On America’s Independence Day, we enjoyed some great presentations from Stephen Redmond and Jim Dowling at another fantastic Data Science Ireland meetup.
Making the trip over from Sweden, CEO of Logical Clocks Jim Dowling spoke to us about creating the world’s fastest and most scalable Hadoop distribution platform Hops and how they make the machine learning practitioner experience as close to the desktop data scientist experience with a Python-First platform for writing machine-learning workflows on Big Data.
Jim told us about the importance of deep learning in creating Hops and how PySpark helps them run it on many executors. However, this poses the questions how do you get Python Libs on these executors? The solution is to write all your documents on dockerfile in the cloud and set up Python in that environment which then allows you to describe the cluster you want and create your projects.
Jim also spoke on how to debug applications using Tensorboard and how to examine logs, monitor training and inferencing with parallel experimentation and Tensorflow Model Serving. In summary, Jim said the future of deep learning is distributed and that Hops is a new Data Platform with first-class support for Python, Deep Learning, ML and Data Governance and GPUs. For more about Jim’s thoughts on the future of deep learning, you can read more at: https://www.oreilly.com/ideas/distributed-tensorflow
Next up we had a guy who is well used to speaking regularly on anything data as Accenture’s Big Data and AI lead Stephen Redmond presented on how data visualisation is vital for data scientists from the humble histogram to complex multi-dimensional visualisations. He presented some of his work from his thesis on a method of displaying bipartite graph data that allows both the data scientist and the business user to discover relationships in their data.
One of most interesting parts of Stephen’s presentation was the unsupervised clustering and how as a business user you can investigate and allow automation to determine what clusters are there. Stephen described how key terms can be extracted using graph-of-words and how each were matched to term nodes which allows the weight-centroid of the matched nodes to be calculated. This allows you to get an absolute result on your data with the highest k-NN score. His example was to ask the question: who wrote the Acts of the Apostles? By using this method, it was discovered that Luke was the most likely to have done so.
We hope all our guests really enjoyed the meetup and that you took something away from Jim and Stephen’s talks. Apologies about having to cut short Eric Risser’s presentation on how AI is disrupting the future of content creation for games but we are delighted that he has agreed to return for the next one on Wednesday 1 August.
You can register for the next meetup at this link and we hope to see you all back again at Huckletree D2 next month.
AI Awards 2018: Applications Now Open
Connecting Ireland’s AI ecosystem for a higher purpose
Applications are now open for you to take part in the inaugural AI Awards and showcase your skills and expertise to the AI community on the island of Ireland.
The AI Awards is a not-for-profit organisation dedicated to promoting, advancing and rewarding corporate and academic excellence in Artificial Intelligence (AI) innovation, research & development and products.
It is our mission to support the AI community by recognising the hard work and dedication of those working in the field and to help promote and market Ireland globally as the first-choice home for AI investment and job creation in research and product development.
For more information about the awards,head over to www.aiawards.ie