision system is composed of a head-mounted camera and a chest-mounted camera and the machine learning model is composed of two convolutional neural networks, one for each camera. Read More
AI and the Workforce
Artificial intelligence will transform the nature of work and affect virtually all aspects of the economy. However, artificial intelligence is not the first technology to have such wide-reaching impacts on the workforce, and the United States has gone through various technological transitions in the past.
For instance, the steam engine helped give rise to the Industrial Revolution. Initially developed in the 18th century to pump water out of mines, the technology behind the steam engine was quickly found to have other uses that spurred industry and innovation. The steam engine disrupted the jobs of many workers and made certain skillsets less in demand or even obsolete, but it also created many new manufacturing jobs. Read More
Artificial Intelligence and National Security
Artificial intelligence will have immense implications for national and international security, and AI’s potential applications for defense and intelligence have been identified by the federal government as a major priority.
There are, however, significant bureaucratic and technical challenges to the adoption and scaling of AI across U.S. defense and intelligence organizations. Moreover, other nations—particularly China and Russia—are also investing in military AI applications. As the strategic competition intensifies, the pressure to deploy untested and poorly understood systems to gain competitive advantage could lead to accidents, failures, and unintended escalation. Read More
Why is Tik-Tok being forced to sell to Microsoft?
So, what is the real purpose of suppressing Tik-Tok?
It is a fight over “discourse power” on the Internet.
— Luke Wen via Jeffrey Ding — Read More
On the Use of AI for Satellite Communications
This document presents an initial approach to the investigation and development of artificial intelligence (AI) mechanisms in satellite communication (SatCom) systems. We first introduce the nowadays SatCom operations which are strongly dependent on the human intervention. Along with those use cases, we present an initial way of automatizing some of those tasks and we show the key AI tools capable of dealing with those challenges.Finally, the long term AI developments in the SatCom sector is discussed. Read More
Rick's Cafe AI 