Data Science During Covid-19
Panel debate at Toronto Machine Learning Summit 2019
I am a community organizer for a group of 7,500 machine learning professionals. I work full-time organizing conferences, seminars, workshops and social initiatives. We began gathering in 2016 as a group of 30 people in the lobby of a co-working space. The meetups grew rapidly and in 2019 we were hosting events for 500 to 2,000 data scientists, machine learning engineers, data architects, researchers and entrepreneurs.
Not surprisingly, Covid-19 has rocked our world.
Not only has COVID put a complete stop on these kinds of face-to-face gatherings, but there is no re-start in sight. When we do meet again in person, it will not be under the same circumstances. It will be with measured caution, ample spacing of seating, reduced room capacity, and strict procedural guidelines. Inevitably will see elevated costs for event insurance and decreased availability of sponsorship.
In short, it will become very expensive to run large gatherings.
From a community organizer’s perspective, this is a major setback, but in truth, prior to COVID, we were already seeing trends that indicated a change was in order. COVID has merely accelerated a shift in the marketplace for data practitioners, and companies approaching AI.
Also, it has forced us to re-assessed the effectiveness of our gatherings, and adopt new ways for our community to meet and network. The good news is, these changes offer new opportunities for those looking to build skills and network effectively with peers. It just takes some creativity and an open mind.
Covid-19 Market Conditions for Data Practitioners and AI Teams
2019 Career fair
Each year, thousands of new AI embedded products enter the market, and companies get a more realistic understanding of the challenges and complexities of putting models into production. Still, companies will underestimate the time requirements, financial costs, and the nuanced team roles that are required to create AI products. Most proof of concepts (POC’s) do not make it off the ground.
The lack of commonly shared principles, methodologies and best practices mean that level setting stakeholder expectations, in particular during Covid-19, is an ongoing process, fraught with micro-management and increased accountability. Understanding the complexity of machine learning in production is still in its early stage. For many, it feels as though we are still in the wild wild west.
Throughout our community discussions and committee gatherings, two main themes have continued to dominate agendas, and have increased in importance during Covid-19 times. For technical teams, emphasis on constructing end-to-end ML pipelines and model governance takes center stage. For non-technical teams, understanding the scope of ‘change-management’ is the biggest reported hurdle.
This is reflected throughout the hiring market. In November 2019 we ran an AI career fair 70 companies. Start-ups, scale-ups and large enterprises were hiring for various roles, mostly labelled as data scientist positions, although expectations varied.
Despite the titles, almost all positions put emphasis on software development and engineering skills, with hopes that candidates had experience putting predictive models into production environments.
What does this mean for data practitioners?
If we were seeing decreased appetite for data science “generalists”, there is even less so now in 2020 due to Covid-19.
Whether you are a data practitioner, scientist, engineer, analyst etc. it’s become much more important to understand how your work fits into the larger picture and cohesive team structure.
Often, job-descriptions do not reflect the real expectations and requirements of the hiring companies. The need for reproducibility, transparency, model explainability, established methods for clear code documentation, emphasis on reducing pipeline debt, understanding computational costs, and the myriad of other potential project-killers are seldom listed, let alone mapped out and planned for.
“AI-first” companies like Google, Amazon, Netflix have had the opportunity to play around with team formations, and experiment with processes that can work around these challenges. Traditional companies continue to struggle.
Your up-skilling and networking?
Brain-date session at break
To help further AI projects, companies spend thousands of dollars sending practitioners around the world to attend conferences, with costs for premium events that can range from $600-$1,299 USD
COVID has accelerated a switch towards virtual gatherings and removed many of these barriers to entry.
Virtual events are hosted from all over the world and reflect new and different markets, as well as, diverse AI ecosystems. This opens the door to a variety of specialty themed meetups. Interested in cloud architecture? Join the Cloud Architect meetup in Singapore. Want to learn about Reinforcement learning? Join the Reinforcement Learning reading group in Oslo. Working on ML in production; Join the ML Ops, Production, and Engineering conference.
Face-to-face networking at these events is a huge part of the value proposition. In virtual settings, conference hall conversations are replaced by digital conference platforms offering targeted virtual matchmaking and open attendee lists for direct private messages.
Lot’s of problem-solving can be addressed using Reddit, Google or Youtube content, but speaking with somebody who has tackled a similar project, and understands the nuances of your particular use-case, is still possible during home isolation. It just takes patience, and open-mind as you develop online relationships with peers and embrace virtual platforms.
Virtual events are offering many benefits that traditional events cannot:
Increased access to participate in Q+A sessions
Social platforms allow for more networking interactions at scale
Conference ‘hours’ that were 1–2 days are now offered weeks leading up to the events
Opportunity to participate, or host virtual groups and themed networking sessions
Lots of these tools already existed, but we are seeing them embraced at unprecedented levels.
Getting familiar with these tools, and using them to your advantage is a worthwhile investment. What has worked successfully during social distancing times, will continue to be incorporated and embraced by data science communities in the form of physical/virtual hybrid events.
If you are open to the idea of connecting on these new virtual platforms, get a head start and give it a try. There are lots to gain!
We wish you luck.
Talks from our past machine learning events have been hosted on Towards Data Science’s official YouTube page. You can also attend our upcoming events here