Although significant efforts have been made into tackling statistical data heterogeneity,the diversity in the processing capabilities andnetworkbandwidth ofclients,termedassystemheterogeneity,hasremained largelyunexplored.

Federated Learning (FL) has been gaining significant traction across different ML tasks, ranging from vision to keyboard predictions. In large-scale deployments, client heterogeneity is a fact and constitutes a primary problem for fairness, training performance and accuracy. Although significant efforts have been made into tackling statistical data heterogeneity, the diversity in the processing capabilities and network bandwidth of clients, termed system heterogeneity, has remained largely unexplored. Current solutions either disregard a large portion of available devices or set a uniform limit on the model's capacity, restricted by the least capable participants.In this work, we introduce Ordered Dropout, a mechanism that achieves an ordered, nested representation of knowledge in Neural Networks and enables the extraction of lower footprint submodels without the need for retraining. We further show that for linear maps our Ordered Dropout is equivalent to SVD. We employ this technique, along with a self-distillation methodology, in the realm of FL in a framework called FjORD.

Federated Learning (FL) has been gaining significant traction across different ML tasks, ranging from vision to keyboard predictions. In large-scale deployments, client heterogeneity is a fact and constitutes a primary problem for fairness, training performance and accuracy. Although significant efforts have been made into tackling statistical data heterogeneity, the diversity in the processing capabilities and network bandwidth of clients, termed system heterogeneity, has remained largely unexplored. Current solutions either disregard a large portion of available devices or set a uniform limit on the model's capacity, restricted by the least capable participants.In this work, we introduce Ordered Dropout, a mechanism that achieves an ordered, nested representation of knowledge in Neural Networks and enables the extraction of lower footprint submodels without the need for retraining. We further show that for linear maps our Ordered Dropout is equivalent to SVD.

Zero emissions and, soon, zero crew: the world's first fully electric autonomous cargo vessel was unveiled in Norway, a small but promising step toward reducing the maritime industry's climate footprint. By shipping up to 120 containers of fertilizer from a plant in the southeastern town of Porsgrunn to the Brevik port a dozen kilometres (about eight miles) away, the much-delayed Yara Birkeland, shown off to the media on Friday, will eliminate the need for around 40,000 truck journeys a year that are now fuelled by polluting diesel. "Of course, there have been difficulties and setbacks," said Svein Tore Holsether, chief executive of Norwegian fertiliser giant Yara. "But then it feels even more rewarding to stand here today in front this ship and see that we were able to do it," he said, with the sleek blue-and-white vessel moored behind him in an Oslo dock, where it had been sailed for the event. The 80-metre, 3,200-deadweight tonne ship will soon begin two years of working trials during which it will be fine-tuned to learn to manoeuvre on its own.

Zero emissions and, soon, zero crew: the world's first fully electric autonomous cargo vessel was unveiled in Norway, a small but promising step toward reducing the maritime industry's climate footprint. By shipping up to 120 containers of fertilizer from a plant in the southeastern town of Porsgrunn to the Brevik port a dozen kilometres (about eight miles) away, the much-delayed Yara Birkeland, shown off to the media on Friday, will eliminate the need for around 40,000 truck journeys a year that are now fuelled by polluting diesel. "Of course, there have been difficulties and setbacks," said Svein Tore Holsether, chief executive of Norwegian fertiliser giant Yara. "But then it feels even more rewarding to stand here today in front this ship and see that we were able to do it," he said, with the sleek blue-and-white vessel moored behind him in an Oslo dock, where it had been sailed for the event. The 80-metre, 3,200-deadweight tonne ship will soon begin two years of working trials during which it will be fine-tuned to learn to manoeuvre on its own.

"Everything should be made as simple as possible, but no simpler." Designed Intelligence is Fjord and Accenture's approach to unlocking the full potential of human collaboration with AI. In our previous articles we talk about how AI technologies can help augment strategic decision making and build better experiences. Empowerment is the third pillar of Designed Intelligence and focusses on how design can make intelligent systems more transparent, more adaptable and ultimately more resilient. We live in a world of increasing complexity.

During WIRED Live 2017, Accenture and Fjord ran a Live Innovation looking at the ethics of AI and how we can address the consequences of badly designed AI, to co-create an AI ethics manifesto. Take a look at the result here.