arXiv subjects

Zhen Jia

Publications and source records attributed to Zhen Jia.

At least 37 records · Page 2Linked to original sources

arXiv2024

Faithful Temporal Question Answering over Heterogeneous Sources

Temporal question answering (QA) involves time constraints, with phrases such as "... in 2019" or "... before COVID". In the former, time is an explicit condition, in the latter it is implicit. State-of-the-art methods have limitations along three dimensions. First, with neural inference, time constraints are merely soft-matched, giving room to invalid or inexplicable answers. Second, questions with implicit time are poorly supported. Third, answers come from a single source: either a knowledge base (KB) or a text corpus. We propose a temporal QA system that addresses these shortcomings. First, it enforces temporal constraints for faithful answering with tangible evidence. Second, it properly handles implicit questions. Third, it operates over heterogeneous sources, covering KB, text and web tables in a unified manner. The method has three stages: (i) understanding the question and its temporal conditions, (ii) retrieving evidence from all sources, and (iii) faithfully answering the question. As implicit questions are sparse in prior benchmarks, we introduce a principled method for generating diverse questions. Experiments show superior performance over a suite of baselines.

cs.IR↗

arXiv2023

DynaPipe: Optimizing Multi-task Training through Dynamic Pipelines

Multi-task model training has been adopted to enable a single deep neural network model (often a large language model) to handle multiple tasks (e.g., question answering and text summarization). Multi-task training commonly receives input sequences of highly different lengths due to the diverse contexts of different tasks. Padding (to the same sequence length) or packing (short examples into long sequences of the same length) is usually adopted to prepare input samples for model training, which is nonetheless not space or computation efficient. This paper proposes a dynamic micro-batching approach to tackle sequence length variation and enable efficient multi-task model training. We advocate pipeline-parallel training of the large model with variable-length micro-batches, each of which potentially comprises a different number of samples. We optimize micro-batch construction using a dynamic programming-based approach, and handle micro-batch execution time variation through dynamic pipeline and communication scheduling, enabling highly efficient pipeline training. Extensive evaluation on the FLANv2 dataset demonstrates up to 4.39x higher training throughput when training T5, and 3.25x when training GPT, as compared with packing-based baselines. DynaPipe's source code is publicly available at https://github.com/awslabs/optimizing-multitask-training-through-dynamic-pipelines.

cs.DC↗

arXiv2023

Multi-Semantic Fusion Model for Generalized Zero-Shot Skeleton-Based Action Recognition

Generalized zero-shot skeleton-based action recognition (GZSSAR) is a new challenging problem in computer vision community, which requires models to recognize actions without any training samples. Previous studies only utilize the action labels of verb phrases as the semantic prototypes for learning the mapping from skeleton-based actions to a shared semantic space. However, the limited semantic information of action labels restricts the generalization ability of skeleton features for recognizing unseen actions. In order to solve this dilemma, we propose a multi-semantic fusion (MSF) model for improving the performance of GZSSAR, where two kinds of class-level textual descriptions (i.e., action descriptions and motion descriptions), are collected as auxiliary semantic information to enhance the learning efficacy of generalizable skeleton features. Specially, a pre-trained language encoder takes the action descriptions, motion descriptions and original class labels as inputs to obtain rich semantic features for each action class, while a skeleton encoder is implemented to extract skeleton features. Then, a variational autoencoder (VAE) based generative module is performed to learn a cross-modal alignment between skeleton and semantic features. Finally, a classification module is built to recognize the action categories of input samples, where a seen-unseen classification gate is adopted to predict whether the sample comes from seen action classes or not in GZSSAR. The superior performance in comparisons with previous models validates the effectiveness of the proposed MSF model on GZSSAR.

