Ensuring recommenders are meaningful, personalized, and relevant to a single customer is not easy. Scaling a personalized recommender experience to hundreds of thousands, or millions of customers, comes with unique challenges that data scientists and machine learning engineers tackle every day. Scaling challenges often provide obstacles to effective ETL, training, retraining, or deploying models into production.
To tackle these challenges, machine learning engineers and data scientists within the industry utilize a combination, or hybrid of tools, techniques, and algorithms. NVIDIA is committed to help streamline recommender workflows with Merlin, an open-source framework that is interoperable and designed to support machine learning engineers and data scientists with preprocessing, feature engineering, training, and inference. Merlin supports industry leaders who are tackling common recommender challenges as they provide relevant, impactful, and fresh recommenders at scale.
Here a few key sessions from industry leaders in media, delivery-on-demand, and retail at GTC Spring 2021.
- Learn How Tencent Deployed An Advertising System On The Merlin GPU Recommender Framework
Learn how Tencent deployed their real advertising recommendation training with Merlin and achieved more than 7x speedup over the original TensorFlow solution on the same GPU platform.
- AI-First Social Media Feeds: A View From the Trenches
ShareChat discusses efficient training of extremely large recommender models with billions of parameters distributed across multiple GPUs and workers as well as the importance of continual model updates in near-real time to deal with the key challenge of concept drift. The session includes how solutions in NVIDIA’s Merlin stack resolve key bottlenecks faced in general purpose deep learning frameworks.
- Personalization and Discovery at Postmates
Covers the machine learning building blocks, relevant models developed in production, model optimizations to align with Postmates’ business objectives, and more.
- GPU-Accelerated Processing of Large Tabular Datasets and Distributed Neural Network Training with NVTabular+Dask+TensorFlow for Dynamic Pricing at Walmart
Demonstrates how to use open-source tools like Dask, NVTabular, and TensorFlow to horizontally scale up GPU-accelerated training pipelines for large neural networks trained using tabular data.
- End-to-End Deployment of GPU Accelerated Recommender Systems: From ETL to Training to Inference
Upon completion of this 4-hour in-depth workshop, attendees will be able to apply NVIDIA Merlin (NVTabular, HugeCTR and Triton Inference Server) to accelerate and scale their ETL, training and inference systems with GPUs.
Registration is free, visit the GTC website for more information.
