About Me

Overview

I currently work at Meta and lead the machine learning effort for Threads. I previously led Instagram's Home Feed team building large scale recommendation systems and worked on Meta Generative AI team focusing on video generations.

Before Meta, I worked at AWS on its machine learning platform Amazon SageMaker and NLP search service Amazon Kendra. I co-founded and worked at a couple of startups (Compass and Aperio.AI).

Before doing software and machine learning, I was a physicist and a semiconductor chip designer. I received my Ph.D. at the University of Washington and spent several years commercializing my research through a hardware startup Elenion, later acquired by Nokia.

I hold 9 US patents and published more than 80 peer-reviewed academic papers^{Google Scholar}.

You can contact me on LinkedIn or Instagram DM (@magicfeature).

Work Experience

Facebook / Meta

Ran led Instagram's Feed team working on the ranking and delivery of Home Feed -the largest product surface of Instagram and one of the largest recommendation system on the planet serving billions of users every day.

Ran worked on Meta's Generative AI, focusing on image and video generation before heading back to Instagram to lead the machine learning effort for Threads app shortly after its launch in July 2023.

Compass

Ran kickstarted Compass’ machine learning effort and built its first machine learning team. The team delivered several machine learning services and features end-to-end (real-time recommenders, search ranking, valuation, CRM). In the process, the team built the necessary machine learning infrastructure (event-driven feature store, training and serving pipeline) and defined machine learning development process.

AWS

Ran joined AWS AI to work on infinitely scalable machine learning algorithms and platforms in Amazon SageMaker. His main focus was neural variational inference and its applications in Natural Language Processing (NLP)^{AWS blog}^,^webinar. Later on, he became a founding member of Amazon Kendra, making state-of-the-art deep learning based semantic search and question answering available to the public through this new service.

Representative work:

Coherence-Aware Neural Topic Modeling (EMNLP'18)
Weakly Semi-Supervisited Neural Topic Models (ICLR'19, workshop)
Topic Modeling with Wasserstein Autoencoders (ACL'19)
Technical reports: Generative Models, Language Modeling, Neural Variational Inference

Aperio.AI

Aperio.AI was a solid-state LIDAR 3D sensing startup. The thesis was to use Silicon Photonics (SiPh) for LIDAR sensing along with integrated front-end circuits and computer vision algorithms to alleviate the stringent requirements for high-definition near real-time LIDAR sensing needs for applications in autonomous vehicles. This approach ultimately did not deliver the orders of magnitude improvements over conventional methods and as the startup was shut down.

PhD and Elenion

Ran Ding received his Ph.D. in EE from the University of Washington (UW), Seattle in 2014. With support from Intel, his Ph.D. research contributed to the creation of an open-access foundry platform OpSIS^{Nature article}, which made silicon photonics technology accessible to more than 150 research and industry institutions. For example, researchers at CalTech and MIT used chips fabricated at OpSIS for cutting-edge research in 3D LiDAR sensing, quantum computing, and deep learning computation acceleration.

Representative work:

During his PhD, Ran co-founded a fabless (Design-as-a-Service) startup commercializing silicon photonics technology. The startup was acquired in 2014 and later rebranded^{OVUM article} as Elenion, where he continued to lead a 15-person core technical group and delivered the companies first commercial product^IMRA in 2017. Nokia acquired Elenion in March 2020.

Research Interests

Ran's main interests in machine learning include natural language processing (NLP) (language modeling, topic models, semantic matching, text classification, etc), deep generative models (VAE, GAN, adversarially regularized autoencoders), and learning methods (weak supervision, transfer learning, meta/few-shot learning, domain adaptation).

Ran currently holds 9 US patents and has authored and co-authored more than 80 peer-reviewed academic papers^{Google Scholar}.