Partial Inference

Partial inference focuses on efficiently recovering only a portion of the complete solution to a complex problem, rather than computing the entire solution, thereby reducing computational cost and improving efficiency. Current research explores this concept across diverse applications, employing techniques like generative flow networks with learned energy decompositions, variational autoencoders for inverse inference, and mobile-cloud collaborative inference strategies that leverage compressed feature tensors. These advancements offer significant potential for improving the scalability and practicality of AI models in areas such as personalized medicine (e.g., cardiac digital twins), robotics (e.g., manipulation of deformable objects), and resource-constrained mobile devices.

Papers

October 5, 2023

Learning Energy Decompositions for Partial Inference of GFlowNets
Hyosoon Jang, Minsu Kim, Sungsoo Ahn
Generative Flow Network Flow Based Energy Based GFlowNets Energy Based Boltzmann Distribution Partial Inference

July 10, 2023

Towards Enabling Cardiac Digital Twins of Myocardial Infarction Using Deep Computational Models for Inverse Inference
Lei Li, Julia Camps, Zhinuo, Wang, Abhirup Banerjee, Marcel Beetz, Blanca Rodriguez, Vicente Grau
Deep Model Cardiac Magnetic Resonance Synthetic Electrocardiogram Cardiac Digital Twin Partial Inference

June 24, 2023

Mobile-Cloud Inference for Collaborative Intelligence
Mateen Ulhaq
Inference Latency Deep Learning Inference Collaborative Intelligence Mobile Inference Partial Inference

June 6, 2023

Partial Inference

Papers

Learning Energy Decompositions for Partial Inference of GFlowNets

Towards Enabling Cardiac Digital Twins of Myocardial Infarction Using Deep Computational Models for Inverse Inference

Mobile-Cloud Inference for Collaborative Intelligence

Partial Inference in Structured Prediction

Learning to Simulate Tree-Branch Dynamics for Manipulation