Arkade: k-Nearest Neighbor Search With Non-Euclidean Distances using GPU Ray Tracing
Durga Mandarapu, Vani Nagarajan, Artem Pelenitsyn, Milind Kulkarni
2024 International Conference on Supercomputing (ICS '24) [Best paper Award]
Arkade: k-Nearest Neighbor Search With Non-Euclidean Distances using GPU Ray Tracing
Durga Mandarapu, Vani Nagarajan, Artem Pelenitsyn, Milind Kulkarni
2024 International Conference on Supercomputing (ICS '24) [Best paper Award]
Prior work can only accelerate neighbor searches using Euclidean distance
Arkade Filter-Refine and Arkade Monotone Transformation reduce non-Euclidean distance-based nearest neighbor queries to Euclidean distance
Showed that our approach upto 200x faster than GPU-accelerated baselines
Read the paper here: Arkade
Vani Nagarajan, Durga Mandarapu, Milind Kulkarni
2023 International Conference on Supercomputing (ICS '23)
Prior work can only translate fixed-radius Nearest Neighbor Searches (NNS) to ray tracing queries
Accelerated unbounded k-Nearest Neighbor Search (kNNS) applications by removing fixed-radius constraint
Showed that our approach is orders of magnitude faster than other RT-accelerated neighbor search baseline
Read the paper here: TrueKNN
Vani Nagarajan, Milind Kulkarni
2023 International Parallel and Distributed Processing Symposium (IPDPS '23)
Introduced RT-DBSCAN, the first RT-accelerated clustering algorithm
Created a primitive, RT-findNeighbor, that automatically translates DBSCAN's nearest neighbor queries to ray tracing queries that can be launched on the Ray Tracing hardware
Showed that offloading DBSCAN's distance computations to RT cores while allowing shader cores to perform other CUDA operations resulted in significant performance improvements
Achieved 1.3x to 4.5x speedup over current state-of-the-art GPU implementations
Read the paper here: RT-DBSCAN