Tony Tran

Project Overview: This post introduces TrashDet, a family of efficient object detectors designed specifically for resource-constrained edge devices (TinyML). By employing an Iterative Neural Architecture Search (NAS) framework, we successfully deployed waste detection models on microcontrollers, reducing energy consumption by up to 88% while outperforming existing baselines on the TACO... [Read More]

Project Overview: This post explores the implementation of Alternating Direction Method of Multipliers (ADMM) for pruning lightweight Deep Neural Networks (DNNs). We compare this optimization-based approach against traditional magnitude-based pruning on the CIFAR10 dataset. [Read More]

Project Overview: This project analyzes traffic accident severity in Texas (2016-2023) using a suite of machine learning models. We compare interpretable linear baselines (Logistic Regression, LDA) against non-linear tree-based methods (Random Forest, LightGBM) and explore natural accident groupings using unsupervised K-Means clustering. [Read More]

Abstract: Deploying high-performance object detectors on TinyML platforms poses significant challenges due to tight hardware constraints and the modular complexity of modern detection pipelines. Neural Architecture Search (NAS) offers a path toward automation, but existing methods either restrict optimization to individual modules, sacrificing cross-module synergy, or require global searches that... [Read More]

Project Overview: This post explores the experimental application of 1-Lipschitz layers—typically used to certify robustness in classification tasks—to the more complex domain of object detection. We analyze the trade-offs between robustness, expressiveness, and accuracy using the SVHN dataset on a custom Tinier SSD architecture. [Read More]