cs.CV↗

arXiv2023

Arbitrary point cloud upsampling via Dual Back-Projection Network

Point clouds acquired from 3D sensors are usually sparse and noisy. Point cloud upsampling is an approach to increase the density of the point cloud so that detailed geometric information can be restored. In this paper, we propose a Dual Back-Projection network for point cloud upsampling (DBPnet). A Dual Back-Projection is formulated in an up-down-up manner for point cloud upsampling. It not only back projects feature residues but also coordinates residues so that the network better captures the point correlations in the feature and space domains, achieving lower reconstruction errors on both uniform and non-uniform sparse point clouds. Our proposed method is also generalizable for arbitrary upsampling tasks (e.g. 4x, 5.5x). Experimental results show that the proposed method achieves the lowest point set matching losses with respect to the benchmark. In addition, the success of our approach demonstrates that generative networks are not necessarily needed for non-uniform point clouds.

cs.CV↗

arXiv2023

Soft-IntroVAE for Continuous Latent space Image Super-Resolution

Continuous image super-resolution (SR) recently receives a lot of attention from researchers, for its practical and flexible image scaling for various displays. Local implicit image representation is one of the methods that can map the coordinates and 2D features for latent space interpolation. Inspired by Variational AutoEncoder, we propose a Soft-introVAE for continuous latent space image super-resolution (SVAE-SR). A novel latent space adversarial training is achieved for photo-realistic image restoration. To further improve the quality, a positional encoding scheme is used to extend the original pixel coordinates by aggregating frequency information over the pixel areas. We show the effectiveness of the proposed SVAE-SR through quantitative and qualitative comparisons, and further, illustrate its generalization in denoising and real-image super-resolution.

eess.IV↗

arXiv2023

RAF: Holistic Compilation for Deep Learning Model Training

As deep learning is pervasive in modern applications, many deep learning frameworks are presented for deep learning practitioners to develop and train DNN models rapidly. Meanwhile, as training large deep learning models becomes a trend in recent years, the training throughput and memory footprint are getting crucial. Accordingly, optimizing training workloads with compiler optimizations is inevitable and getting more and more attentions. However, existing deep learning compilers (DLCs) mainly target inference and do not incorporate holistic optimizations, such as automatic differentiation and automatic mixed precision, in training workloads. In this paper, we present RAF, a deep learning compiler for training. Unlike existing DLCs, RAF accepts a forward model and in-house generates a training graph. Accordingly, RAF is able to systematically consolidate graph optimizations for performance, memory and distributed training. In addition, to catch up to the state-of-the-art performance with hand-crafted kernel libraries as well as tensor compilers, RAF proposes an operator dialect mechanism to seamlessly integrate all possible kernel implementations. We demonstrate that by in-house training graph generation and operator dialect mechanism, we are able to perform holistic optimizations and achieve either better training throughput or larger batch size against PyTorch (eager and torchscript mode), XLA, and DeepSpeed for popular transformer models on GPUs.

cs.LG↗

arXiv2022

Learning Domain Invariant Representations for Generalizable Person Re-Identification

Generalizable person Re-Identification (ReID) has attracted growing attention in recent computer vision community. In this work, we construct a structural causal model among identity labels, identity-specific factors (clothes/shoes color etc), and domain-specific factors (background, viewpoints etc). According to the causal analysis, we propose a novel Domain Invariant Representation Learning for generalizable person Re-Identification (DIR-ReID) framework. Specifically, we first propose to disentangle the identity-specific and domain-specific feature spaces, based on which we propose an effective algorithmic implementation for backdoor adjustment, essentially serving as a causal intervention towards the SCM. Extensive experiments have been conducted, showing that DIR-ReID outperforms state-of-the-art methods on large-scale domain generalization ReID benchmarks.

