Since just‐in‐time (JIT) has considerable overhead to detect hot spots and compile them at runtime, using sophisticated optimization techniques for embedded devices means that any resulting performance improvements will be limited. In this paper, we introduce a novel static Dalvik bytecode optimization framework, as a complementary compilation of the Dalvik virtual machine, to improve the performance of Android applications. Our system generates optimized Dalvik bytecodes by using Low Level Virtual Machine (LLVM). A major obstacle in using LLVM for optimizing Dalvik bytecodes is determining how to handle the high‐level language features of the Dalvik bytecode in LLVM IR and how to optimize LLVM IR conforming to the language information of the Dalvik bytecode. To this end, we annotate the high‐level language features of Dalvik bytecode to LLVM IR and successfully optimize Dalvik bytecodes through instruction selection processes. Our experimental results show that our system with JIT improves the performance of Android applications by up to 6.08 times, and surpasses JIT by up to 4.34 times.

This paper presents a hierarchical dispatcher architecture designed to efficiently schedule the execution of multiple deep neural networks (DNNs) on edge devices with heterogeneous processing units (PUs). The proposed architecture is applicable to systems where PUs are either integrated on a single edge device or distributed across multiple devices. We separate the dispatcher and scheduling policy. The dispatcher in our framework acts as a mechanism for allocating, executing, and managing subgraphs of DNNs across various PUs, and the scheduling policy generates optimized scheduling sequences. We formalize a hierarchical structure consisting of high-level and low-level dispatchers, which together provide scalable and flexible scheduling support for diverse DNN workloads. The high-level dispatcher oversees the partitioning and distribution of subgraphs, while the low-level dispatcher handles the execution and coordination of subgraphs on allocated PUs. This separation of responsibilities allows the architecture to efficiently manage workloads in both homogeneous and heterogeneous environments. Through case studies on edge devices, we demonstrate the practicality of the proposed architecture. By integrating appropriate scheduling policies, our approach achieves an average performance improvement of 51.6%, providing a scalable and adaptable solution for deploying deep learning models on heterogeneous edge systems.

Smart TV is expected to bring cloud services based on virtualization technologies to the home environment with hardware and software support. Although most physical resources can be shared among virtual machines (VMs) using a time sharing approach, allocating the proper amount of memory to VMs is still challenging. In this paper, we propose a novel mechanism to dynamically balance the memory allocation among VMs in virtualized Smart TV systems. In contrast to previous studies, where a virtual machine monitor (VMM) is solely responsible for estimating the working set size, our mechanism is symbiotic. Each VM periodically reports its memory usage pattern to the VMM. The VMM then predicts the future memory demand of each VM and rebalances the memory allocation among the VMs when necessary. Experimental results show that our mechanism improves performance by up to 18.28 times and reduces expensive memory swapping by up to 99.73% with negligible overheads (0.05% on average).

We propose a method to estimate the classification accuracy of a machine learning model without requiring any ground-truth labels, by leveraging the agreement between two models on unlabeled data. Unlike prior work that interprets model agreement as an upper bound or average of model accuracies, we are the first to demonstrate that in heterogeneous, label-free settings, agreement reliably approximates the performance of the weaker model in the pair. Our method is architecture-agnostic and does not require any labeled data, assumptions on model calibration, or prior performance information. We introduce a principled estimator that combines hard-label agreement, probability-level consistency, and a correction term for class imbalance and calibration bias. This estimator remains robust across diverse model types—including convolutional networks and Transformers—and performs reliably on standard image and text classification benchmarks. Experimental results confirm that our estimator closely approximates the accuracy of the weaker model in the pair, often within 1–2% of the ground-truth. Our approach operates entirely on in-distribution unlabeled data, offering a practical and reliable solution for model evaluation in real-world scenarios where labeled validation sets are unavailable.