Search

To match the blooming demand of generative AI workloads, GPU designers have so far been trying to pack more and more compute and memory into single complex and expensive packages. However, there is growing uncertainty about the scalability of individual GPUs and thus AI clusters, as state-of-the-art GPUs are already displaying packaging, yield, and cooling limitations. We propose to rethink the design and scaling of AI clusters through efficiently-connected large clusters of Lite-GPUs, GPUs with single, small dies and a fraction of the capabilities of larger GPUs. We think recent advances in co-packaged optics can be key in overcoming the communication challenges of distributing AI workloads onto more Lite-GPUs. In this paper, we present the key benefits of Lite-GPUs on manufacturing cost, blast radius, yield, and power efficiency; and discuss systems opportunities and challenges around resource, workload, memory, and network management., Comment: 5+ pages, 4 figures

AI clusters today are one of the major uses of High Bandwidth Memory (HBM). However, HBM is suboptimal for AI workloads for several reasons. Analysis shows HBM is overprovisioned on write performance, but underprovisioned on density and read bandwidth, and also has significant energy per bit overheads. It is also expensive, with lower yield than DRAM due to manufacturing complexity. We propose a new memory class: Managed-Retention Memory (MRM), which is more optimized to store key data structures for AI inference workloads. We believe that MRM may finally provide a path to viability for technologies that were originally proposed to support Storage Class Memory (SCM). These technologies traditionally offered long-term persistence (10+ years) but provided poor IO performance and/or endurance. MRM makes different trade-offs, and by understanding the workload IO patterns, MRM foregoes long-term data retention and write performance for better potential performance on the metrics important for these workloads., Comment: 8 pages (5 content + 3 refs); 1 figure

3D terrain reconstruction with remote sensing imagery achieves cost-effective and large-scale earth observation and is crucial for safeguarding natural disasters, monitoring ecological changes, and preserving the environment.Recently, learning-based multi-view stereo~(MVS) methods have shown promise in this task. However, these methods simply modify the general learning-based MVS framework for height estimation, which overlooks the terrain characteristics and results in insufficient accuracy. Considering that the Earth's surface generally undulates with no drastic changes and can be measured by slope, integrating slope considerations into MVS frameworks could enhance the accuracy of terrain reconstructions. To this end, we propose an end-to-end slope-aware height estimation network named TS-SatMVSNet for large-scale remote sensing terrain reconstruction.To effectively obtain the slope representation, drawing from mathematical gradient concepts, we innovatively proposed a height-based slope calculation strategy to first calculate a slope map from a height map to measure the terrain undulation. To fully integrate slope information into the MVS pipeline, we separately design two slope-guided modules to enhance reconstruction outcomes at both micro and macro levels. Specifically, at the micro level, we designed a slope-guided interval partition module for refined height estimation using slope values. At the macro level, a height correction module is proposed, using a learnable Gaussian smoothing operator to amend the inaccurate height values. Additionally, to enhance the efficacy of height estimation, we proposed a slope direction loss for implicitly optimizing height estimation results. Extensive experiments on the WHU-TLC dataset and MVS3D dataset show that our proposed method achieves state-of-the-art performance and demonstrates competitive generalization ability.

Automating radiology report generation can significantly reduce the workload of radiologists and enhance the accuracy, consistency, and efficiency of clinical documentation.We propose a novel cross-modal framework that uses MedCLIP as both a vision extractor and a retrieval mechanism to improve the process of medical report generation.By extracting retrieved report features and image features through an attention-based extract module, and integrating them with a fusion module, our method improves the coherence and clinical relevance of generated reports.Experimental results on the widely used IU-Xray dataset demonstrate the effectiveness of our approach, showing improvements over commonly used methods in both report quality and relevance.Additionally, ablation studies provide further validation of the framework, highlighting the importance of accurate report retrieval and feature integration in generating comprehensive medical reports., Comment: Accepted in IEEE Big Data 2024

Controller Area Networks (CANs) are widely adopted in real-time automotive control and are increasingly standard in factory automation. Considering their critical application in safety-critical systems, The error rate of the system must be accurately predicted and guaranteed. Through simulation, it is possible to obtain a low-precision overview of the system's behavior. However, for low-probability events, the required number of samples in simulation increases rapidly, making it difficult to conduct a sufficient number of simulations in practical applications, and the statistical results may deviate from the actual outcomes. Therefore, a formal analysis is needed to evaluate the error rate of the system. This paper improves the worst-case probability response time analysis by using convolution-based busy-window and backlog techniques under the error retransmission protocol of CANs. Empirical analysis shows that the proposed method improves upon existing methods in terms of accuracy and efficiency.

