Runtime Performance Research Articles

Leveraging Incremental Machine Learning for Reconfigurable Systems Modeling under Dynamic Workloads

Dynamic workload orchestration is one of the main concerns when working with heterogeneous computing infrastructures in the edge-cloud continuum. In this context, FPGA-based computing nodes can take advantage of their improved flexibility, performance and energy efficiency provided that they use proper resource management strategies. In this regard, many state-of-the-art systems rely on proactive power management techniques and task scheduling decisions, which in turn require deep knowledge about the applications to be accelerated and the actual response of the target reconfigurable fabrics when executing them. While acquiring this knowledge at design time was more or less feasible in the past, with applications mostly being static task graphs that did not change at run time, the highly dynamic nature of current workloads in the edge-cloud continuum, where tasks can be deployed on any node and at any time, has removed this possibility. As a result, being able to derive such information at run time to make informed decisions has become a must. This paper presents an infrastructure to build incremental ML models that can be used to obtain run-time power consumption and performance estimations in FPGA-based reconfigurable multi-accelerator systems operating under dynamic workloads. The proposed infrastructure features a novel stop-and-restart resource-aware mechanism to monitor and control the model training and evaluation stages during normal system operation, enabling low-overhead updates in the models to account for either unexpected acceleration requests (i.e., tasks not considered previously by the models) or model drift (e.g., fabric degradation). Experimental results show that the proposed approach induces a maximal additional error of 3.66% compared to a continuous training alternative. Furthermore, the proposed approach incurs only a 4.49% execution time overhead, compared to the 20.91% overhead induced by the continuous training alternative. The proposed modeling strategy enables innovative scheduling approaches in reconfigurable systems. This is exemplified by the conflict-aware scheduler introduced in this work, which achieves up to a 1.35 times speedup in executing the experimental workload. Additionally, the proposed approach demonstrates superior adaptability compared to other methods in the literature, particularly in response to significant changes in workload and to mitigate the effects of model overfitting. The portability of the proposed modeling methodology and monitoring infrastructure is also shown through their application to both Zynq-7000 and Zynq UltraScale+ devices.

RepSGG: Novel Representations of Entities and Relationships for Scene Graph Generation.

Scene Graph Generation (SGG) has achieved significant progress recently. However, most previous works rely heavily on fixed-size entity representations based on bounding box proposals, anchors, or learnable queries. As each representation's cardinality has different trade-offs between performance and computation overhead, extracting highly representative features efficiently and dynamically is both challenging and crucial for SGG. In this work, a novel architecture called RepSGG is proposed to address the aforementioned challenges, formulating a subject as queries, an object as keys, and their relationship as the maximum attention weight between pairwise queries and keys. With more fine-grained and flexible representation power for entities and relationships, RepSGG learns to sample semantically discriminative and representative points for relationship inference. Moreover, the long-tailed distribution also poses a significant challenge for generalization of SGG. A run-time performance-guided logit adjustment (PGLA) strategy is proposed such that the relationship logits are modified via affine transformations based on run-time performance during training. This strategy encourages a more balanced performance between dominant and rare classes. Experimental results show that RepSGG achieves the state-of-the-art or comparable performance on the Visual Genome and Open Images V6 datasets with fast inference speed, demonstrating the efficacy and efficiency of the proposed methods.

Efficient compiler optimization by modeling passes dependence

Selecting the optimal combination of compiler passes is a significant challenge to enhance performance and reduce the code size of compiled binaries. While a well-selected sequence of compiler passes can yield considerable benefits, the large number of potential combinations and the scarcity of effective ones make this task prohibitively complex. To tackle this problem, we propose a novel approach to group compiler passes into a small set of sub-sequences. This approach translates the task of identifying the right compiler passes combination into determining the appropriate combination of these sub-sequences. We apply our approach to CBench and PolyBench, demonstrating remarkable performance improvements. Our approach enhances runtime performance by 22% compared to the default LLVM ‘O3’ option, and achieves a code size reduction of 24% compared to the ‘Oz’ option. Our approach also outperforms state-of-the-art across various optimization tasks and hardware platforms.