cs.CV↗

arXiv2022

Focal and Efficient IOU Loss for Accurate Bounding Box Regression

In object detection, bounding box regression (BBR) is a crucial step that determines the object localization performance. However, we find that most previous loss functions for BBR have two main drawbacks: (i) Both $\ell_n$-norm and IOU-based loss functions are inefficient to depict the objective of BBR, which leads to slow convergence and inaccurate regression results. (ii) Most of the loss functions ignore the imbalance problem in BBR that the large number of anchor boxes which have small overlaps with the target boxes contribute most to the optimization of BBR. To mitigate the adverse effects caused thereby, we perform thorough studies to exploit the potential of BBR losses in this paper. Firstly, an Efficient Intersection over Union (EIOU) loss is proposed, which explicitly measures the discrepancies of three geometric factors in BBR, i.e., the overlap area, the central point and the side length. After that, we state the Effective Example Mining (EEM) problem and propose a regression version of focal loss to make the regression process focus on high-quality anchor boxes. Finally, the above two parts are combined to obtain a new loss function, namely Focal-EIOU loss. Extensive experiments on both synthetic and real datasets are performed. Notable superiorities on both the convergence speed and the localization accuracy can be achieved over other BBR losses.

cs.CV↗

arXiv2022

DRAformer: Differentially Reconstructed Attention Transformer for Time-Series Forecasting

Time-series forecasting plays an important role in many real-world scenarios, such as equipment life cycle forecasting, weather forecasting, and traffic flow forecasting. It can be observed from recent research that a variety of transformer-based models have shown remarkable results in time-series forecasting. However, there are still some issues that limit the ability of transformer-based models on time-series forecasting tasks: (i) learning directly on raw data is susceptible to noise due to its complex and unstable feature representation; (ii) the self-attention mechanisms pay insufficient attention to changing features and temporal dependencies. In order to solve these two problems, we propose a transformer-based differentially reconstructed attention model DRAformer. Specifically, DRAformer has the following innovations: (i) learning against differenced sequences, which preserves clear and stable sequence features by differencing and highlights the changing properties of sequences; (ii) the reconstructed attention: integrated distance attention exhibits sequential distance through a learnable Gaussian kernel, distributed difference attention calculates distribution difference by mapping the difference sequence to the adaptive feature space, and the combination of the two effectively focuses on the sequences with prominent associations; (iii) the reconstructed decoder input, which extracts sequence features by integrating variation information and temporal correlations, thereby obtaining a more comprehensive sequence representation. Extensive experiments on four large-scale datasets demonstrate that DRAformer outperforms state-of-the-art baselines.

cs.LG↗

arXiv2022

Manifoldron: Direct Space Partition via Manifold Discovery

A neural network with the widely-used ReLU activation has been shown to partition the sample space into many convex polytopes for prediction. However, the parameterized way a neural network and other machine learning models use to partition the space has imperfections, \textit{e}.\textit{g}., the compromised interpretability for complex models, the inflexibility in decision boundary construction due to the generic character of the model, and the risk of being trapped into shortcut solutions. In contrast, although the non-parameterized models can adorably avoid or downplay these issues, they are usually insufficiently powerful either due to over-simplification or the failure to accommodate the manifold structures of data. In this context, we first propose a new type of machine learning models referred to as Manifoldron that directly derives decision boundaries from data and partitions the space via manifold structure discovery. Then, we systematically analyze the key characteristics of the Manifoldron such as manifold characterization capability and its link to neural networks. The experimental results on 4 synthetic examples, 20 public benchmark datasets, and 1 real-world application demonstrate that the proposed Manifoldron performs competitively compared to the mainstream machine learning models. We have shared our code in \url{https://github.com/wdayang/Manifoldron} for free download and evaluation.

cs.LG↗

arXiv2021

Complex Temporal Question Answering on Knowledge Graphs

Question answering over knowledge graphs (KG-QA) is a vital topic in IR. Questions with temporal intent are a special class of practical importance, but have not received much attention in research. This work presents EXAQT, the first end-to-end system for answering complex temporal questions that have multiple entities and predicates, and associated temporal conditions. EXAQT answers natural language questions over KGs in two stages, one geared towards high recall, the other towards precision at top ranks. The first step computes question-relevant compact subgraphs within the KG, and judiciously enhances them with pertinent temporal facts, using Group Steiner Trees and fine-tuned BERT models. The second step constructs relational graph convolutional networks (R-GCNs) from the first step's output, and enhances the R-GCNs with time-aware entity embeddings and attention over temporal relations. We evaluate EXAQT on TimeQuestions, a large dataset of 16k temporal questions we compiled from a variety of general purpose KG-QA benchmarks. Results show that EXAQT outperforms three state-of-the-art systems for answering complex questions over KGs, thereby justifying specialized treatment of temporal QA.

