English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

The deep neural network (DNN) based AI applications on the edge require bothlow-cost computing platforms and high-quality services. However, the limitedmemory, computing resources, and power budget of the edge devices constrain theeffectiveness of the DNN algorithms. Developing edge-oriented AI algorithms andimplementations (e.g., accelerators) is challenging. In this paper, wesummarize our recent efforts for efficient on-device AI development from threeaspects, including both training and inference. First, we present on-devicetraining with ultra-low memory usage. We propose a novel rank-adaptivetensor-based tensorized neural network model, which offers orders-of-magnitudememory reduction during training. Second, we introduce an ultra-low bitwidthquantization method for DNN model compression, achieving the state-of-the-artaccuracy under the same compression ratio. Third, we introduce an ultra-lowlatency DNN accelerator design, practicing the software/hardware co-designmethodology. This paper emphasizes the importance and efficacy of training,quantization and accelerator design, and calls for more research breakthroughsin the area for AI on the edge.

3U-EdgeAI: Ultra-Low Memory Training, Ultra-Low BitwidthQuantization, and Ultra-Low Latency Acceleration