arXiv subjects

Minyi Guo

Publications and source records attributed to Minyi Guo.

At least 145 records · Page 8Linked to original sources

arXiv2021

Architectural Implications of Graph Neural Networks

Graph neural networks (GNN) represent an emerging line of deep learning models that operate on graph structures. It is becoming more and more popular due to its high accuracy achieved in many graph-related tasks. However, GNN is not as well understood in the system and architecture community as its counterparts such as multi-layer perceptrons and convolutional neural networks. This work tries to introduce the GNN to our community. In contrast to prior work that only presents characterizations of GCNs, our work covers a large portion of the varieties for GNN workloads based on a general GNN description framework. By constructing the models on top of two widely-used libraries, we characterize the GNN computation at inference stage concerning general-purpose and application-specific architectures and hope our work can foster more system and architecture research for GNNs.

cs.AR↗

arXiv2021

Characterizing and Demystifying the Implicit Convolution Algorithm on Commercial Matrix-Multiplication Accelerators

Many of today's deep neural network accelerators, e.g., Google's TPU and NVIDIA's tensor core, are built around accelerating the general matrix multiplication (i.e., GEMM). However, supporting convolution on GEMM-based accelerators is not trivial. The naive method explicitly lowers the convolution to GEMM, commonly known as im2col, which introduces significant performance and memory overhead. Existing implicit im2col algorithms require unscalable hardware and are inefficient in supporting important convolution variants such as strided convolution. In this paper, we propose a memory-efficient and hardware-friendly implicit im2col algorithm used by Google's TPU, which dynamically converts a convolution into a GEMM with practically zero performance and memory overhead, fully unleashing the power of GEMM engines. Through comprehensive experimental results, we quantitatively argue that this algorithm has been adopted in commercial closed-source platforms, and we are the first to describe its high-level idea and implementation details. Finally, we show that our algorithm can also be generally applied to Nvidia's Tensor Cores (TC), matching and out-performing the measured performance on TCs.

cs.DC↗

arXiv2021

Dubhe: Towards Data Unbiasedness with Homomorphic Encryption in Federated Learning Client Selection

Federated learning (FL) is a distributed machine learning paradigm that allows clients to collaboratively train a model over their own local data. FL promises the privacy of clients and its security can be strengthened by cryptographic methods such as additively homomorphic encryption (HE). However, the efficiency of FL could seriously suffer from the statistical heterogeneity in both the data distribution discrepancy among clients and the global distribution skewness. We mathematically demonstrate the cause of performance degradation in FL and examine the performance of FL over various datasets. To tackle the statistical heterogeneity problem, we propose a pluggable system-level client selection method named Dubhe, which allows clients to proactively participate in training, meanwhile preserving their privacy with the assistance of HE. Experimental results show that Dubhe is comparable with the optimal greedy method on the classification accuracy, with negligible encryption and communication overhead.

cs.CR↗

arXiv2021

Pagurus: Eliminating Cold Startup in Serverless Computing with Inter-Action Container Sharing

Serverless computing provides fine-grain resource sharing between Cloud tenants through containers. Each function invocation (action) runs in an individual container. When there is not an already started container for a user function, a new container has to be created for it. However, the long cold startup time of a container results in the long response latency of the action. Our investigation shows that the containers for some user actions share most of the software packages. If an action that requires a new container can ``borrow'' a similar warm container from other actions, the long cold startup can be eliminated. Based on the above finding, we propose Pagurus, a runtime container management system for eliminating the cold startup in serverless computing. Pagurus is comprised of an inter-action container scheduler and an intra-action container scheduler for each action. The inter-action container scheduler schedules shared containers among actions. The intra-action container scheduler deals with the management of the container lifecycle. Our experimental results show that Pagurus effectively eliminates the time-consuming container cold startup. An action may start to run in 10ms with Pagurus, even if there is not warm container for it.

cs.DC↗

arXiv2021

ZIPPER: Exploiting Tile- and Operator-level Parallelism for General and Scalable Graph Neural Network Acceleration

Graph neural networks (GNNs) start to gain momentum after showing significant performance improvement in a variety of domains including molecular science, recommendation, and transportation. Turning such performance improvement of GNNs into practical applications relies on effective and efficient execution, especially for inference. However, neither CPU nor GPU can meet these needs if considering both performance and energy efficiency. That's because accelerating GNNs is challenging due to their excessive memory usage and arbitrary interleaving of diverse operations. Besides, the semantics gap between the high-level GNN programming model and efficient hardware makes it difficult in accelerating general-domain GNNs. To address the challenge, we propose Zipper, an efficient yet general acceleration system for GNNs. The keys to Zipper include a graph-native intermediate representation (IR) and the associated compiler. By capturing GNN primitive operations and representing with GNN IR, Zipper is able to fit GNN semantics into hardware structure for efficient execution. The IR also enables GNN-specific optimizations including sparse graph tiling and redundant operation elimination. We further present an hardware architecture design consisting of dedicated blocks for different primitive operations, along with a run-time scheduler to map a IR program to the hardware blocks. Our evaluation shows that Zipper achieves 93.6x speedup and 147x energy reduction over Intel Xeon CPU, and 1.56x speedup and 4.85x energy reduction over NVIDIA V100 GPU on averages.

