Recommenders / Personalization

Democratizing Deep Learning Recommenders Resources

Deep learning work is iterative, experimental, and often time consuming for new and established machine learning practitioners, data scientists, and engineers. Yet, when deep learning is applied to certain application domains, like recommenders, industry members have the potential to provide better predictions at scale over existing commercial recommenders.

As recommenders impact daily decisions regarding the music we listen to, what we read, how we connect via social media, and how we shop; industry members have the tremendous responsibility of building recommenders that are relevant and trustworthy.

Democratizing development with SDKs is core to NVIDIA’s DNA. 

GTC Fall Keynote: Recommenders — The Personalization Engine of the Internet 

At GTC Fall 2020, NVIDIA CEO Jensen Huang, announced the open beta of NVIDIA Merlin, an open source end-to-end framework that democratizes the development of large scale deep learning recommenders.

GTC Fall Keynote: Announcing NVIDIA Merlin Open Beta 

With NVIDIA Merlin, data scientists, machine learning engineers, and researchers can accelerate their pipelines for data ingestion, training, and deployment on GPU-accelerated recommenders.

As GTC Fall 2020 included over 1000 sessions, this blog post provides a distilled guide to support development of deep learning recommenders:

A Deep Dive into NVIDIA Merlin Recommendation Framework – GTC Session: overview provides insight into the feature engineering and preprocessing common to recommendation datasets, deep learning-based recommendation models to enable fast experimentation and production retraining, and how Merlin provides low latency, high throughput inference.

Announcing The NVIDIA NVTabular Open Beta with Multi-GPU Support and New Data Loaders: NVTabular is the Extract, Transform, Load (ETL) as a component of Merlin that addresses common data pipeline pain points. Learn how and why Merlin includes an iterable data loader and discover examples of how NVTabular data loaders optimized for PyTorch and TensorFlow accelerate workflows.

Accelerated Recommender Systems Training with NVIDIA Merlin Open Beta: HugeCTR is the training component of Merlin and has an enriched feature set to ease optimization and interoperability. Uncover how HugeCTR’s the latest updates reaffirms NVIDIA’s commitment to accelerating workflows for large-scale deep learning recommender systems. 

Winning the RecSys Challenge: NVIDIA’s interdisciplinary team won 1st place for their collaboration on predicting tweet interaction. Read more about their learnings from a technical perspective or watch the GTC session video

Discuss (0)