Monthly Archives: August 2019
How Wimbledon and Watson use AI to edit autofocus for video (Part 3)
This collaborative approach really helps to show the role of people in training AI systems – at least so far. How far away are we to automate the data collection process? Boyden said that we could have done 70% or 80% of them in an automated way – this number is positive when coaching, but the challenge we have at Wimbledon is that the league requires 100% accuracy at all times.
In the future, a series of changes will be possible and Watson will gradually take on more responsibilities. Currently, it is only supporting people, but in the future, it may begin to make its own recommendations, which will then be confirmed by a tennis expert.
Objectives from this partnership:
Increase the fan experience on the platform by accelerating and improving the match focus production process over a 13-day period.
Expand the content by creating a solution that makes it easy to scale, providing highlights from the Wimbledon stadiums that have never been used before. Support Wimbledon’s strategic goal, reduce dependence on third parties, thereby helping Wimbledon enhance control of the output quality of the content.
Create quality outputs to hand over to international broadcast partners.
Reduce the pressure on editorial and content groups, allowing them to focus on other areas that add value to the content. Contribute to improving video performance on all Wimbledon platforms in the Championship.
In the competitive digital and sports environment, Cognitive Highlighs helped attract 69.9 million digital hits in the Championship (69.4 million in 2016) and 436 million page views (395 million in 2016)
14.4 million views of new accent videos were created without human intervention.
A total of 250 premium packages were created, an increase of 252% from the 2016 Championship.
Reduce turnaround time from a minimum of one hour to 15 minutes.
Automating the highlighting process of the match allows the content team to focus on serving the digital community, creating a total of 200 million video views on all platforms.
Outstanding scenes that are not used (due to limited rights) continue to be utilized by the team to maintain the audience’s engagement outside the Championship.
The new Cognitive Highlight approach drew the attention of the press with Wimbledon’s digital platforms, with 247 articles earning over 31 countries up 21%.
This is the first technical solution that uses existing IBM patents.
How Wimbledon and Watson use AI to edit autofocus for video (Part 2)
Applications of AI in Wimbledon
To do this, AI analyzes factors such as gestures and players’ reactions (such as punches on the air celebrating and greeting by touching two fists on each other) and listening to the sound of the crowd, like exhaling and cheering. For the highlights of Wimbledon 2019, IBM said it trained Watson to take advantage of more acoustical data, such as detecting every shot of the ball – making automatic cropping of match highlights closer around the main activities in the innings and can also help determine the series of pay backs.
Misleading AI
Of course, there are problems with deviations in the AI we face for a long time. Wimbledon has many situations that lead to fooling the automated system – the center of the stadium will have more fans (and therefore more noise), while a superstar like Roger Federer might have many fans, creating more cheering sounds than the 89-ranked players in the world. To detect the best action, regardless of these external factors, IBM trained Watson to better identify trends and patterns in acoustics – creating a fourth type of data point to reduce risk of deviation.
This result is enabled via AI OpenScale. IBM launched the AI OpenScale last year, providing a set of automated detection and mitigation tools – constantly monitoring bias-based decision-making based on AI applications. Specifically, for Wimbledon, this technology helps Watson plan to “learn” the overall crowd noise and use it as a neutral mechanism, from which other heights will be measured based on that basic measure.
Ace Index (to score the goal directly)
During the production of video accents, Watson also exploited match statistics for more context. These statistics are created by tennis players (usually players at the district level). They are people who are hired to sit next to the stage to identify subjective data points such as unforced error, other statistical information – such as handing the ball directly or double errors – and important moments in tennis matches, such as points that bring break points, set points, match points.