I indulged most of my childhood watching Japanese sci-fi anime like A Farewell to Weapons and Ghost in the Shell in Beijing, my hometown. When I came to the U.S. at the age of 15, I was attracted to the field of artificial intelligence on Kaggle. I was fortunate to study computer science at Carnegie Mellon University, where I learned rigorous problem-solving skills. I'm interested in mapping components of business decision-making processes to intelligent systems from the silicon up to complex software. I'm a member of the CMU automated learning systems group (Catalyst) and Neo. I'm currently working on Praxis, helping teams to integrate large, difficult external datasets.
I indulged most of my childhood watching Japanese sci-fi anime like A Farewell to Weapons and Ghost in the Shell in Beijing, my hometown. When I came to the U.S. at the age of 15, I was attracted to the field of artificial intelligence on Kaggle. I was fortunate to study computer science at Carnegie Mellon University, where I learned rigorous problem-solving skills. I'm interested in mapping components of business decision-making processes to intelligent systems from the silicon up to complex software. I'm a member of the CMU automated learning systems group (Catalyst) and Neo. I'm currently working on Praxis, helping teams to integrate large, difficult external datasets.





















Google Scholar    ORCID    Semantic Scholar

My research aims to further our understanding of neural-network-powered intelligent systems and how they could be more efficiently and reliably built in the resource-constrained world.

  • To empower real-world use cases, we need to better understand neural network properties to align them to our intuitions.
  • To build intelligent systems more efficiently, we need to deeply understand system and hardware instantiations of different types of neural networks and utilize various forms of tapped and untapped computational resources.
  • To make intelligent systems more reliable, we need to elucidate the role of data, factual or fictitious, in the end-to-end lifecycle of algorithmic development, benchmarking, delivery, and maintenance. 

Concretely, I’m interested in a wide range of topics in computing, including (but not limited to; percentage in parenthesis indicates my perceived depth of exploration in the literature):

  • (?%) foundation models in entertainment
  • (60%) a communication-optimized scheduler enabling fine-grained chunking of neural network workloads 
  • (5%) an incentive structure motivating a decentralized mesh of consumer compute 
  • (30%) an information-theoretic framework to identify adversarial/passive gradients in said structure
  • (25%) proof of neural network computation (forward & backward) in privacy-preserving & succinct scenarios 
  • (80%) datasets that train better neural networks and better benchmarks for neural networks 
  • (?%) curation & watermarking of data artifacts that are distinctively non-generated

Some more generic ones:

  • make neural networks run faster, sometimes on weird hardware
  • tradeoff space of many things we care about in machine learning (accuracy, "privacy," etc.) 
  • using neural networks to super-charge resource-limited fields, sometimes non-STEM

If you have any questions regarding our work or would like to collaborate, please send all correspondences to peiyuan [at] praxispioneering [dot] com.

selected publications
 services