There is a fast-growing literature on estimating optimal treatment rules directly by maximizing the expected outcome. In biomedical studies and operations applications, censored survival outcome is frequently observed, in which case the restricted mean survival time and survival probability are of great interest. In this paper, we propose two robust criteria for learning optimal treatment rules with censored survival outcomes; the former one targets at an optimal treatment rule maximizing the restricted mean survival time, where the restriction is specified by a given quantile such as median; the latter one targets at an optimal treatment rule maximizing buffered survival probabilities, where the predetermined threshold is adjusted to account the restricted mean survival time. We provide theoretical justifications for the proposed optimal treatment rules and develop a sampling-based difference-of-convex algorithm for learning them. In simulation studies, our estimators show improved performance compared to existing methods. We also demonstrate the proposed method using AIDS clinical trial data.

In the fast-growing field of Remote Sensing (RS) image analysis, the gap between massive unlabeled datasets and the ability to fully utilize these datasets for advanced RS analytics presents a significant challenge. To fill the gap, our work introduces an innovative auto-labeling framework named ALPS (Automatic Labeling for Pre-training in Segmentation), leveraging the Segment Anything Model (SAM) to predict precise pseudo-labels for RS images without necessitating prior annotations or additional prompts. The proposed pipeline significantly reduces the labor and resource demands traditionally associated with annotating RS datasets. By constructing two comprehensive pseudo-labeled RS datasets via ALPS for pre-training purposes, our approach enhances the performance of downstream tasks across various benchmarks, including iSAID and ISPRS Potsdam. Experiments demonstrate the effectiveness of our framework, showcasing its ability to generalize well across multiple tasks even under the scarcity of extensively annotated datasets, offering a scalable solution to automatic segmentation and annotation challenges in the field. In addition, the proposed a pipeline is flexible and can be applied to medical image segmentation, remarkably boosting the performance. Note that ALPS utilizes pre-trained SAM to semi-automatically annotate RS images without additional manual annotations. Though every component in the pipeline has bee well explored, integrating clustering algorithms with SAM and novel pseudo-label alignment significantly enhances RS segmentation, as an off-the-shelf tool for pre-training data preparation. Our source code is available at: https://github.com/StriveZs/ALPS.

Few-mode multi-core fiber (FM-MCF) based Space-Division Multiplexing (SDM) systems possess the potential to maximize the number of multiplexed spatial channels per fiber by harnessing both the space (fiber cores) and mode (optical mode per core) dimensions. However, to date, no SDM transmissions over field-deployed FM-MCFs in realistic outdoor settings have been reported, which contrasts with SDM schemes demonstrated using single-mode multi-core fibers (SM-MCFs) installed in practical fiber cable ducts. In this paper, we present the successful demonstration of bidirectional SDM transmission over a 5-km field-deployed seven ring-core fiber (7-RCF) with a cladding diameter of 178 ${\mu}$m, achieving a Spectral Efficiency (SE) of 2$\times$201.6 bit/s/Hz. This work establishes a new record for the highest SE attained in SDM demonstrations utilizing field-deployed fiber cables, achieving an approximate 10x increase compared to the SE of reported field-deployed optical fiber cable transmission systems. Notably, these results are realized through the utilization of small-scale modular 4$\times$4 multiple-input multiple-output (MIMO) processing with a time-domain equalization (TDE) tap number not exceeding 15, maintaining a complexity per unit capacity comparable to that of MIMO equalization in SDM demonstrations employing weakly coupled SM-MCF cables. These results underscore the significant potential for achieving heightened SE and expanding capacity per individual fiber using SDM techniques in practical applications., Comment: 17 pages, 8 figures

This paper proposes a Heaviside composite optimization approach and presents a progressive (mixed) integer programming (PIP) method for solving multi-class classification and multi-action treatment problems with constraints. A Heaviside composite function is a composite of a Heaviside function (i.e., the indicator function of either the open $( \, 0,\infty )$ or closed $[ \, 0,\infty \, )$ interval) with a possibly nondifferentiable function. Modeling-wise, we show how Heaviside composite optimization provides a unified formulation for learning the optimal multi-class classification and multi-action treatment rules, subject to rule-dependent constraints stipulating a variety of domain restrictions. A Heaviside composite function has an equivalent discrete formulation, and the resulting optimization problem can in principle be solved by integer programming (IP) methods. Nevertheless, for constrained learning problems with large data sets, a straightforward application of off-the-shelf IP solvers is usually ineffective in achieving global optimality. To alleviate such a computational burden, our major contribution is the proposal of the PIP method by leveraging the effectiveness of state-of-the-art IP solvers for problems of modest sizes. We provide the theoretical advantage of the PIP method with the connection to continuous optimization and show that the computed solution is locally optimal for a broad class of Heaviside composite optimization problems. The numerical performance of the PIP method is demonstrated by extensive computational experimentation.

