From robots amusing the audience with improvised poems to getting a deep dive into how AI is being used to optimise the cleaning of polluted beaches around the world, this year’s Dataiku Everyday AI Conference in London really set the bar high.
The conference kicked off with a morning plenary session where we learnt all about the vision that Florian Douetteau, Co-founder and CEO at Dataiku, has for the company, as well as some testimonials from businesses that have already introduced Dataiku to systematise the use of AI across their organisation. Following the first talk, the lead EMEA Evangelist at Snowflake, Eva Murray, hosted a thought-provoking panel discussion on democratising the data industry, emphasizing the need for data to be in the hands of more diverse people.
The last morning session was particularly inspiring. Creativity and science go hand in hand, and Improbotics is the perfect example of this. Unique and original ideas are often the outcome of combining passions from very different sides of the disciplinary spectrum. Improbotics have designed and developed an AI-based conversational agent (chatbot) that generates the lines of one of the performers in a scene. How? The improvisation chatbot is based on HumanMachine’s A.L.Ex. It uses the GPT-2 model which was retrained on movie dialogue for OpenSubtitles. During the performance, the operator of the chatbot types a sentence into it, which in turn generates sentences to be said for one of the performers. The result? A highly entertaining experience that left the audience in laughter, whilst also demonstrating the role AI might play in the future of experimental art.
Another theme of the conference was the upcoming release of Dataiku 11, and it’s promised to be a big one according to Iulia Feroli, Alex Alfille, and Serena Yuen, three key speakers at the event. New features include code studios, experiment tracking, visual deep learning, and managed labelling. So how can some of these features help your business?
Experiment tracking keeps track of all models in order to store and compare them. This feature provides a visual aid in the comparison of models, making it easier to choose the best current model as well as seeing general trends of all models. For example, you might note a particular measure is underperforming across models, which could be an indication of an inherent fault in the model or even the dataset itself.
The feature store is a dedicated zone where teams can access and share reference datasets containing features that might be suitable for others. It allows your company to get the most use out of its data, and to not lose valuable time redefining features. On top of that, a feature store allows for a central definition of a feature, eliminating the risk of separate teams defining and naming features differently, which in turn increases data quality.
Visual deep learning delivers object detection and image classification use cases out-of-the-box. This could be used for automatic document, passport or contract scanning and/or recognition, as well as for handwriting recognition. Computer vision is also used in e-KYC, an online version of KYC that reduces risk of fraud and identity theft by using secure digital identity based on computer vision facial recognition technology.
These new features add strong tools to the existing AutoML toolbox that is Dataiku, and the talks mentioned here are only a handful of the many insightful and inspirational sessions that formed the Everyday AI conference 2022. Dataiku has again shown its strong capabilities in ML, as to be expected from the re-awarded ML partner of the year by Snowflake, which was announced at the recent Snowflake Summit in Las Vegas.
In our opinion, the most interesting development revealed at the conference was the use case of Improbotics, and the engaging demonstration got us excited about the potential for natural language processing (NLP) in Financial Services. Overall, we thoroughly enjoyed the event and look forward to more to come.
As one of Dataiku’s partners, we have already used Dataiku in several use cases across Financial Services. If you would like to speak to us about how we can help you and your business, please get in touch with one of our experts here.