The question of whether technology is good or bad depends on how it’s developed and used. Nowhere is that more topical than in technologies using artificial intelligence.
When developed and used appropriately, artificial intelligence (AI) has the potential to transform the way we live, work, communicate and travel.
New AI-enabled medical technologies are being developed to improve patient care. There are persuasive indications that autonomous vehicles will improve safety and reduce the road toll. Machine learning and automation are streamlining workflows and allowing us to work smarter.
Around the world, AI-enabled technology is increasingly being adopted by individuals, governments, organisations and institutions. But along with the vast potential to improve our quality of life, comes a risk to our basic human rights and freedoms.
Appropriate oversight, guidance and understanding of the way AI is used and developed in Australia must be prioritised. Read More
Daily Archives: April 5, 2019
A Hitchhiker’s Guide On Distributed Training of Deep Neural Networks
Deep learning has led to tremendous advancements in the field of Artificial Intelligence. One caveat however is the substantial amount of compute needed to train these deep learning models. Training a benchmark dataset like ImageNet on a single machine with a modern GPU can take up to a week,distributing training on multiple machines has been observed to drastically bring this time down. Recent work has brought down ImageNet training time to a time as low as 4 minutes by using a cluster of 2048 GPUs. This paper surveys the various algorithms and techniques used to distribute training and presents the current state of the art for a modern distributed training framework. More specifically, we explore the synchronous and asynchronous variants of distributed Stochastic Gradient Descent, various All Reduce gradient aggregation strategies and best practices for obtaining higher throughout and lower latency over a cluster such as mixed precision training, large batch training and gradient compression. Read More
Platform Strategy Survey
This article provides a brief survey of the platform strategy literature and is organized around launch strategies, governance, including private ordering, and competition. A platform strategy is the mobilization of a networked business platform to expand into and operatein a given market. A business platform, in turn, is a nexus of rules and infrastructure that facilitate interactions among network users. A platform may also be viewed as a published standard,together with a governance model, that facilitates third party participation. Platforms provide building blocks that serve as the foundation for complementary products and services. They also match buyers with suppliers, who transact directly with each other using system resources and are generally subject to network effects. Examples include operating systems, game consoles, payment systems, ride sharing platforms, smart grids, healthcare networks, and social networks. Read More
Two-Sided Network Effects: A Theory of Information Product Design
How can firms profitably give away free products? This paper provides a novel answer and articulates trade-offs in a space of information product design. We introduce a formal model of two-sided network externalities based in textbook economics—a mix of Katz and Shapiro network effects, price discrimination,and product differentiation. Externality-based complements, however, exploit a different mechanism than either tying or lock-in even as they help to explain many recent strategies such as those of firms selling operating systems, Internet browsers, games, music, and video.The model presented here argues for three simple but useful results. First, even in the absence of competition,a firm can rationally invest in a product it intends to give away into perpetuity. Second, we identify distinct markets for content providers and end consumers and show that either can be a candidate for a free good.Third, product coupling across markets can increase consumer welfare even as it increases firm profits.The model also generates testable hypotheses on the size and direction of network effects while offering insights to regulators seeking to apply antitrust law to network markets. Read More
Rick's Cafe AI 