The 3D scene editing method based on neural implicit field has gained wide attention. It has achieved excellent results in 3D editing tasks. However, existing methods often blend the interaction between objects and scene environment. The change of scene appearance like shadows is failed to be displayed in the rendering view. In this paper, we propose an Object and Scene environment Interaction aware (OSI-aware) system, which is a novel two-stream neural rendering system considering object and scene environment interaction. To obtain illuminating conditions from the mixture soup, the system successfully separates the interaction between objects and scene environment by intrinsic decomposition method. To study the corresponding changes to the scene appearance from object-level editing tasks, we introduce a depth map guided scene inpainting method and shadow rendering method by point matching strategy. Extensive experiments demonstrate that our novel pipeline produce reasonable appearance changes in scene editing tasks. It also achieve competitive performance for the rendering quality in novel-view synthesis tasks.

Recommender systems are essential information technologies today, and recommendation algorithms combined with deep learning have become a research hotspot in this field. The recommendation model known as LFM (Latent Factor Model), which captures latent features through matrix factorization and gradient descent to fit user preferences, has given rise to various recommendation algorithms that bring new improvements in recommendation accuracy. However, collaborative filtering recommendation models based on LFM lack flexibility and has shortcomings for real-time recommendations, as they need to redo the matrix factorization and retrain using gradient descent when new users arrive. In response to this, this paper innovatively proposes a Latent Factor Generator (LFG) network, and set the movie recommendation as research theme. The LFG dynamically generates user latent factors through deep neural networks without the need for re-factorization or retrain. Experimental results indicate that the LFG recommendation model outperforms traditional matrix factorization algorithms in recommendation accuracy, providing an effective solution to the challenges of real-time recommendations with LFM., Comment: 9 pages, 1 figure, 4 tables. Source code would be uploaded to github soon

Since ChatGPT released its API for public use, the number of applications built on top of commercial large language models (LLMs) increase exponentially. One popular usage of such models is leveraging its in-context learning ability and generating responses given user queries leveraging knowledge obtained by retrieval augmentation. One problem of deploying commercial retrieval-augmented LLMs is the cost due to the additionally retrieved context that largely increases the input token size of the LLMs. To mitigate this, we propose a token compression scheme that includes two methods: summarization compression and semantic compression. The first method applies a T5-based model that is fine-tuned by datasets generated using self-instruct containing samples with varying lengths and reduce token size by doing summarization. The second method further compresses the token size by removing words with lower impact on the semantic. In order to adequately evaluate the effectiveness of the proposed methods, we propose and utilize a dataset called Food-Recommendation DB (FRDB) focusing on food recommendation for women around pregnancy period or infants. Our summarization compression can reduce 65% of the retrieval token size with further 0.3% improvement on the accuracy; semantic compression provides a more flexible way to trade-off the token size with performance, for which we can reduce the token size by 20% with only 1.6% of accuracy drop., Comment: EMNLP 2023 Findings

Multi-View Stereo~(MVS) is a fundamental problem in geometric computer vision which aims to reconstruct a scene using multi-view images with known camera parameters. However, the mainstream approaches represent the scene with a fixed all-pixel depth range and equal depth interval partition, which will result in inadequate utilization of depth planes and imprecise depth estimation. In this paper, we present a novel multi-stage coarse-to-fine framework to achieve adaptive all-pixel depth range and depth interval. We predict a coarse depth map in the first stage, then an Adaptive Depth Range Prediction module is proposed in the second stage to zoom in the scene by leveraging the reference image and the obtained depth map in the first stage and predict a more accurate all-pixel depth range for the following stages. In the third and fourth stages, we propose an Adaptive Depth Interval Adjustment module to achieve adaptive variable interval partition for pixel-wise depth range. The depth interval distribution in this module is normalized by Z-score, which can allocate dense depth hypothesis planes around the potential ground truth depth value and vice versa to achieve more accurate depth estimation. Extensive experiments on four widely used benchmark datasets~(DTU, TnT, BlendedMVS, ETH 3D) demonstrate that our model achieves state-of-the-art performance and yields competitive generalization ability. Particularly, our method achieves the highest Acc and Overall on the DTU dataset, while attaining the highest Recall and $F_{1}$-score on the Tanks and Temples intermediate and advanced dataset. Moreover, our method also achieves the lowest $e_{1}$ and $e_{3}$ on the BlendedMVS dataset and the highest Acc and $F_{1}$-score on the ETH 3D dataset, surpassing all listed methods.Project website: https://github.com/zs670980918/ARAI-MVSNet

