DSSVirtual at Venture Beat’s Transform 2021 took place virtually on July 12th and featured nine data science experts who shared their strategies to improve the current state-of-the-art and the future of machine learning and data science in the enterprise.
The conference brought together the most diverse community of tech leaders to give insights into AI and ML best practices in the most prevalent industries, such as media, healthcare, e-commerce, and finance.
Here's an event recap including the main takeaways and highlights. Let’s dive in!
The event starts with Yuling Ma, CTO at FreeWheel, talking about how data can connect media and advertising businesses to help regulate the viewer frequency and run the advertising platform more successfully. FreeWheel being a data-driven marketplace uses machine learning to increase the efficiency between its buyers and sellers.
Speaking about ML systems in the media space, Vijay Pappu who is a Senior ML Manager at Peloton, says that the important question we need to ask ourselves is - How to improve the ML models further even after it performs very well in the initial A/B testing? One of the simple ways he proposed is cohort analysis, which is being used at Peloton. This method is an easy way to visualize the data and targets efficiently so that teams can make decisions on including more features or parameters.
We also hear from Tim Yoo, Head of Analytics at Roku, and Sriram Subramanian, the Head of Data Science and Engineering at Condé Nast, talking about how to use data to improve the content acquisition cycle and recommendation system strategies respectively. The speakers highlight that ML models are evolving with a higher amount of data as that facilitates the algorithms to make more accurate assertions. Imagine you can turn that power to advertising and show advertisements precisely to the right people at all the right times. If the ads become more and more relevant, an increasingly large number of viewers click on the ads and buy your products. The longer the campaign runs, the more your profits would increase and the less you could spend on ineffective ads.
Out of all industries, healthcare is a sector where machine learning can have the most impact. From the healthcare perspective, the event has Saira Kazmi, a Senior Director at CVS Health, to discuss the model development lifecycle of an ML project. She says that understanding business problems should be the initial step to define what and why we are trying to automate.
Once the problem statement is clearly defined, then comes the design phase, data annotation, model training, and finally the deployment phase. She discusses the tools dedicated to making this entire process easy to be deployed. The deployment phase can be the most challenging part since teams have to work on containerization, artifact management, and serving frameworks.
Whenever it makes sense to use a pre-built model at any stage, it should be used to its full potential.
Almost every major e-commerce and entertainment platform has recommendation systems to attract more customers. Stitch Fix is one of those companies that leveraged their highly actionable data and turned it into their main focus. June Andrews, Data Science Manager of Style Discovery at Stitch Fix, shares how predictive algorithms helped stylists to successfully serve their clients.
The challenge, in this case, was to figure out if a style would perform well before they get it onto the ecosystem. To solve this, Stitch Fix focused on building a recommender algorithm on millions of data rather than billions. By keeping the data minimal, they developed a system that can handle the complexity. By overspecifying a system, it may cause damage in the long run.
Talking about computer vision, Appu Shaji, CEO and Chief Scientist at Mobius Labs says that more than 50% of our brain cortex is devoted to visual information and hence it is an important area to be tackled. Their team is mainly focused on easy feature mapping, building efficient algorithms, and creating new models with very little training data.
He also stresses that learnability should be the key to move forward as we aim for faster and better AI systems. MobiusLabs also released Superhuman Vision, an AI-powered computer vision platform, which is a no-code training interface that should make it easy for everyone to add computer vision to their applications.
There is no doubt in saying that machine learning brought a lot of changes to how enterprises fundamentally operate. But there are also challenges in ML that could affect the workflow. While talking about innovation in machine learning, Reed Peterson, Field CTO Telecom Strategy at DataStax, shares two main challenges that need to be dealt with. The first is limited data sources and the second being timeliness of the data. This means that training models are affected by data quality which doesn’t have any value.
Fortunately, DataStax provides flexible APIs for any kind of data model. For example, Cassandra is a hybrid, multi-cloud server that replicates the data in real-time and deals with the pipelines’ complexity.
Dun & Bradstreet is another company leveraging its valuable data sources. Since it is a 200-year-old company, they have an advantage in terms of data availability. Rochelle March, Head of Analytics at D&B, discusses how ESG data has matured and proven to be valuable by investors and other stakeholders. D&B uses its vast data cloud business information to provide ESG intelligence and best NLP practices.
The main takeaway from the event was that automation is more achievable than ever, with the vast amount of data available and companies constantly trying to speed up every possible step in their production process.
You can register for free to access all on-demand sessions from DSSVirtual at VBTransform 2021 here.