We propose Delayed Gradient Averaging (DGA), which delays the averaging step to improve efficiency and allows local computation in parallel to communication.

PointAcc is a novel point cloud deep learning accelerator. It introduces a configurable sorting-based mapping unit that efficiently supports diverse operations in point cloud networks. PointAcc further exploits simplified caching and layer fusion specialized for point cloud models, effectively reducing the DRAM access.

Multi-sensor fusion is important in real-world robotics systems, but aligning different sensors through calibration is challenging and requires hours of human efforts. To this end, we propose SemAlign that does not require ground-truth calibration annotations and automates the process of camera-3D calibration.

Pruning and Quantization for Transformer models such as BERT and GPT

SPVNAS enhances Point-Voxel Convolution in large-scale outdoor scenes with sparse convolutions. With 3D Neural Architecture Search (3D-NAS), it efficiently and effectively searches the optimal 3D neural network architecture under a given resource constraint.

A general-purpose compression framework for reducing the inference time and model size of the generator in conditional GANs.