Coming together: The 3rd FARFETCH Data Science Summit

FARFETCH is a data-driven company, meaning data is the heart and soul of our community. Thus, to "spread the data word”, we created a special place to share and discuss the fantastic work done last year: our Data Science Summit! Prior to the global pandemic, we held these events twice a year. However, COVID meant we had to postpone them in 2020 and 2021. This year we decided to embrace the challenge of organising a hybrid summit that reunited our data science community.

At FARFETCH, we have 74 data scientists spread across eight offices in seven different countries. Despite embracing our Todos Juntos value, meaning "all together" in Portuguese, it’s not always straightforward to collaborate with colleagues across our vast network. So one key goal of our event was to motivate everyone to share their work, building a true data community. According to Erika Nitsch, a member of our event committee: "With so many new joiners over the last 20 months, it was the first time we got all of our data scientists across the company together. Even though the event was spread out across different countries, we had a sense of coming together to learn from each other for the first time.”

We wanted a global stage to present to the broader community what we had been working on, and our plans for the year to come. To do this we organised poster sessions and roadmap presentations. It was amazing to see 28 poster presentations distributed across four sessions, talking about the advances and projects developed last year. We also wanted to bring in exciting and challenging external speakers, to start compelling conversations, coupled with hands-on workshops where we could get our hands dirty with some state-of-the-art methods, #todosjuntos.

Organising all of this brought with it several challenges. We wanted everyone to have a chance to reach out to anyone, regardless of team, office or time-zone; additionally, the impact of COVID is felt at different degrees in different countries, so we had to be mindful of existing and potential health and safety guidelines in order to guarantee everyone's safety, while not restricting anyone’s access to the wonderful content we had planned. We embraced a hybrid model: every talk was done via Zoom, but we made arrangements so that we could listen at FARFETCH offices and enjoy coffee and lunch breaks together (and to high praise from our attendees, food was made available when possible!). Talks were recorded and immediately made available so that whoever wanted to catch up could do so without any major delay. Poster sessions had to be reimagined for this context as well; we opted for short, five-minute presentations (live or recorded) linked to a discussion thread on a dedicated slack channel, to make sure that discussions could happen asynchronously, meaning that your preferred location or time-zone didn't stop you from asking (or answering) any burning questions! Additionally, we made sure to mix the schedule up so that no matter where you were, you could listen live to a keynote talk, a masterclass or a poster presentation and follow discussions.

One of the core goals of the summit was to foster creativity by having events where we could come together to listen, inspire and challenge keynote speakers. We were extremely honoured to have hosted Prof. Ricardo Baeza-Yates, from the Institute for Experiential AI, to talk about Ethics in AI and Mounia Lalmas-Roelleke, director of research at Spotify, who shared how Spotify works to personalise and diversify the listening experience. We are very proud of the engaging discussions that followed (both live and on Slack) and of the ideas that were sparked from listening to these two very interesting speakers.

Of course, inspiration can also come from within, and that was definitely obvious throughout the talks about our Innovation team by Eugene Fisher, who pushed us to consider how innovation can be perceived through different lenses and showcased some great examples of innovative work being done at FARFETCH, and by Abie Cohen and Eva Robledo Sierra, who dove into the Machine Learning Platform and all the possibilities, it can open to data scientists at FARFETCH.

However, our best truly shone through in the poster sessions. We were absolutely blown away by how much great work is being carried out by our colleagues! Ben Lewis, Head of Data Science - Product, said, "It was an amazing opportunity for our Data Scientists to showcase the huge amount of research and development that goes on behind the scenes, generating really healthy dialogue across team members - we all learnt something new”. Beatriz Quintino, João Faria and Evgeny Sherkhonov, who co-authored a poster, added that "the poster sessions, in particular, were a great opportunity to engage with the FARFETCHData Community. We not only learned about the challenges faced by other teams and how they are tackling them, but we were also ableto share and give wider visibility to our team’s projects as well". The work showcased in these sessions included two contributions to RecSys 2021 from the Recommendations team, examples of how our Operations team are using predictive models to optimise returns and prevent fraud, plus several efforts in fashion image understanding and semantic search. Of course, as is usual in scientific conferences, we had a "best poster award”, won by Daniel Lin, Andonis Pavlidis and Kishan Manani, with a poster on how their team optimises the prices for poor performing products. In Daniel’s words: "The DS Summit and poster sessions were truly inspiring. It's always insightful to see the wide variety of projects that FARFETCHis tackling in data science and the creative approaches that our teams take. Sharing our great work in the Commercial DS team was a pleasure too!”.

We have to shout out the full list of winners of our poster competition:

Agility - How we use price recommendations to optimise prices for our poor performing products

Daniel Lin, Andonis Pavlidis and Kishan Manani

Fraud Detection

Ana Mendes, Mariana Henriques & Pedro Henrique Guimarães

Classifying Images with Fashion Taxonomy Attributes - dealing with Noisy Labels

Beatriz Quintino, João Faria & Evgeny Sherkhonov

The posters were a fantastic way to look back into what had been done over the past year. We also looked at our future together via a session where each team went over their plans for next year from a more technical point of view; this was a great way to, as Erika put it, "make connections and find people working on similar techniques, and take these connections back into their regular jobs”. Todos juntos in practice! 

We talked a lot about opportunities to engage with new perspectives and connect with other people, whilst also covering more day-to-day practical concerns. We organised workshops and masterclasses that allowed us to learn about new technologies and gave us space for in-depth discussions: Ian Ozsvald, co-founder of PyData London and a data scientist at Mor Consulting, spoke to us about high performance in Python followed by a direct conversation about those topics, plus inviting us to discuss what the challenges that are specific to data scientists. We also heard about Data Genie directly from the company founder Karthikeyan Muthukumar and had the chance to try out Transformers4Rec, the library that produced the winning solution of the SIGIR ‘21 eCommerce Challenge, in a workshop guided by Gabriel Souza and Ronay Ak, Senior Data Scientist and Senior Research Scientist at NVIDIA, respectively. Last but not least, we closed the session with a hands-on tutorial on the statistical method behind our internal experimentation platform – FABS, the FARFETCH AB testing Service – by our very own João Martins.

Overall, we are very encouraged by the feedback from the community. We asked our attendees and were pleased to see how much people enjoyed the variety of talks, topics and speakers, how much buzz the poster sessions generated and all the great suggestions for future iterations of the DS Summit – particularly, we recognise that we have to come up with better ways for people to socialise with each other (an obvious challenge in these times), and now fully understand how important pizza is for this sort of event. We had two intense days in which we got motivated, learned and showcased our work, but this is only the start of our journey. Ana Mendes, Mariana Henriques and Pedro Henrique Guimarães, the team who won second prize in our poster competition, shared a little bit about their experience at this edition: "It was an exciting experience for us! We had the opportunity to be in touch with such amazing and unique projects that are being developed at FARFETCH and also to better understand how our daily work contributes to the company's goal. We hope this initiative keeps occurring so we can keep sharing all the knowledge among the Data Community at FARFETCH.” Some of our posters will evolve into papers, and our community’s motivation will allow us to do what has never been done!