cs.IR↗

arXiv2021

TEQUILA: Temporal Question Answering over Knowledge Bases

Question answering over knowledge bases (KB-QA) poses challenges in handling complex questions that need to be decomposed into sub-questions. An important case, addressed here, is that of temporal questions, where cues for temporal relations need to be discovered and handled. We present TEQUILA, an enabler method for temporal QA that can run on top of any KB-QA engine. TEQUILA has four stages. It detects if a question has temporal intent. It decomposes and rewrites the question into non-temporal sub-questions and temporal constraints. Answers to sub-questions are then retrieved from the underlying KB-QA engine. Finally, TEQUILA uses constraint reasoning on temporal intervals to compute final answers to the full question. Comparisons against state-of-the-art baselines show the viability of our method.

cs.IR↗

arXiv2018

Data Motif-based Proxy Benchmarks for Big Data and AI Workloads

For the architecture community, reasonable simulation time is a strong requirement in addition to performance data accuracy. However, emerging big data and AI workloads are too huge at binary size level and prohibitively expensive to run on cycle-accurate simulators. The concept of data motif, which is identified as a class of units of computation performed on initial or intermediate data, is the first step towards building proxy benchmark to mimic the real-world big data and AI workloads. However, there is no practical way to construct a proxy benchmark based on the data motifs to help simulation-based research. In this paper, we embark on a study to bridge the gap between data motif and a practical proxy benchmark. We propose a data motif-based proxy benchmark generating methodology by means of machine learning method, which combine data motifs with different weights to mimic the big data and AI workloads. Furthermore, we implement various data motifs using light-weight stacks and apply the methodology to five real-world workloads to construct a suite of proxy benchmarks, considering the data types, patterns, and distributions. The evaluation results show that our proxy benchmarks shorten the execution time by 100s times on real systems while maintaining the average system and micro-architecture performance data accuracy above 90%, even changing the input data sets or cluster configurations. Moreover, the generated proxy benchmarks reflect consistent performance trends across different architectures. To facilitate the community, we will release the proxy benchmarks on the project homepage http://prof.ict.ac.cn/BigDataBench.

cs.DC↗

arXiv2018

FFT Convolutions are Faster than Winograd on Modern CPUs, Here is Why

Winograd-based convolution has quickly gained traction as a preferred approach to implement convolutional neural networks (ConvNet) on various hardware platforms because it requires fewer floating point operations than FFT-based or direct convolutions. This paper compares three highly optimized implementations (regular FFT--, Gauss--FFT--, and Winograd--based convolutions) on modern multi-- and many--core CPUs. Although all three implementations employed the same optimizations for modern CPUs, our experimental results with two popular ConvNets (VGG and AlexNet) show that the FFT--based implementations generally outperform the Winograd--based approach, contrary to the popular belief. To understand the results, we use a Roofline performance model to analyze the three implementations in detail, by looking at each of their computation phases and by considering not only the number of floating point operations, but also the memory bandwidth and the cache sizes. The performance analysis explains why, and under what conditions, the FFT--based implementations outperform the Winograd--based one, on modern CPUs.

cs.PF↗

arXiv2017

A Dwarf-based Scalable Big Data Benchmarking Methodology

Different from the traditional benchmarking methodology that creates a new benchmark or proxy for every possible workload, this paper presents a scalable big data benchmarking methodology. Among a wide variety of big data analytics workloads, we identify eight big data dwarfs, each of which captures the common requirements of each class of unit of computation while being reasonably divorced from individual implementations. We implement the eight dwarfs on different software stacks, e.g., OpenMP, MPI, Hadoop as the dwarf components. For the purpose of architecture simulation, we construct and tune big data proxy benchmarks using the directed acyclic graph (DAG)-like combinations of the dwarf components with different weights to mimic the benchmarks in BigDataBench. Our proxy benchmarks preserve the micro-architecture, memory, and I/O characteristics, and they shorten the simulation time by 100s times while maintain the average micro-architectural data accuracy above 90 percentage on both X86 64 and ARMv8 processors. We will open-source the big data dwarf components and proxy benchmarks soon.

