PAPIs Europe 2018

What is recognizable about a particular artist’s style? What parts can be delegated to an assistant? Can AI play the role of assistant? Of collaborator? How would we get enough training data? How little could we get away with?



Exploring these questions using GANs, image to image translation and extreme augmentation of very small data sets, a series of experiments in human/AI collaborative cartoon drawing provides a gentle introduction to machine learning and computer vision while presenting augmentation and rule based techniques for the more experienced user.

Machine learning has been making headlines with its sometimes alarming progress in skills previously thought to be the preserve of the human. What will become of music now that machines are capable of composing it? In our engineering and creative research, we are applying deep learning methods to model transcriptions of traditional music of Ireland and the UK. These models, trained on over 23,000 tunes, can then become creative partners in making new music. Several composers have worked with our models to compose dozens of new pieces, both within and outside the conventions of the traditional music on which the models were trained. This shows the exciting potential for such machines for augmenting human creativity, and opening new avenues for music practice.

At Adobe we have more than 500 engineers and data scientists working on features that use machine learning and AI. In fact, AI is not just the fastest growing skill set, it is also one of Adobe's four innovation drivers. With Adobe Sensei, we want to democratize AI, so that intelligence can be part of every app, every tool, and every experience. The way to make AI scale at Adobe: APIs. We share our lessons learned in weaving AI into all our technology, what it takes to build an API layer for AI, and how to market AI at a Fortune 500 scale.

Artificial intelligence is instrumental in changing the way new generations will interact with the world around them. In nearly every field, great new technologies have emerged that benefit society in both practical and personal ways ranging from self-driving cars to smart bot opponents in video games. However, the field that could be reaping the most benefit is also the most reluctant to accept this new wave of technology. The medical field stands to save more lives, spend less public money on data analysis, and free up physician time for more patient interactions among many other benefits. Yet, the most we have seen by way of widespread incorporation of smart technologies in the system was a failed experiment called MYCIN which recommended antibiotics based on bacteria identification. We will focus on the barriers to moving out of the research phase into more practical applications, current research in the field, and future steps needed to advance health technologies.

This is a "Data from the trenches" story about optimizing infrastructure inventory in the energy sector by automatically flagging unnecessary field visits. We will explain how to turn an unsupervised problem with messy data into a successful project used on the field. We will discuss challenges we met along the way: working with business experts to create a training set from disparate structured and unstructured data sources; and how to implement unsupervised algorithms in a scalable way. Finally, we will show that messy data can be turned from a hindrance into an opportunity for data scientists.

I will present research that I conducted during my Ph.D. at University College London and in collaboration with Google. My primary interest lays in the development of neural architectures for 2D imagery problems in big scale. Will present the recently published analysis of different upsampling methods in the decoder part of visual architectures, together with last week ongoing extension for GANs. Will discuss attention mechanism for text recognition and review for what kind of application it can be useful (automatically updating Google Maps based on Google Street View imagery). I will explain the idea behind Inception and what had we change in inception-v3 to have it the best single model on ImageNet 2015 and how does it compare to Resnet architecture which was published 2 weeks after. Together with inception, will present our winning submission to MS COCO 2016 detection challenge and the extensive analysis of different models and backbone architectures inside. At the end will shortly review our UCL effort working with 4096x4096 images at The Digital Mammography DREAM Challenge for breast cancer recognition, where we have achieved 9th among 1375 teams worldwide and 2nd place in the community phase.

Ethics is increasingly becoming a buzz word in the world of machine learning. While small and large companies are continuing to innovate their operations with data-driven, predictive algorithms, it remains a challenge for companies to identify what ethical issues are relevant to them and how much risk the issues pose to their business.
In this talk, we share the findings and lessons learned from generating an ethics roadmap at the onset of integrating real-time predictions into the operations of Technical Safety BC, a safety regulator in Canada.