A5079905894- Privacy-Preserving Technologies in Data
- Stochastic Gradient Optimization Techniques
- Industrial Vision Systems and Defect Detection
- Embedded Systems Design Techniques
- Context-Aware Activity Recognition Systems
- Cloud Computing and Resource Management
- Advanced MIMO Systems Optimization
- Vehicle License Plate Recognition
- IoT and Edge/Fog Computing
- Advanced Optical Network Technologies
- Cryptography and Data Security
- Caching and Content Delivery
- Advanced Neural Network Applications
- Energy Efficient Wireless Sensor Networks
- Wireless Networks and Protocols
- Augmented Reality Applications
- Radio Frequency Integrated Circuit Design
The University of Texas at Austin
2020-2024
Polytechnic University of Timişoara
2020
This paper presents a hardware prototype and framework for new communication-aware model compression distributed on-device inference. Our approach relies on Knowledge Distillation (KD) achieves orders of magnitude ratios large pre-trained teacher model. The consists multiple student models deployed Raspberry-Pi 3 nodes that run Wide ResNet VGG the CIFAR10 dataset real-time image classification. We observe significant reductions in memory footprint (50×), energy consumption (14×), latency...
A central challenge in machine learning deployment is maintaining accurate and updated models as the environment changes over time. We present a hardware/software framework for simultaneous training inference monocular depth estimation on edge devices. Our proposed frame-work can be used co-design tool that enables continual online federated results show real-time performance, demonstrating feasibility of
Hierarchical Federated Learning (HFL) has shown great promise over the past few years, with significant improvements in communication efficiency and overall performance. However, current research for HFL predominantly centers on supervised learning. This focus becomes problematic when dealing semi-supervised learning, particularly under non-IID scenarios. In order to address this gap, our paper critically assesses performance of straightforward adaptations state-of-the-art FL (SSFL)...
This paper presents a new hardware prototype to explore how centralized and hierarchical federated learning systems are impacted by real-world devices distribution, availability, heterogeneity. Our results show considerable performance degradation wasted energy during training when users mobility is accounted for. Hence, we provide that can be used as design exploration tool better design, calibrate evaluate FL for deployment.
The recent developments in Federated Learning (FL) focus on optimizing the learning process for data, hardware, and model heterogeneity. However, most approaches assume all devices are stationary, charging, always connected to Wi-Fi when training local data. We argue that real move around, FL is negatively impacted device energy spent communication increased. To mitigate such effects, we propose a dynamic community selection algorithm which improves efficiency two new aggregation strategies...
We have built a hardware prototype specifically designed for teaching edge artificial intelligence (AI), enabling students to easily develop software devices. This complements the University of Texas curriculum with foundational AI course where can experiment real devices directly.
Most Federated Learning (FL) algorithms proposed to date obtain the global model by aggregating multiple local models that typically share same architecture, thus overlooking impact on hardware heterogeneity of edge devices. To address this issue, we propose a model-architecture co-design framework for FL optimization based new concept elasticity. More precisely, enable devices train different belonging architecture family, selected match resource budgets (e.g., latency, memory, power)...
RF system development is traditionally constrained by a restrictive trade-off between power efficiency and programmatic flexibility. We outline path towards achieving both, thereby enabling range of new concepts that better utilize limited resources. As an example, for many future applications, we consider convergence – reusing the same spectrum waveforms to achieve multiple distributed functions goals, simultaneously. To enable this next step in processing, develop novel framework includes...
This paper presents a method for detecting receipt fraud by implementing an Object Character Recognition (OCR) algorithm composed of Image Processing Techniques and Convolutional Neural Networks (CNNs). We implemented two CNN models into smartphone application that gives customers the option to take pictures products they intend buy (also crop their price tags) while present in hypermarket/supermarket as well paid succeeds automatically identify compare all prices (multiple digits including...