A Biochemistry-Inspired Algorithm for Path Planning in Unmanned Ground Vehicles

Unmanned ground vehicles (UGVs) have gained significant attention due to their extensive applications in both military and civilian sectors. For effective UGV deployment, path planning algorithms must prioritize computational efficiency, solution reliability, and runtime performance while maintaining path quality. Autonomous path planning remains a critical challenge in UGV navigation, as conventional methods, while effective, often suffer from considerable computational overhead. To address this issue, we propose a novel biochemistry-inspired path planning algorithm designed specifically for static grid-based scenarios. MetaPath demonstrates remarkable computational efficiency while maintaining solution quality across different obstacle densities in benchmark environments. Specifically, the algorithm achieves path lengths within ±5% of all benchmark algorithms while dramatically reducing the exploration space, visiting up to 10% of the cells explored by conventional approaches such as A*. This superior efficiency translates into exceptional runtime performance, executing up to 3000 times faster than bio-inspired algorithms like Ant Colony Optimization (ACO) and the Genetic Algorithm (GA), performing nearly three times faster than the widely used A* algorithm, and maintaining competitive performance with efficient algorithms like Breadth-First Search (BFS) and Particle Swarm Optimization (PSO), thereby establishing the algorithm as a highly efficient pathfinding solution. Most notably, MetaPath introduces a novel approach as the first chemistry-inspired pathfinding algorithm, guaranteeing path discovery when one exists within reasonable computational time, a crucial advantage over some benchmark algorithms that may fail to converge or require excessive computational resources in complex scenarios.

TargetCall: eliminating the wasted computation in basecalling via pre-basecalling filtering.

Basecalling is an essential step in nanopore sequencing analysis where the raw signals of nanopore sequencers are converted into nucleotide sequences, that is, reads. State-of-the-art basecallers use complex deep learning models to achieve high basecalling accuracy. This makes basecalling computationally inefficient and memory-hungry, bottlenecking the entire genome analysis pipeline. However, for many applications, most reads do not match the reference genome of interest (i.e., target reference) and thus are discarded in later steps in the genomics pipeline, wasting the basecalling computation. To overcome this issue, we propose TargetCall, the first pre-basecalling filter to eliminate the wasted computation in basecalling. TargetCall's key idea is to discard reads that will not match the target reference (i.e., off-target reads) prior to basecalling. TargetCall consists of two main components: (1) LightCall, a lightweight neural network basecaller that produces noisy reads, and (2) Similarity Check, which labels each of these noisy reads as on-target or off-target by matching them to the target reference. Our thorough experimental evaluations show that TargetCall 1) improves the end-to-end basecalling runtime performance of the state-of-the-art basecaller by while maintaining high recall in keeping on-target reads, 2) maintains high accuracy in downstream analysis, and 3) achieves better runtime performance, throughput, recall, precision, and generality than prior works. TargetCall is available at https://github.com/CMU-SAFARI/TargetCall.

RT-APT: A real-time APT anomaly detection method for large-scale provenance graph

Advanced Persistent Threats (APTs) are prevalent in the field of cyber attacks, where attackers employ advanced techniques to control targets and exfiltrate data without being detected by the system. Existing APT detection methods heavily rely on expert rules or specific training scenarios, resulting in the lack of both generality and reliability. Therefore, this paper proposes a novel real-time APT attack anomaly detection system for large-scale provenance graphs, named RT-APT. Firstly, a provenance graph is constructed with kernel logs, and the WL subtree kernel algorithm is utilized to aggregate contextual information of nodes in the provenance graph. In this way we obtain vector representations. Secondly, the FlexSketch algorithm transforms the streaming provenance graph into a sequence of feature vectors. Finally, the K-means clustering algorithm is performed on benign feature vector sequences, where each cluster represents a different system state. Thus, we can identify abnormal behaviors during system execution. Therefore RT-APT enables to detect unknown attacks and extract long-term system behaviors. Experiments have been carried out to explore the optimal parameter settings under which RT-APT can perform best. In addition, we compare RT-APT and the state-of-the-art approaches on three datasets, Laboratory, StreamSpot and Unicorn. Results demonstrate that our proposed method outperforms the state-of-the-art approaches from the perspective of runtime performance, memory overhead and CPU usage.

TriHuman: A Real-time and Controllable Tri-plane Representation for Detailed Human Geometry and Appearance Synthesis

Creating controllable, photorealistic, and geometrically detailed digital doubles of real humans solely from video data is a key challenge in Computer Graphics and Vision, especially when real-time performance is required. Recent methods attach a neural radiance field (NeRF) to an articulated structure, e.g., a body model or a skeleton, to map points into a pose canonical space while conditioning the NeRF on the skeletal pose. These approaches typically parameterize the neural field with a multi-layer perceptron (MLP) leading to a slow runtime. To address this drawback, we propose TriHuman a novel human-tailored, deformable, and efficient tri-plane representation, which achieves real-time performance, state-of-the-art pose-controllable geometry synthesis as well as photorealistic rendering quality. At the core, we non-rigidly warp global ray samples into our undeformed tri-plane texture space, which effectively addresses the problem of global points being mapped to the same tri-plane locations. We then show how such a tri-plane feature representation can be conditioned on the skeletal motion to account for dynamic appearance and geometry changes. Our results demonstrate a clear step toward higher quality in terms of geometry and appearance modeling of humans as well as runtime performance.