There are many significant applied contexts that require the solution of discontinuous optimization problems in finite dimensions. Yet these problems are very difficult, both computationally and analytically. With the functions being discontinuous and a minimizer (local or global) of the problems, even if it exists, being impossible to verifiably compute, a foremost question is what kind of ''stationary solutions'' one can expect to obtain; these solutions provide promising candidates for minimizers; i.e., their defining conditions are necessary for optimality. Motivated by recent results on sparse optimization, we introduce in this paper such a kind of solution, termed ''pseudo B- (for Bouligand) stationary solution'', for a broad class of discontinuous piecewise continuous optimization problems with objective and constraint defined by indicator functions of the positive real axis composite with functions that are possibly nonsmooth. We present two approaches for computing such a solution. One approach is based on lifting the problem to a higher dimension via the epigraphical formulation of the indicator functions; this requires the addition of some auxiliary variables. The other approach is based on certain continuous (albeit not necessarily differentiable) piecewise approximations of the indicator functions and the convergence to a pseudo B-stationary solution of the original problem is established. The conditions for convergence are discussed and illustrated by an example.

Focusing on stochastic programming (SP) with covariate information, this paper proposes an empirical risk minimization (ERM) method embedded within a nonconvex piecewise affine decision rule (PADR), which aims to learn the direct mapping from features to optimal decisions. We establish the nonasymptotic consistency result of our PADR-based ERM model for unconstrained problems and asymptotic consistency result for constrained ones. To solve the nonconvex and nondifferentiable ERM problem, we develop an enhanced stochastic majorization-minimization algorithm and establish the asymptotic convergence to (composite strong) directional stationarity along with complexity analysis. We show that the proposed PADR-based ERM method applies to a broad class of nonconvex SP problems with theoretical consistency guarantees and computational tractability. Our numerical study demonstrates the superior performance of PADR-based ERM methods compared to state-of-the-art approaches under various settings, with significantly lower costs, less computation time, and robustness to feature dimensions and nonlinearity of the underlying dependency.

In-memory ordered key-value stores are an important building block in modern distributed applications. We present Honeycomb, a hybrid software-hardware system for accelerating read-dominated workloads on ordered key-value stores that provides linearizability for all operations including scans. Honeycomb stores a B-Tree in host memory, and executes SCAN and GET on an FPGA-based SmartNIC, and PUT, UPDATE and DELETE on the CPU. This approach enables large stores and simplifies the FPGA implementation but raises the challenge of data access and synchronization across the slow PCIe bus. We describe how Honeycomb overcomes this challenge with careful data structure design, caching, request parallelism with out-of-order request execution, wait-free read operations, and batching synchronization between the CPU and the FPGA. For read-heavy YCSB workloads, Honeycomb improves the throughput of a state-of-the-art ordered key-value store by at least 1.8x. For scan-heavy workloads inspired by cloud storage, Honeycomb improves throughput by more than 2x. The cost-performance, which is more important for large-scale deployments, is improved by at least 1.5x on these workloads.

Cybersecurity breaches are the common anomalies for distributed cyber-physical systems (CPS). However, the cyber security breach classification is still a difficult problem, even using cutting-edge artificial intelligence (AI) approaches. In this paper, we study the multi-class classification problem in cyber security for attack detection. A challenging multi-node data-censoring case is considered. In such a case, data within each data center/node cannot be shared while the local data is incomplete. Particularly, local nodes contain only a part of the multiple classes. In order to train a global multi-class classifier without sharing the raw data across all nodes, the main result of our study is designing a multi-node multi-class classification ensemble approach. By gathering the estimated parameters of the binary classifiers and data densities from each local node, the missing information for each local node is completed to build the global multi-class classifier. Numerical experiments are given to validate the effectiveness of the proposed approach under the multi-node data-censoring case. Under such a case, we even show the out-performance of the proposed approach over the full-data approach., Comment: 22 pages. Submitted to ACM Transactions on Sensor Networks (TOSN)

CoqQ is a framework for reasoning about quantum programs in the Coq proof assistant. Its main components are: a deeply embedded quantum programming language, in which classic quantum algorithms are easily expressed, and an expressive program logic for proving properties of programs. CoqQ is foundational: the program logic is formally proved sound with respect to a denotational semantics based on state-of-art mathematical libraries (mathcomp and mathcomp analysis). CoqQ is also practical: assertions can use Dirac expressions, which eases concise specifications, and proofs can exploit local and parallel reasoning, which minimizes verification effort. We illustrate the applicability of CoqQ with many examples from the literature.

