Related papers: Dr. Top-k: Delegate-Centric Top-k on GPUs

Expert Parallelism (EP) permits Mixture of Experts (MoE) models to scale beyond a single GPU. To address load imbalance across GPUs in EP, existing approaches aim to balance the number of tokens each GPU processes. Surprisingly, we find…

Edge computing offers the distinct advantage of harnessing compute capabilities on resources located at the edge of the network to run workloads of relatively weak user devices. This is achieved by offloading computationally intensive…

The problem of finding the $k$ most critical nodes, referred to as the $top\text{-}k$ problem, is a very important one in several contexts such as information diffusion and preference aggregation in social networks, clustering of data…

Large Language Models (LLMs) are increasingly prevalent in the field of long-context modeling, however, their inference computational costs have become a critical bottleneck hindering the advancement of tasks such as agents and multimodal…

k-nearest neighbor graph is a key data structure in many disciplines such as manifold learning, machine learning and information retrieval, etc. NN-Descent was proposed as an effective solution for the graph construction problem. However,…

We explore how the big-three computing paradigms -- symmetric multi-processor (SMC), graphical processing units (GPUs), and cluster computing -- can together be brought to bare on large-data Gaussian processes (GP) regression problems via a…

Computational Pangenomics is an emerging field that studies genetic variation using a graph structure encompassing multiple genomes. Visualizing pangenome graphs is vital for understanding genome diversity. Yet, handling large graphs can be…

Clustering is an important tool in data analysis, with K-means being popular for its simplicity and versatility. However, it cannot handle non-linearly separable clusters. Kernel K-means addresses this limitation but requires a large kernel…

Recent years have witnessed increasing interest in machine learning inferences on serverless computing for its auto-scaling and cost effective properties. Existing serverless computing, however, lacks effective job scheduling methods to…

To improve the quality of computation experience for mobile devices, mobile-edge computing (MEC) is emerging as a promising paradigm by providing computing capabilities within radio access networks in close proximity. Nevertheless, the…

We consider the recently introduced monochromatic reverse top-k queries which ask for, given a new tuple q and a dataset D, all possible top-k queries on D union {q} for which q is in the result. Towards this problem, we focus on designing…

A central problem in graph mining is finding dense subgraphs, with several applications in different fields, a notable example being identifying communities. While a lot of effort has been put on the problem of finding a single dense…

Distributed attention is a fundamental problem for scaling context window for Large Language Models (LLMs). The state-of-the-art method, Ring-Attention, suffers from scalability limitations due to its excessive communication traffic. This…

Tokenization sits at the boundary between high-throughput genomic input and GPU compute, posing challenges in both algorithm design and system throughput. Overlapping k-mer tokenization can introduce information leakage under masked…

In order to improve system performance efficiently, a number of systems choose to equip multi-core and many-core processors (such as GPUs). Due to their discrete memory these heterogeneous architectures comprise a distributed system within…

Flow delegation is a flexible technique to mitigate flow table capacity bottlenecks in Software-defined Networks (SDN). Such bottlenecks occur when SDN switches provide insufficient flow table capacity which leads to performance degradation…

Deep Learning (DL) models have achieved superior performance. Meanwhile, computing hardware like NVIDIA GPUs also demonstrated strong computing scaling trends with 2x throughput and memory bandwidth for each generation. With such strong…

Hypertree decompositions of hypergraphs are a generalization of tree decompositions of graphs. The corresponding hypertree-width is a measure for the cyclicity and therefore tractability of the encoded computation problem. Many NP-hard…

Due to densification of wireless networks, there exist abundance of idling computation resources at edge devices. These resources can be scavenged by offloading heavy computation tasks from small IoT devices in proximity, thereby overcoming…

Going beyond networks, to include higher-order interactions of arbitrary sizes, is a major step to better describe complex systems. In the resulting hypergraph representation, tools to identify structures and central nodes are scarce. We…