Adaptive Sparsity for Efficient and Reliable Machine Learning Inference in Autonomous Vehicles

Name: Danny Abraham

Chair: Prof. Nikil Dutt

Date: June 8, 2026

Time: 10:00 AM

Location: DBH 4011

Committee: Prof. Fadi Kurdahi, Prof. Mohsen Imani

Title: Adaptive Sparsity for Efficient and Reliable Machine Learning Inference in Autonomous Vehicles

Abstract: Autonomous vehicles rely on machine learning models that must make accurate decisions under strict latency, memory, and reliability constraints. However, modern deep learning systems are often designed as fixed, dense inference pipelines, even though real driving conditions vary across compute availability, scene difficulty, sensor structure, and modality access.
This dissertation presents adaptive sparsity as a framework for making autonomous vehicle inference more efficient and responsive to these changing conditions. It explores three complementary forms of sparsity: parameter sparsity for dynamically adjusting model capacity, activation sparsity for exploiting input-dependent computation, and input sparsity for using rich training-time modalities to improve deployment-time models with fewer inputs. Together, these methods show how sparsity can serve not only as a compression tool, but as a mechanism for adaptive, efficient, and reliable inference in autonomous vehicle systems.

Bio: Danny Abraham is a PhD student in the School of Computer Science at UC Irvine, working with Dr. Nikil Dutt, in the field of embedded machine learning. His current line of work is researching adaptive machine learning models for resource-constrained embedded systems, focusing on dynamic runtime reconfiguration. This enables real-time model adjustment and optimizes the tradeoff between accuracy, latency, and energy consumption particularly in the context of autonomous vehicles.