The Deep Learning Summit returned today to London, accompanied by the Deep Learning in Healthcare Summit. The dual-track event once again brought together some of the leading experts in their fields to present and discuss the latest technological advancements and breakthroughs in a range of industries including healthcare, social networking platforms, finance, academia and more.
The summit started with networking and technical discussion taking place over breakfast between both first time RE•WORK-ers and regular summit attendees. Early dialogue focussed on desired outcomes from the summit alongside day one presentations which were of great interest:
“It’s my first time at a RE•WORK event and it’s been great so far, much busier than I expected. I work in medicine but I am really looking forward to the NASA talk as I have a great personal interest in space.” Tomasz Grzywalski, Machine Learning Engineer, StethoMe
The morning presentations got off to a strong start with Fabrizio Silvestri, Software Engineer at Facebook, recounting the embedding techniques needed to implement image recognition into the ‘real world’. The question on everybody’s lips was posed during the post-presentation Q&A which queried the influence of GDPR (General Data Protection Regulation) on image recognition going forward:
“The right to be forgotten is simple as there’s a list we cannot show, the biggest issue is when you decide your data is not for use after you have joined Facebook, but we do have models to remove this information” Fabrizio Silvestri, Facebook
Fabrizio’s talk was followed by Agata Lapedriza, Professor at Oberta de Catalunya, who discussed emotional recognition in images. The monumental advancement in Computer Vision was soon made clear by Agata, stating “The computer vision industry has improved incredibly over the last 6 years. Scene recognition can now be done with 75% accuracy, which 10 years ago was unheard of.”
The high standard of content and presentations was mirrored on the Healthcare track, which kicked off with discussion surrounding AI in the NHS and Clinically Applicable Deep Learning Techniques from Sarah Culkin of the NHS and Trevor Back from DeepMind Health respectively. Trevor’s current research includes that of cancer detection and predictive patient deterioration, with Trevor stressing the need to collaborate with the NHS and fellow health providers to make a substantial impact on the healthcare industry.
What else did we learn in this morning’s sessions?
“If you go from hospital-level data to regional level data, then can then show what happens on a larger scale and we can see that some really amazing things happened through that test, and we should expand this work further’ - Sarah Culkin, Strategic Data Lead, NHS
“Over 80% of all vision impairments could be avoided if detected earlier. This opened up a huge opportunity for Deep Learning to be applied.” Trevor Back, Product Manager, DeepMind
“We have figured out it will take 1141 years to create a detailed map of the moon” James Parr, NASA Frontier Development Lab
Alongside the presentations, interviews and networking sessions were interactive workshops, focussing on Ethics within AI, Predictive and Generative Deep Learning for graphs and a roundtable discussion including Machine Learning researchers, strategic data leads and postdoctoral research associates.
The Mapping workshop, run by Cansu Canca and Laura Haaber Ihle looked at how we can structure ethical problems at hand, delving into the underlying principles before attempting to solve the issues at hand. The engaging workshop involved attendee collaboration and assessment of practical implementation methods, gradually working toward the creation of practical solutions.
“Should search results be based on the prominence of an individual, or the number of other results. For example if you search ‘John Smith’, should a famous professor come up, or a list of other John Smiths? Is this ethical?”
Other workshops throughout the day posed questions including the potential need for alteration of search engine results to reduce bias, and whether search engine results should be based on the prominence of an individual, or the number of results available, with the example of famous sports stars maintaining prominence in search engine results over namesakes of equal importance. Tomorrow, we’re looking forward to learning from Accenture in their workshop session, ‘How to Design AI with Human Centricity’.
After breaking for lunch and more networking opportunities, the afternoon presentations continued with talks including multimodal and multilingual context models from Lucia Specia, Professor of NLP at Sheffield University, which preceded a discussion surrounding Custom Vision Services and their standing against manually developed machine learning models from Tom Szumowski at Urban Outfitters. Tom’s presentation included in-depth technical discussion and numeric representations of data, however, it was stressed that without domain knowledge and a real understanding of your product, the high-level models available will be significantly restricted. Tom’s talk also set a milestone for RE•WORK as the first presentation to be live streamed on the RE•WORK YouTube channel. If you missed the stream, fear not, as Tom’s presentation, alongside all other talks, will be uploaded to the RE•WORK Digital Content Hub post-event.
Further highlights from the afternoon presentations included Content Discovery with Semantic flow from Jian Li at Sky, Managing Alcohol Misuse by Automation from Daniel Leightley at King Centre for Military Health Research and Event Prediction with Deep Learning from Janahan Ramana of Borealis AI.
A number of podcasts and interviews were completed throughout day one of the events, further discussing presentation topics, current research and what to expect next! Day one was brought to a close with networking drinks for attendees, followed by the official attendee dinner and further discussion surrounding the presentations seen on day one! We will be back tomorrow with another day of technical discussion and presentations, including talks from Pipeline AI, MedicScan, Google Brain, Imperial College London and more.
For day two updates, make sure you’re following our official event Twitter accounts and hashtags on @reworkDL, @reworkHEALTH, and give our Twitter Ambassadors a follow @TuuliaTimonen, @DeepTech2_0, @verdictUK, @Yssybyl, @TechHub.