Starfish

Hu, Pan; Im, Junha; Asgar, Zain

doi:10.1145/3384419.3430769

Cited by 17 publications

(4 citation statements)

References 44 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The study [ 30 ] proposed an image compression framework built specifically for running in resource-constrained IoT devices. The solution uses a DNN that avoids inefficient image transmissions being projected to low-memory and processing IoT devices.…”

Section: Related Workmentioning

confidence: 99%

“…In such scenarios, multimedia data necessitate special considerations. Applying traditional codecs to compress multimedia data reduces the data’s quality, thereby impacting its performance when utilized by AI models, as evidenced by studies such as [ 30 , 34 ]. To address this issue, researchers are turning to deep learning approaches such as Convolutional Neural Networks (CNN) and Deep Neural Networks (DNN) for handling this type of data.…”

Section: Proposed Frameworkmentioning

confidence: 99%

“…To address this issue, researchers are turning to deep learning approaches such as Convolutional Neural Networks (CNN) and Deep Neural Networks (DNN) for handling this type of data. As an example, in their work, the authors of [ 30 , 34 ] employed Deep Neural Networks (DNN) to handle multimedia data. Opting to compress image data using traditional image compression codecs before transmission is not advisable, as this approach leads to the removal of crucial image structures that DNN models rely on.…”

Section: Proposed Frameworkmentioning

confidence: 99%

See 2 more Smart Citations

Towards an AI-Driven Data Reduction Framework for Smart City Applications

Pioli,

de Macedo,

Costa

et al. 2024

Sensors

View full text Add to dashboard Cite

The accelerated development of technologies within the Internet of Things landscape has led to an exponential boost in the volume of heterogeneous data generated by interconnected sensors, particularly in scenarios with multiple data sources as in smart cities. Transferring, processing, and storing a vast amount of sensed data poses significant challenges for Internet of Things systems. In this sense, data reduction techniques based on artificial intelligence have emerged as promising solutions to address these challenges, alleviating the burden on the required storage, bandwidth, and computational resources. This article proposes a framework that exploits the concept of data reduction to decrease the amount of heterogeneous data in certain applications. A machine learning model that predicts a distortion rate and its corresponding reduction rate of the imputed data is also proposed, which uses the predicted values to select, among many reduction techniques, the most suitable approach. To support such a decision, the model also considers the context of the data producer that dictates the class of reduction algorithm that is allowed to be applied to the input stream. The achieved results indicate that the Huffman algorithm performed better considering the reduction of time-series data, with significant potential applications for smart city scenarios.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Proposed Frameworkmentioning

confidence: 99%

Section: Proposed Frameworkmentioning

confidence: 99%

See 1 more Smart Citation

Towards an AI-Driven Data Reduction Framework for Smart City Applications

Pioli,

de Macedo,

Costa

et al. 2024

Sensors

View full text Add to dashboard Cite

show abstract

“…Therefore, Egeria adopts post-training quantization [24] to instantly generate a reference model with the same structure. [32,90,92]. It reduces the precision of model's parameters (e.g., from 32-bit floating-point to 8-bit integers).…”

Section: Generating and Updating The Reference Modelmentioning

confidence: 99%

Egeria: Efficient DNN Training with Knowledge-Guided Layer Freezing

Wang

Sun

Chen

et al. 2023

Proceedings of the Eighteenth European Conference on Computer Systems

View full text Add to dashboard Cite

Training deep neural networks (DNNs) is time-consuming. While most existing solutions try to overlap/schedule computation and communication for efficient training, this paper goes one step further by skipping computing and communication through DNN layer freezing. Our key insight is that the training progress of internal DNN layers differs significantly, and front layers often become well-trained much earlier than deep layers. To explore this, we first introduce the notion of training plasticity to quantify the training progress of internal DNN layers. Then we design Egeria, a knowledgeguided DNN training system that employs semantic knowledge from a reference model to accurately evaluate individual layers' training plasticity and safely freeze the converged ones, saving their corresponding backward computation and communication. Our reference model is generated on the fly using quantization techniques and runs forward operations asynchronously on available CPUs to minimize the overhead. In addition, Egeria caches the intermediate outputs of the frozen layers with prefetching to further skip the forward computation. Our implementation and testbed experiments with popular vision and language models show that Egeria achieves 19%-43% training speedup w.r.t. the state-of-the-art without sacrificing accuracy.

show abstract