EVENT - Thursday
Location1 Curtain Place, London EC2A 3AN
When6:00 PM - 9:00 PM
DSF Meetup with River Island
6:45: Martin Speed and Gareth Jones
7:05: Sophie Shawdon
7:25: Virginie Bonnefond
7:45: Michael Leznik
8:05: Networking, food and drinks
8:45-9:00: Leave the building
Martin Speed – Safety and Loss Programs Manager at River Island
Bio: Martin’s PhD is in criminology and he built his experience building crime risk models and researching for the sentencing advisory council, the Home Office, and the Ministry of Justice.
Gareth Jones – Safety and Loss Analyst at River Island
Bio: Gareth has a Master’s in Maths and just completed a Master’s in Data Science. He has been working as a senior analyst at River Island for 5 years and his previous modelling projects included predicting Stock Loss at stores. With Martin he has formed the new River Island Data Science team.
Summary: as a recently created data science team, our focus is to provide actionable insight quickly. Size gaps on the shop floor that do not currently get replenished represent a £2.5m opportunity and our RFID data can now identify exactly where they occur. Modelling the factors associated with the gaps occurring gave insight into how the issue could be addressed.
Sophie Shawdon – Senior Data Analyst at ClearScore
Bio: Sophie is the lead analyst on ClearScore’s international
business, and is passionate about helping our new to credit and
perceived ‘riskier’ customers gain access to and learn to manage
credit. Since joining the business three years ago, she has worked
with bureaux and credit providers around the world, and has seen
firsthand not just where the algorithms can go wrong, but also how
they can be made right.
Summary: Talk Data To Me: Using Machine Learning to Tell Customer Stories. As businesses become more number-driven, how do you ensure that the qualitative data – and in particular what your users are telling you – does not get left behind? In this talk, Sophie will talk through ClearScore’s recent work on using machine learning to better understand its customers; and what we learned; and why there’s so much value in ‘dormant’ data.
Virginie Bonnefond – Data Scientist at Hummingbird Technologies
Bio: French Signal and Image Processing Engineer graduated from ENSEEIHT Toulouse in 2017. I joined Hummingbird Technologies in the early days when we were 8 employees and we are now over 60. My work has been mostly concentrated on developing conventional and Deep Learning Computer Vision based products from prototyping to production based on UAV and Satellites images.
Summary: The recent progress in deep learning is slowly but consistently shifting the paradigm that was followed in remote sensing data processing over the last decades. AI applications are starting to become available for a wide range of applications, from on-board data reduction to fine-grained analytics. Agriculture is one of the areas with the largest social impact of this technological revolution, through a two-pronged strategy. Firstly, AI technology is used to replace the typical flat-rate applications of chemicals with targeted applications only to regions of need, thus alleviating the environmental stress caused by intensive farming practices. Secondly, AI is used to substitute agronomic input across the season, something particularly meaningful in parts of the world where access to agronomical input is limited. In this talk, we are going to present in detail the related topics, while also discussing the technology bottlenecks, limitations and future directions.
Michael Leznik – Data Scientist at Product Madness
Bio: As the Head of Data Science, Michael oversees all scientific data related analysis in the Product Madness. For the last ten years, he helped to create and managed interdisciplinary and data science related teams. He was Chief Research Scientist at one of the UK’s leading agencies Greenlight, spent a substantial period of time at King where he helped to build data science from almost ground zero, managed the London studio and later marketing data science teams. In between, he spent some time consulting digital real money casino companies. He holds a PhD degree in Operations Research from the Hertfordshire University where he also used to teach several programming and numerically oriented disciplines. Working in several startups in the digital marketing area helped him to be on the cutting edge of marketing technology, and observe the dramatic technological developments of the market for the last 20 years. Michael lives in the UK but also worked and taught in several different countries throughout his career. He enjoys reading, watching films, and British ales
Summary: Bayesian Decision making becoming the widely adopted methodology of choice in the industry. Ability to make your conclusions based on probability distributions instead of arbitrary binary p-values seems to be much preferable when it comes to decision making. The methodology allows to accept null values and estimate statistical power. Using non-parametric Bayesian helps to deal with non-standard models and avoid reduction of statistical power when the probability distribution isn’t normal. The latter is a major drawback when using regular t-tests. Non-binary; significant vs non-significant conclusions provide decision-makers with a plethora of outcome and wide field of possibilities.