By Marc Górriz Blanch & Luka Murn
On November 27th and 28th, we attended the “AI in production and distribution” workshop organised by the European Broadcasting Union (EBU) at the BBC’s MediaCity facilities in Manchester. Under the scope of the world’s leading alliance of Public Service Media (PSM), this workshop explored how AI is changing the media landscape. Insights were provided on how media organisations are finding potential to leverage the technology. Experts from both the editorial and tech sectors had a look at specific use cases involving AI in the media industry. Some discussion topics included AI in the newsroom as well as opportunities and practical difficulties in using AI for media creation, distribution and consumption.
During the two-day event, several sessions addressed use cases of AI in media production and distribution, including technical talks about image processing and deep fakes and AI interpretability. In conjunction, we explored more ethical frameworks for the use of AI, in the context of Public Service Media values, and best practices for business metrics and KPIs.
As one of the keynote speakers, Gabriel Straub (Head of Data Science and Architecture at the BBC) discussed the importance of a responsible use of machine learning in a public service organisation, providing some successful day-to-day approaches used in the British Broadcasting Corporation. Among the two dozen talks, additional insights from the media sector were provided by Grant Franklin Totten (Head of Media & Emerging Platforms for Al Jazeera Media Network)describing their approach for transforming the media workflow in sense of creating more efficient methods for content creation and production and empowering the newsrooms and journalists to tell more compelling stories. Furthermore, Léonard Bouchet, Head of the Swiss-French Public Media Data and Archives Department, provided an approach for multimedia archives, exploring how AI can enrich and optimise essential metadata from audiovisual archived content.
We participated in demo sessions during breaks presenting our research work at JOLT, having the opportunity to discuss our results with attendees and experts in the AI and media sector. Our demos showcased our advances on automatic colourisation using AI, and the possible applications in media production and enhancement of archived content. We showed the potential benefits of our novel algorithm based on Generative Adversarial Networks, capable of boosting the colourfulness and perceptual meaningfulness of the generated predictions, and improving the state-of-the-art performance.
Moreover, we showcased the importance of interpretability of neural networks. Neural networks are increasingly used in solving complex tasks and network models are usually applied in a black-box manner, with little transparency on how these methods arrive at their predictions. Explainability is an important topic because it allows verification of such predictions. For EBU members, an essential task that can benefit from applying machine learning is video coding & compression. Although the complexity of neural network models is still too high to replace traditional methods of image and video processing, we can follow the mathematical principles of deep learning to understand how and what network models learn. We have shown that we can use these principles to supplement common video compression techniques and bring higher quality content to audiences.