cs.AR↗

arXiv2020

DLFusion: An Auto-Tuning Compiler for Layer Fusion on Deep Neural Network Accelerator

Many hardware vendors have introduced specialized deep neural networks (DNN) accelerators owing to their superior performance and efficiency. As such, how to generate and optimize the code for the hardware accelerator becomes an important yet less explored problem. In this paper, we perform the compiler-stage optimization study using a novel and representative Cambricon DNN accelerator and demonstrate that the code optimization knobs play an important role in unleashing the potential of hardware computational horsepower. However, even only two studied code optimization knobs, namely the number of cores and layer fusion scheme, present an enormous search space that prevents the naive brute-force search. This work introduces a joint, auto-tuning optimization framework to address this challenge. We first use a set of synthesized DNN layers to study the interplay between the hardware performance and layer characteristics. Based on the insights, we extract the operation count and feature map channel size as each layer's characteristics and derive a joint optimization strategy to decide the performance-optimal core number and fusion scheme. We evaluate the performance of the proposed approach using a set of representative DNN models and show that it achieves the minimal of 3.6x and the maximal of 7.9x performance speedup compared to no optimization baseline. We also show that the achieved speedup is close to the oracle case that is based on a reduced brute-force search but with much less search time.

cs.DC↗

arXiv2020

How Far Does BERT Look At:Distance-based Clustering and Analysis of BERT$'$s Attention

Recent research on the multi-head attention mechanism, especially that in pre-trained models such as BERT, has shown us heuristics and clues in analyzing various aspects of the mechanism. As most of the research focus on probing tasks or hidden states, previous works have found some primitive patterns of attention head behavior by heuristic analytical methods, but a more systematic analysis specific on the attention patterns still remains primitive. In this work, we clearly cluster the attention heatmaps into significantly different patterns through unsupervised clustering on top of a set of proposed features, which corroborates with previous observations. We further study their corresponding functions through analytical study. In addition, our proposed features can be used to explain and calibrate different attention heads in Transformer models.

cs.CL↗

arXiv2020

Accelerating Sparse DNN Models without Hardware-Support via Tile-Wise Sparsity

Network pruning can reduce the high computation cost of deep neural network (DNN) models. However, to maintain their accuracies, sparse models often carry randomly-distributed weights, leading to irregular computations. Consequently, sparse models cannot achieve meaningful speedup on commodity hardware (e.g., GPU) built for dense matrix computations. As such, prior works usually modify or design completely new sparsity-optimized architectures for exploiting sparsity. We propose an algorithm-software co-designed pruning method that achieves latency speedups on existing dense architectures. Our work builds upon the insight that the matrix multiplication generally breaks the large matrix into multiple smaller tiles for parallel execution. We propose a tiling-friendly "tile-wise" sparsity pattern, which maintains a regular pattern at the tile level for efficient execution but allows for irregular, arbitrary pruning at the global scale to maintain the high accuracy. We implement and evaluate the sparsity pattern on GPU tensor core, achieving a 1.95x speedup over the dense model.

cs.DC↗

arXiv2020

Ptolemy: Architecture Support for Robust Deep Learning

Deep learning is vulnerable to adversarial attacks, where carefully-crafted input perturbations could mislead a well-trained Deep Neural Network to produce incorrect results. Today's countermeasures to adversarial attacks either do not have capability to detect adversarial samples at inference time, or introduce prohibitively high overhead to be practical at inference time. We propose Ptolemy, an algorithm-architecture co-designed system that detects adversarial attacks at inference time with low overhead and high accuracy.We exploit the synergies between DNN inference and imperative program execution: an input to a DNN uniquely activates a set of neurons that contribute significantly to the inference output, analogous to the sequence of basic blocks exercised by an input in a conventional program. Critically, we observe that adversarial samples tend to activate distinctive paths from those of benign inputs. Leveraging this insight, we propose an adversarial sample detection framework, which uses canary paths generated from offline profiling to detect adversarial samples at runtime. The Ptolemy compiler along with the co-designed hardware enable efficient execution by exploiting the unique algorithmic characteristics. Extensive evaluations show that Ptolemy achieves higher or similar adversarial example detection accuracy than today's mechanisms with a much lower runtime (as low as 2%) overhead.