In this paper, we introduce VeriQBench -- an open source benchmark for quantum circuits. It offers high-level quantum circuit abstractions of various circuit types, including 1) combinational, 2) dynamic, 3) sequential, and 4) variational quantum circuits, which cover almost all existing types of quantum circuits in the literature. Meanwhile, VeriQBench is a versatile benchmark which can be used in verifying quantum software for different applications, as is evidenced by the existing works including quantum circuit verification (e.g., equivalence checking [Hon+21a; WLY21] and model checking [Yin21]), simulation (e.g., fault simulation), testing (e.g., test pattern generation [CY22]) and debugging (e.g., runtime assertions [Li+20b]). All the circuits are described in OpenQASM and are validated on Qiskit and QCOR simulators. With the hope that it can be used by other researchers, VeriQBench is released at: https://github.com/Veri-Q/Benchmark.

We introduce isQ, a new software stack for quantum programming in an imperative programming language, also named isQ. The aim of isQ is to make the programmers write quantum programs as conveniently as possible. In particular: 1) The isQ language and its compiler contain many features, including some not well supported by (most) other quantum programming platforms, e.g. classical control flow such as recursion; decomposition of selfdefined unitary gates; and oracle programming and its circuit realization. 2) To make it flexible, an isQ program can be compiled into several kinds of intermediate representation, including OpenQASM 3.0, QIR and QCIS (specially tailored for the superconducting quantum hardware at USTC). 3) Besides interfacing isQ with true superconducting hardware, a QIR simulator is also developed for demonstration and testing of isQ programs., Comment: This old version was not well written. And a new version of this paper is published (open access) on TQE. DOI:doi.org/10.1109/TQE.2023.3275868

We propose a series of quantum algorithms for computing a wide range of quantum entropies and distances, including the von Neumann entropy, quantum R\'{e}nyi entropy, trace distance, and fidelity. The proposed algorithms significantly outperform the prior best (and even quantum) ones in the low-rank case, some of which achieve exponential speedups. In particular, for $N$-dimensional quantum states of rank $r$, our proposed quantum algorithms for computing the von Neumann entropy, trace distance and fidelity within additive error $\varepsilon$ have time complexity of $\tilde O(r/\varepsilon^2)$, $\tilde O(r^5/\varepsilon^6)$ and $\tilde O(r^{6.5}/\varepsilon^{7.5})$, respectively. By contrast, prior quantum algorithms for the von Neumann entropy and trace distance usually have time complexity $\Omega(N)$, and the prior best one for fidelity has time complexity $\tilde O(r^{12.5}/\varepsilon^{13.5})$. The key idea of our quantum algorithms is to extend block-encoding from unitary operators in previous work to quantum states (i.e., density operators). It is realized by developing several convenient techniques to manipulate quantum states and extract information from them. The advantage of our techniques over the existing methods is that no restrictions on density operators are required; in sharp contrast, the previous methods usually require a lower bound on the minimal non-zero eigenvalue of density operators., Comment: Final version. 58 pages, 5 tables, 1 figure. Minor corrections to Theorem 3.1, Theorem 3.4, and Corollary 3.5

Chance-constrained programs (CCPs) constitute a difficult class of stochastic programs due to its possible nondifferentiability and nonconvexity even with simple linear random functionals. Existing approaches for solving the CCPs mainly deal with convex random functionals within the probability function. In the present paper, we consider two generalizations of the class of chance constraints commonly studied in the literature; one generalization involves probabilities of disjunctive nonconvex functional events and the other generalization involves mixed-signed affine combinations of the resulting probabilities; together, we coin the term affine chance constraint (ACC) system for these generalized chance constraints. Our proposed treatment of such an ACC system involves the fusion of several individually known ideas: (a) parameterized upper and lower approximations of the indicator function in the expectation formulation of probability; (b) external (i.e., fixed) versus internal (i.e., sequential) sampling-based approximation of the expectation operator; (c) constraint penalization as relaxations of feasibility; and (d) convexification of nonconvexity and nondifferentiability via surrogation. The integration of these techniques for solving the affine chance-constrained stochastic program (ACC-SP) with various degrees of practicality and computational efforts is the main contribution of this paper.