There has been a recent surge in the field of Explainable AI (XAI) which
tackles the problem of providing insights into the behavior of black-box
machine learning models. Within this field, \textit{feature attribution}
encompasses methods which assign relevance scores to input features and
visualize them as a heatmap. Designing flexible accelerators for multiple such
algorithms is challenging since the hardware mapping of these algorithms has
not been studied yet. In this work, we first analyze the dataflow of gradient
backpropagation based feature attribution algorithms to determine the resource
overhead required over inference. The gradient computation is optimized to
minimize the memory overhead. Second, we develop a High-Level Synthesis (HLS)
based configurable FPGA design that is targeted for edge devices and supports
three feature attribution algorithms. Tile based computation is employed to
maximally use on-chip resources while adhering to the resource constraints.
Representative CNNs are trained on CIFAR-10 dataset and implemented on multiple
Xilinx FPGAs using 16-bit fixed-point precision demonstrating flexibility of
our library. Finally, through efficient reuse of allocated hardware resources,
our design methodology demonstrates a pathway to repurpose inference
accelerators to support feature attribution with minimal overhead, thereby
enabling real-time XAI on the edge.

本研究为了实现边缘设备上的实时可解释人工智能（XAI），利用可配置的 FPGA 设计和优化算法提高多种特征归因算法的灵活性和性能，并演示了通过分配硬件资源的有效重复使用的路径，以最小化部分准确性损失。

基于梯度反向传播的特征归因，实现边缘智能可解释性

Gradient Backpropagation based Feature Attribution to Enable Explainable-AI on the Edge

This paper presents the FPGA design of a convolutional neural network (CNN)
based road segmentation algorithm for real-time processing of LiDAR data. For
autonomous vehicles, it is important to perform road segmentation and obstacle
detection such that the drivable region can be identified for path planning.
Traditional road segmentation algorithms are mainly based on image data from
cameras, which is subjected to the light condition as well as the quality of
road markings. LiDAR sensor can obtain the 3D geometry information of the
vehicle surroundings with very high accuracy. However, it is a computational
challenge to process a large amount of LiDAR data at real-time. In this work, a
convolutional neural network model is proposed and trained to perform semantic
segmentation using the LiDAR sensor data. Furthermore, an efficient hardware
design is implemented on the FPGA that can process each LiDAR scan in 16.9ms,
which is much faster than the previous works. Evaluated using KITTI road
benchmarks, the proposed solution achieves high accuracy of road segmentation.

本研究提出了一种基于 FPGA 的卷积神经网络 (CNN) 设计，用于实时处理激光雷达数据的道路分割算法。通过在 LiDAR 传感器数据上训练神经网络模型，本文实现了高效的硬件设计，并成功实现每个 LiDAR 扫描的实时处理，实现了道路分割的高精度。