cs.AR↗

arXiv2015

Characterization and Architectural Implications of Big Data Workloads

Big data areas are expanding in a fast way in terms of increasing workloads and runtime systems, and this situation imposes a serious challenge to workload characterization, which is the foundation of innovative system and architecture design. The previous major efforts on big data benchmarking either propose a comprehensive but a large amount of workloads, or only select a few workloads according to so-called popularity, which may lead to partial or even biased observations. In this paper, on the basis of a comprehensive big data benchmark suite---BigDataBench, we reduced 77 workloads to 17 representative workloads from a micro-architectural perspective. On a typical state-of-practice platform---Intel Xeon E5645, we compare the representative big data workloads with SPECINT, SPECCFP, PARSEC, CloudSuite and HPCC. After a comprehensive workload characterization, we have the following observations. First, the big data workloads are data movement dominated computing with more branch operations, taking up to 92% percentage in terms of instruction mix, which places them in a different class from Desktop (SPEC CPU2006), CMP (PARSEC), HPC (HPCC) workloads. Second, corroborating the previous work, Hadoop and Spark based big data workloads have higher front-end stalls. Comparing with the traditional workloads i. e. PARSEC, the big data workloads have larger instructions footprint. But we also note that, in addition to varied instruction-level parallelism, there are significant disparities of front-end efficiencies among different big data workloads. Third, we found complex software stacks that fail to use state-of-practise processors efficiently are one of the main factors leading to high front-end stalls. For the same workloads, the L1I cache miss rates have one order of magnitude differences among diverse implementations with different software stacks.

cs.DC↗

arXiv2015

Benchmarking Big Data Systems: State-of-the-Art and Future Directions

The great prosperity of big data systems such as Hadoop in recent years makes the benchmarking of these systems become crucial for both research and industry communities. The complexity, diversity, and rapid evolution of big data systems gives rise to various new challenges about how we design generators to produce data with the 4V properties (i.e. volume, velocity, variety and veracity), as well as implement application-specific but still comprehensive workloads. However, most of the existing big data benchmarks can be described as attempts to solve specific problems in benchmarking systems. This article investigates the state-of-the-art in benchmarking big data systems along with the future challenges to be addressed to realize a successful and efficient benchmark.

cs.PF↗

arXiv2015

Understanding Big Data Analytic Workloads on Modern Processors

Big data analytics applications play a significant role in data centers, and hence it has become increasingly important to understand their behaviors in order to further improve the performance of data center computer systems, in which characterizing representative workloads is a key practical problem. In this paper, after investigating three most impor- tant application domains in terms of page views and daily visitors, we chose 11 repre- sentative data analytics workloads and characterized their micro-architectural behaviors by using hardware performance counters, so as to understand the impacts and implications of data analytics workloads on the systems equipped with modern superscalar out-of-order processors. Our study reveals that big data analytics applications themselves share many inherent characteristics, which place them in a different class from traditional workloads and scale-out services. To further understand the characteristics of big data analytics work- loads we performed a correlation analysis of CPI (cycles per instruction) with other micro- architecture level characteristics and an investigation of the big data software stack impacts on application behaviors. Our correlation analysis showed that even though big data ana- lytics workloads own notable pipeline front end stalls, the main factors affecting the CPI performance are long latency data accesses rather than the front end stalls. Our software stack investigation found that the typical big data software stack significantly contributes to the front end stalls and incurs bigger working set. Finally we gave several recommen- dations for architects, programmers and big data system designers with the knowledge acquired from this paper.

cs.DC↗