An official website of the United States government
Clustering is a fundamental task in machine learning. One of the most successful and broadly used algorithms is DBSCAN, a density-based clustering algorithm. DBSCAN requires ϵ-nearest neighbor graphs of the input dataset, which are computed with range-search algorithms and spatial data structures like KD-trees. Despite many efforts to design scalable implementations for DBSCAN, existing work is limited to low-dimensional datasets, as constructing ϵ-nearest neighbor graphs can be expensive in high-dimensions. This article introduces a modified DBSCAN, usingk-nearest neighbor (kNN) graphs to improve efficiency. We outline conditions forkNN-DBSCAN to match DBSCAN’s results and present a parallel implementation using OpenMP and MPI for shared and distributed memory systems. Testing on datasets up to 32 dimensions, we achieve remarkable scalability. Our implementation clusters one billion 3D points in under one second on 28K cores at TACC’s Frontera system. In a larger run, we cluster 65 billion points in 20 dimensions in under 40 seconds using 114,688 cores. Our method is up to 37× faster than state-of-the-art parallel DBSCAN on a 20-dimensional dataset with 4 million points. Code is available athttps://github.com/ut-padas/knndbscan.
more »
« less
Randomized Algorithms Accelerated over CPU-GPU for Ultra-High Dimensional Similarity Search
We present FLASH (F ast L SH A lgorithm for S imilarity search accelerated with H PC), a similarity search system for ultra-high dimensional datasets on a single machine, that does not require similarity computations and is tailored for high-performance computing platforms. By leveraging a LSH style randomized indexing procedure and combining it with several principled techniques, such as reservoir sampling, recent advances in one-pass minwise hashing, and count based estimations, we reduce the computational and parallelization costs of similarity search, while retaining sound theoretical guarantees. We evaluate FLASH on several real, high-dimensional datasets from different domains, including text, malicious URL, click-through prediction, social networks, etc. Our experiments shed new light on the difficulties associated with datasets having several million dimensions. Current state-of-the-art implementations either fail on the presented scale or are orders of magnitude slower than FLASH. FLASH is capable of computing an approximate k-NN graph, from scratch, over the full webspam dataset (1.3 billion nonzeros) in less than 10 seconds. Computing a full k-NN graph in less than 10 seconds on the webspam dataset, using brute-force (n2D), will require at least 20 teraflops. We provide CPU and GPU implementations of FLASH for replicability of our results.
more »
« less
- PAR ID:
- 10065988
- Date Published:
- Journal Name:
- Proceedings of the 2018 International Conference on Management of Data
- Page Range / eLocation ID:
- 889 to 903
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Processing large amounts of data, especially in learning algorithms, poses a challenge for current embedded computing systems. Hyperdimensional (HD) computing (HDC) is a brain-inspired computing paradigm that works with high-dimensional vectors called hypervectors . HDC replaces several complex learning computations with bitwise and simpler arithmetic operations at the expense of an increased amount of data due to mapping the data into high-dimensional space. These hypervectors, more often than not, cannot be stored in memory, resulting in long data transfers from storage. In this article, we propose Store-n-Learn, an in-storage computing solution that performs HDC classification and clustering by implementing encoding, training, retraining, and inference across the flash hierarchy. To hide the latency of training and enable efficient computation, we introduce the concept of batching in HDC. We also present on-chip acceleration for HDC encoding in flash planes. This enables us to exploit the high parallelism provided by the flash hierarchy and encode multiple data points in parallel in both batched and non-batched fashion. Store-n-Learn also implements a single top-level FPGA accelerator with novel implementations for HDC classification training, retraining, inference, and clustering on the encoded data. Our evaluation over 10 popular datasets shows that Store-n-Learn is on average 222× (543×) faster than CPU and 10.6× (7.3×) faster than the state-of-the-art in-storage computing solution, INSIDER for HDC classification (clustering).more » « less
-
Neural architecture search (NAS) is a promising technique to design efficient and high-performance deep neural networks (DNNs). As the performance requirements of ML applications grow continuously, the hardware accelerators start playing a central role in DNN design. This trend makes NAS even more complicated and time-consuming for most real applications. This paper proposes FLASH, a very fast NAS methodology that co-optimizes the DNN accuracy and performance on a real hardware platform. As the main theoretical contribution, we first propose the NN-Degree, an analytical metric to quantify the topological characteristics of DNNs with skip connections (e.g., DenseNets, ResNets, Wide-ResNets, and MobileNets). The newly proposed NN-Degree allows us to do training-free NAS within one second and build an accuracy predictor by training as few as 25 samples out of a vast search space with more than 63 billion configurations. Second, by performing inference on the target hardware, we fine-tune and validate our analytical models to estimate the latency, area, and energy consumption of various DNN architectures while executing standard ML datasets. Third, we construct a hierarchical algorithm based on simplicial homology global optimization (SHGO) to optimize the model-architecture co-design process, while considering the area, latency, and energy consumption of the target hardware. We demonstrate that, compared to the state-of-the-art NAS approaches, our proposed hierarchical SHGO-based algorithm enables more than four orders of magnitude speedup (specifically, the execution time of the proposed algorithm is about 0.1 seconds). Finally, our experimental evaluations show that FLASH is easily transferable to different hardware architectures, thus enabling us to do NAS on a Raspberry Pi-3B processor in less than 3 seconds.more » « less
-
We study the top-k set similarity search problem using semantic overlap. While vanilla overlap requires exact matches between set elements, semantic overlap allows elements that are syntactically different but semantically related to increase the overlap. The semantic overlap is the maximum matching score of a bipartite graph, where an edge weight between two set elements is defined by a user-defined similarity function, e.g., cosine similarity between embeddings. Common techniques like token indexes fail for semantic search since similar elements may be unrelated at the character level. Further, verifying candidates is expensive (cubic versus linear for syntactic overlap), calling for highly selective filters. We propose Koios, the first exact and efficient algorithm for semantic overlap search. Koios leverages sophisticated filters to minimize the number of required graph-matching calculations. Our experiments show that for medium to large sets less than 5% of the candidate sets need verification, and more than half of those sets are further pruned without requiring the expensive graph matching. We show the efficiency of our algorithm on four real datasets and demonstrate the improved result quality of semantic over vanilla set similarity search.more » « less
-
The emerging Ray-tracing cores on GPUs have been repurposed for non-ray-tracing tasks by researchers recently. In this paper, we explore the benefits and effectiveness of executing graph algorithms on RT cores. We re-design breadth-first search and triangle counting on the new hardware as graph algorithm representatives. Our implementations focus on how to convert the graph operations to bounding volume hierarchy construction and ray generation, which are computational paradigms specific to ray tracing. We evaluate our RT-based methods on a wide range of real-world datasets. The results do not show the advantage of the RT-based methods over CUDA-based methods. We extend the experiments to the set intersection workload on synthesized datasets, and the RT-based method shows superior performance when the skew ratio is high. By carefully comparing the RT-based and CUDA-based binary search, we discover that RT cores are more efficient at searching for elements, but this comes with a constant and non-trivial overhead of the execution pipeline. Furthermore, the overhead of BVH construction is substantially higher than sorting on CUDA cores for large datasets. Our case studies unveil several rules of adapting graph algorithms to ray-tracing cores that might benefit future evolution of the emerging hardware towards general-computing tasks.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1145/3183713.3196925
