Â鶹ԼÅÄ

On recommender systems

Polina has continued to work on measuring how useful existing tags are in making recommendations. Ben is using some work from Andrew and Alexandros that finds missing tags in the metadata which improve the recommendations, and tried to find ways of making sense of those tags so that they could be used by our Editorial colleagues. Which is very difficult!

Ben found that splitting recommender systems into two parts, one which works with the "big data" and compresses it into a sparse matrix, and another part, which actually does the recommender training and evaluation, is much quicker than using the "big data" to do the recommender training and evaluation. Following on from the above, he made versions of the iPlayer recommenders which use those sparse matrices and tuned them using Bayesian Optimization.

Continuing her exploration, Fearn finished off testing xvectors and started to think about applying audioset embeddings to recommending content.

Last week ee officially kicked off a collaboration to explore developing and using recommendation systems in public service media. The project aims to develop clear guidance for how the wider public sector should consider designing and implementing such systems. Miranda's running this from our side.

Explaining AI

Alicia and David were finishing off the initial design and build of a website to showcase better stock images of AI. Henry sent out a brief for these images to a number of artists, and started looking for more artists and photographers. David is also working up his own ideas for new images. Everyone's helping out in reviewing the images that have come from LCC students. More on this on the blog next week.

We all play-tested Galen's second AI-based game, he's also been polishing its interface and starting to think about a third game prototype. This one's theme is the AI in auto-correctautocorrect.

Elsewhere

Libby shaped up a project proposal around content moderation, simulations and social networks. We're planning a few futures workshops to test our process with other teams.

Lunches included a quesadilla from a van in Bristol and a Tesco's meal deal. And, in our team meetings, Ben Jackson from the University of Sussex showed us his work on the Â鶹ԼÅÄ's oral history archive, Libby demonstrated a robot rabbit, Sean explained stack machines and we tried out together.