9 NEWSFEED Updates & upgrades
EDGEWARE’S BIG EVENT STREAMING Edgeware is oering a suite of new features in its CND technology for delivering live OTT services at large scale and, it says, in higher quality than broadcast TV. The technology was developed for operators who have invested in the rights to sporting events or reality shows attracting large numbers of simultaneous viewers. The suite of new features includes Edgeware’s Dynamic Asset Spreading which enables operators to spread popular content across a wider range of servers. The wider the spread the greater the streaming capacity, while reducing the risk of server overload. Another new feature, called Predictive Session Load Balancing, dynamically reacts to any build-up in the queue of requests, checking them against pre-set limits. It controls the way very large numbers of viewers come on-board for the same content and directs new client requests to the next available server to pre-empt network congestion. A third tool, Edge Asset Request Consolidation, keeps streaming servers from sending out many identical requests for the same content if there is a cache-miss. Cache-misses can occur for live content in networks with minimised latency because new segments of content may not have been written to the origin server before the client requests them. This tool can amalgamate thousands of identical requests for new segments into one before sending it to the origin. As well as relieving pressure on the origin server, it also preserves capacity for getting data to a distribution point. The new features are available and already being used in live networks.
IBM’S PROJECT DEBATER GETS ARGUMENTATIVE
Project Debater, an AI developed by IBM, has gone head to head against two humans in a debating contest. Held at a conference room at IBM’s San Francisco oices, the event featured a stage occupied by a tall black monolith on one side and two frail, squishy homo sapiens on the other – 2016 Israeli debate champion Noa Ovadia and debate expert Dan Zafrir. Project Debater faced o against two humans in separate arguments. The first topic was “We should subsidise space exploration”. The second was: “We should increase the use of telemedicine”. The format of the two debates was a four-minute opening statement from each side, followed by four-minute rebuttals, then two-minute concluding summaries of two minutes each. IBM’s stated goal in developing Project Debater was to use AI “to facilitate intelligent debate so we can build well-informed arguments and make better decisions.” The developers hope that technologies like Project Debater might challenge humans to consider options they might not have considered before. The project has been in development at IBM’s lab in Haifa, Israel. The elements which come together to form the argumentative robot include argument mining, in which the AI accesses past evidence, like website, journals and news
reports, to support its position, as well as the development of deep neural nets with weak supervision. The latter is a technique of machine learning which uses less refined, but much larger scale data sets. Weak supervision helps avoid the bottlenecks which occur in an AI’s development by spoon-feeding it more specific, but limited, data sets. Natural Language Processing algorithms are one of the key elements in the Project Debater design. A back and forth using real human language is one of the projects main goals. Some of the NLP elements integrated into Project Debater are automatic speech recognition, the ability to detect argumentative structures, and identification of thematically related sentences. A simple word like “should”, which humans take for granted, doesn’t come as naturally to an AI as discrete nouns like “cat” or “podium”. The final results of the debates were: Project Debater 1, Humans 1. Although it’s worth noting that the judgement was made by a voting audience of humans! The topic for the next debate: “Are humans prejudiced against AI’s – and if so, what should be done to stop them?” ■ Read more at IBM’s Project Debater website: https://www.research.ibm.com/ artificial-intelligence/project-debater/ research.html
feedzine feed.zine feedmagazine.tv
Powered by FlippingBook