No abstract

Automated Design Space Exploration (DSE) is a critical part of system-level design. It relies on performance estimation to evaluate design alternatives. However, since a plethora of design alternatives need to be compared, the run-time of performance estimation itself may pose a bottleneck. In DSE, fastest performance estimation is of essence while some accuracy may be sacrificed. Fast estimation can be realised through capturing application demand, as well as Processing Element (PE) supply (later on called weight table) in a matrix each. Then, performance estimation (retargeting) is reduced to a matrix multiplication. However, defining the weight table from a data sheet is impracticle due to the multitude of (micro-) architecture aspects. This paper introduces a novel methodology, WeiCal, for automatically generating Weight Tables in the context of C source-level estimation using application profiling and Linear Programming (LP). LP solving is based on the measured performance of training benchmarks on an actual PE. We validated WeiCal using a synthetic processor and benchmark model, and also analyse the impact of non-observable features on estimation accuracy. We evaluate the efficiency using 49 benchmarks on 2 different processors with varying configurations (multiple memory configurations and software optimizations). On a 3.1 GHz i5-3450 Intel host, 25 million estimations / second can be obtained regardless of the application size and PE complexity. The accuracy is sufficient for early DSE with a 24% average error.

This study presents a comprehensive analysis and improvement of the YOLOv8-n algorithm for object detection, focusing on the integration of Wasserstein Distance Loss, FasterNext, and Context Aggravation strategies. Through a detailed ablation study, each strategy was systematically evaluated individually and collectively to assess its contribution to the model's performance. The results indicate that each strategy uniquely enhances the model's performance, significantly increasing mAP and reducing model complexity when all three are integrated. Visualizations through Grad-CAM further substantiate the improved model's capacity to extract and focus on key object features. Comparisons with existing models, such as YOLOv5-n, YOLOv5-s, YOLOX-n, YOLOX-s, and YOLOv7-tiny, the improved YOLOv8-n model achieves an optimal balance between accuracy and model complexity, outperforming other models in terms of model accuracy, model complexity, and model inference speed. Further image inference tests validate the model's performance, showcasing its superior detection capabilities.

No abstract