cs.AR↗

arXiv2020

Balancing Efficiency and Flexibility for DNN Acceleration via Temporal GPU-Systolic Array Integration

The research interest in specialized hardware accelerators for deep neural networks (DNN) spikes recently owing to their superior performance and efficiency. However, today's DNN accelerators primarily focus on accelerating specific "kernels" such as convolution and matrix multiplication, which are vital but only part of an end-to-end DNN-enabled application. Meaningful speedups over the entire application often require supporting computations that are, while massively parallel, ill-suited to DNN accelerators. Integrating a general-purpose processor such as a CPU or a GPU incurs significant data movement overhead and leads to resource under-utilization on the DNN accelerators. We propose Simultaneous Multi-mode Architecture (SMA), a novel architecture design and execution model that offers general-purpose programmability on DNN accelerators in order to accelerate end-to-end applications. The key to SMA is the temporal integration of the systolic execution model with the GPU-like SIMD execution model. The SMA exploits the common components shared between the systolic-array accelerator and the GPU, and provides lightweight reconfiguration capability to switch between the two modes in-situ. The SMA achieves up to 63% performance improvement while consuming 23% less energy than the baseline Volta architecture with TensorCore.

cs.DC↗

arXiv2020

Towards QoS-Aware and Resource-Efficient GPU Microservices Based on Spatial Multitasking GPUs In Datacenters

While prior researches focus on CPU-based microservices, they are not applicable for GPU-based microservices due to the different contention patterns. It is challenging to optimize the resource utilization while guaranteeing the QoS for GPU microservices. We find that the overhead is caused by inter microservice communication, GPU resource contention and imbalanced throughput within microservice pipeline. We propose Camelot, a runtime system that manages GPU micorservices considering the above factors. In Camelot, a global memory-based communication mechanism enables onsite data sharing that significantly reduces the end-to-end latencies of user queries. We also propose two contention aware resource allocation policies that either maximize the peak supported service load or minimize the resource usage at low load while ensuring the required QoS. The two policies consider the microservice pipeline effect and the runtime GPU resource contention when allocating resources for the microservices. Compared with state-of-the-art work, Camelot increases the supported peak load by up to 64.5% with limited GPUs, and reduces 35% resource usage at low load while achieving the desired 99%-ile latency target.

cs.DC↗

arXiv2019

URSA: Precise Capacity Planning and Contention-aware Scheduling for Public Clouds

Database platform-as-a-service (dbPaaS) is developing rapidly and a large number of databases have been migrated to run on the Clouds for the low cost and flexibility. Emerging Clouds rely on the tenants to provide the resource specification for their database workloads. However, they tend to over-estimate the resource requirement of their databases, resulting in the unnecessarily high cost and low Cloud utilization. A methodology that automatically suggests the "just-enough" resource specification that fulfills the performance requirement of every database workload is profitable. To this end, we propose URSA, a capacity planning and workload scheduling system for dbPaaS Clouds. USRA is comprised by an online capacity planner, a performance interference estimator, and a contention-aware scheduling engine. The capacity planner identifies the most cost-efficient resource specification for a database workload to achieve the required performance online. The interference estimator quantifies the pressure on the shared resource and the sensitivity to the shared resource contention of each database workload. The scheduling engine schedules the workloads across Cloud nodes carefully to eliminate unfair performance interference between the co-located workloads. Our real system experimental results show that URSA reduces 27.5% of CPU usage and 53.4% of memory usage for database workloads while satisfying their performance requirements. Meanwhile, URSA reduces the performance unfairness between the co-located workloads by 42.8% compared with the Kubernetes.

cs.DC↗

arXiv2019

Adversarial Defense Through Network Profiling Based Path Extraction

Recently, researchers have started decomposing deep neural network models according to their semantics or functions. Recent work has shown the effectiveness of decomposed functional blocks for defending adversarial attacks, which add small input perturbation to the input image to fool the DNN models. This work proposes a profiling-based method to decompose the DNN models to different functional blocks, which lead to the effective path as a new approach to exploring DNNs' internal organization. Specifically, the per-image effective path can be aggregated to the class-level effective path, through which we observe that adversarial images activate effective path different from normal images. We propose an effective path similarity-based method to detect adversarial images with an interpretable model, which achieve better accuracy and broader applicability than the state-of-the-art technique.

