Sifeng Lin – Operations Research Scientist at DoorDash

TBD

TBD

Trupti Jadhav – Lead Data Scientist at Childrensalon

Collaborative and content filtering are traditionally used for the Recommender system but multiple domains like NLP, image recognition have seen great success with the deep learning models over traditional models. Amazon, YouTube are powered with complex deep learning systems to deliver very efficient recommender systems. In the fight against COVID-19, economic activities that require close physical contact have been severely restricted. In this context, e-commerce – defined broadly as the sale of goods or services online – is emerging as a major pillar. Most enterprises have started moving to e-commerce platforms to launch themselves in the digital marketplace. Successful e-commerce companies need to leverage personalization technology to give consumers a customized experience. Delivering personal experiences on e-commerce sites is achieved by dynamically showing content, product recommendations, and specific offers based on browsing behavior, previous actions, prior purchase history, customer demographics, and other enriched personal data. Hence the efficient, readily usable recommender system would be the need of the hour. I will explain the use case of how a recommender system can be built when an enterprise has a big product portfolio without explicit customer ratings for products. Additionally, I will talk about, how to build time-based modeling and testing a dataset for training a recommender system without introducing biases and data leakages, what will be the ideal way of measuring the recommendation system model’s accuracy and how entity embeddings improve the recommendation power significantly. I will showcase the comparative analysis of recommender systems over the period leading to improvements in recommendation with different AI algorithms.

Sayan Maity – Senior Research Data Scientist at Roku

How cutting edge Machine Learning techniques can be leveraged to solve the core business needs of expanding the customer base without impacting the brand perception and by minimizing fraud in the context of product based consumer model.

Briana Brown – Content Manager at SafeGraph

2020 was a watershed year for retail and ecommerce. Some brands and stores experienced sharp declines in revenue due to the economic downturn and social distancing; others saw demand for their products skyrocket, putting stress on supply chain operations as they struggled to keep up with rising consumer expectations. While the economy is now on the road to recovery, some of the shifts in consumer demand, behavior, and expectations will remain, creating a new normal for retail. To adjust to these changes, brands are increasingly turning to data science for answers. Geospatial information like points of interest (POIs), building footprints, and mobility data give retailers and ecommerce brands the tools to analyze how consumers interact with physical store locations, as well as places that may serve as leading indicators of demand to come. But one category of data in particular has been a gamechanger for brands as they adapt to a post-pandemic economy: industrial POIs. Industrial POIs refer to distribution centers, warehouses, and manufacturing facilities that are critical for supply chain analysis and demand forecasting, especially in a retail landscape increasingly dominated by ecommerce. Join geographer Briana Brown from SafeGraph as she describes the different types of industrial POIs and how they can be used for essential retail and ecommerce analytics.

Resham Sarkar – Principal Data Scientist at Slice

What if we all had a personal assistant who kept track of our likes and dislikes, when do we like what we like, where is the best place to get what we like? As we continue to ingest astronomical amounts of daily data, personalization has become a necessity for e-commerce companies for retaining high-quality customers and building trust in your brand. But first, we must understand our data. In this talk, I will show how we use physics to decode pizza at Slice and power personalization.

Edward Ratner – Founder & CEO at Edammo

Though in recent years the focus has been on big data, AI can provide critical in sights even when the amount of data is quite limited. In this talk, we will discuss a new approach to AI pioneered by Edammo. The technology provides very accurate models even when the number of training sample is between 100 and 10,000. We will discuss concrete use cases in several verticals including: image analysis, HR Tech, AI on IOT device and marketing/lead generation. This approach will enable many companies to leverage AI that currently can not. The new technology allows models to be created in seconds on standard desktop computers eliminating the need for expensive GPU clusters and other costly hardware.

Murium Iqbal – Senior Data Scientist at Etsy 

Two tower approaches have become prevalent in industry for both Search and Recommendations over the last few years. These methods employ metric learning to enforce a structure on an embedding space which captures a specific type of similarity. Fast retrieval via approximate nearest neighbor look-ups is then available in real time. We will review the loss functions and sampling strategies employed in industry to enable these methods, how they are deployed and why they are so powerful for information retrieval.

Dia Trambitas – Head of Product at JSL

Coming Soon. 

Brian Burns – Manager of Data Science & Analytics – Personalization & Outfitting at Nordstrom 

Nordstrom Digital Stylists create outfits to help our customers look good and feel great. They are asked to create outfits for several reasons: to serve an individual customer, to contribute to a thematic curation, to showcase an individual product, and more! During peak events with lots of new inventory such as the holidays and large sales, demand for their expertise can be enormous. In service of helping our stylists, we have created a machine learning based outfit creation/completion service leveraging Nordstrom’s extensive dataset of expertly created outfits and a hybrid graph based and representation learning approach. Given an initial item or set of items, we create outfits out of available inventory that are difficult to decern from those created by stylists, even by our stylists themselves. Join this talk to hear more about Nordstrom’s approach and results from their recent Anniversary Sale.

TBD

TBD