cs.LG↗

arXiv2019

Position-Aware Convolutional Networks for Traffic Prediction

Forecasting the future traffic flow distribution in an area is an important issue for traffic management in an intelligent transportation system. The key challenge of traffic prediction is to capture spatial and temporal relations between future traffic flows and historical traffic due to highly dynamical patterns of human activities. Most existing methods explore such relations by fusing spatial and temporal features extracted from multi-source data. However, they neglect position information which helps distinguish patterns on different positions. In this paper, we propose a position-aware neural network that integrates data features and position information. Our approach employs the inception backbone network to capture rich features of traffic distribution on the whole area. The novelty lies in that under the backbone network, we apply position embedding technique used in neural language processing to represent position information as embedding vectors which are learned during the training. With these embedding vectors, we design position-aware convolution which allows different kernels to process features of different positions. Extensive experiments on two real-world datasets show that our approach outperforms previous methods even with fewer data sources.

cs.LG↗

arXiv2019

Knowledge Graph Convolutional Networks for Recommender Systems

To alleviate sparsity and cold start problem of collaborative filtering based recommender systems, researchers and engineers usually collect attributes of users and items, and design delicate algorithms to exploit these additional information. In general, the attributes are not isolated but connected with each other, which forms a knowledge graph (KG). In this paper, we propose Knowledge Graph Convolutional Networks (KGCN), an end-to-end framework that captures inter-item relatedness effectively by mining their associated attributes on the KG. To automatically discover both high-order structure information and semantic information of the KG, we sample from the neighbors for each entity in the KG as their receptive field, then combine neighborhood information with bias when calculating the representation of a given entity. The receptive field can be extended to multiple hops away to model high-order proximity information and capture users' potential long-distance interests. Moreover, we implement the proposed KGCN in a minibatch fashion, which enables our model to operate on large datasets and KGs. We apply the proposed model to three datasets about movie, book, and music recommendation, and experiment results demonstrate that our approach outperforms strong recommender baselines.

cs.IR↗

arXiv2019

Multi-Task Feature Learning for Knowledge Graph Enhanced Recommendation

Collaborative filtering often suffers from sparsity and cold start problems in real recommendation scenarios, therefore, researchers and engineers usually use side information to address the issues and improve the performance of recommender systems. In this paper, we consider knowledge graphs as the source of side information. We propose MKR, a Multi-task feature learning approach for Knowledge graph enhanced Recommendation. MKR is a deep end-to-end framework that utilizes knowledge graph embedding task to assist recommendation task. The two tasks are associated by cross&compress units, which automatically share latent features and learn high-order interactions between items in recommender systems and entities in the knowledge graph. We prove that cross&compress units have sufficient capability of polynomial approximation, and show that MKR is a generalized framework over several representative methods of recommender systems and multi-task learning. Through extensive experiments on real-world datasets, we demonstrate that MKR achieves substantial gains in movie, book, music, and news recommendation, over state-of-the-art baselines. MKR is also shown to be able to maintain a decent performance even if user-item interactions are sparse.

cs.IR↗

arXiv2018

RippleNet: Propagating User Preferences on the Knowledge Graph for Recommender Systems

To address the sparsity and cold start problem of collaborative filtering, researchers usually make use of side information, such as social networks or item attributes, to improve recommendation performance. This paper considers the knowledge graph as the source of side information. To address the limitations of existing embedding-based and path-based methods for knowledge-graph-aware recommendation, we propose Ripple Network, an end-to-end framework that naturally incorporates the knowledge graph into recommender systems. Similar to actual ripples propagating on the surface of water, Ripple Network stimulates the propagation of user preferences over the set of knowledge entities by automatically and iteratively extending a user's potential interests along links in the knowledge graph. The multiple "ripples" activated by a user's historically clicked items are thus superposed to form the preference distribution of the user with respect to a candidate item, which could be used for predicting the final clicking probability. Through extensive experiments on real-world datasets, we demonstrate that Ripple Network achieves substantial gains in a variety of scenarios, including movie, book and news recommendation, over several state-of-the-art baselines.

cs.IR↗

arXiv2018

Personalized Exposure Control Using Adaptive Metering and Reinforcement Learning

We propose a reinforcement learning approach for real-time exposure control of a mobile camera that is personalizable. Our approach is based on Markov Decision Process (MDP). In the camera viewfinder or live preview mode, given the current frame, our system predicts the change in exposure so as to optimize the trade-off among image quality, fast convergence, and minimal temporal oscillation. We model the exposure prediction function as a fully convolutional neural network that can be trained through Gaussian policy gradient in an end-to-end fashion. As a result, our system can associate scene semantics with exposure values; it can also be extended to personalize the exposure adjustments for a user and device. We improve the learning performance by incorporating an adaptive metering module that links semantics with exposure. This adaptive metering module generalizes the conventional spot or matrix metering techniques. We validate our system using the MIT FiveK and our own datasets captured using iPhone 7 and Google Pixel. Experimental results show that our system exhibits stable real-time behavior while improving visual quality compared to what is achieved through native camera control.

cs.CV↗