Fully Verified Instruction Scheduling

CompCert project, the state-of-the-art compiler that achieves the first end-to-end formally verified C compiler, does not support fully verified instruction scheduling. Instead, existing research that works on such topics only implements translation validation. This means they do not have direct formal proof that the scheduling algorithm is correct, but only a posterior validation to check each compiling case. Using such a method, CompCert accepts a valid C program and compiles correctly only when the untrusted scheduler generates a correct result. However, it does not guarantee the complete correctness of the scheduler. It also causes compile-time validation overhead in the view of runtime performance. In this work, we present the first achievement in developing a mechanized library for fully verified instruction scheduling while keeping the proof workload acceptably lightweight. The idea to reduce the proof length is to exploit a simple property that the topological reordering of a topological sorted list is equal to a sequence of swapping adjacent unordered elements. Together with the transitivity of semantic simulation relation, the only burden will become proving the semantic preservation of a transition that only swaps two adjacent independent instructions inside one block. After successfully proving this result, proving the correctness of any new instruction scheduling algorithm only requires proof that it preserved the syntax-level dependence among instructions, instead of reasoning about semantics details every time. We implemented a mechanized library of such methods in the Coq proof assistant based on CompCert's library as a framework and used the list scheduling algorithm as a case study to show the correctness can be formally proved using our theory. We show that with our method that abstracts away the semantics details, it is flexible to implement any scheduler that reorders instructions with little extra proof burden. Our scheduler in the case study also abstracts away the outside scheduling heuristic as a universal parameter so it is flexible to modify without touching any correctness proof.

Comparative analysis of lattice-based cryptographic schemes for secure IoT communications

This paper presents comparative analysis of lattice-based cryptographic schemes, focusing on their runtime and memory performance to evaluate their suitability for secure IoT communications. Specifically, analyzing FrodoKEM, Kyber, and sntrup761, assessing their key generation, encapsulation, and decapsulation operations across various IoT processors. Kyber schemes, particularly Kyber512 and Kyber1024, demonstrate the highest efficiency in terms of battery utilization and computational speed, making them ideal for highly constrained IoT devices. FrodoKEM variants, while providing robust security, show moderate performance, suitable for devices with more processing capacity. sntrup761, however, is less efficient, highlighting the need for careful selection of cryptographic schemes based on specific IoT requirements. The findings underscore the importance of balancing security, speed, and resource usage in PQC implementations, with Kyber schemes emerging as the most promising candidates for secure, resource-constrained IoT environments.

Split-explicit external mode solver in the finite volume sea ice–ocean model FESOM2

Abstract. A novel split-explicit (SE) external mode solver for the Finite volumE Sea ice–Ocean Model (FESOM2) and its sub-versions is presented. It is compared with the semi-implicit (SI) solver currently used in FESOM2. The split-explicit solver for the external mode utilizes a dissipative asynchronous (forward–backward) time-stepping scheme that does not require additional filtering during the barotropic sub-cycling. Its implementation with arbitrary Lagrangian–Eulerian vertical coordinates like Z star (z*) and Z tilde (z̃) is explored. The comparisons are performed through multiple test cases involving idealized and realistic global simulations. The SE solver demonstrates lower phase errors and dissipation, while maintaining a simulated mean ocean state very similar to the SI solver. The SE solver is also shown to possess better run-time performance and parallel scalability across all workloads tested.

FL-DSFA: Securing RPL-Based IoT Networks against Selective Forwarding Attacks Using Federated Learning.

The Internet of Things (IoT) is a significant technological advancement that allows for seamless device integration and data flow. The development of the IoT has led to the emergence of several solutions in various sectors. However, rapid popularization also has its challenges, and one of the most serious challenges is the security of the IoT. Security is a major concern, particularly routing attacks in the core network, which may cause severe damage due to information loss. Routing Protocol for Low-Power and Lossy Networks (RPL), a routing protocol used for IoT devices, is faced with selective forwarding attacks. In this paper, we present a federated learning-based detection technique for detecting selective forwarding attacks, termed FL-DSFA. A lightweight model involving the IoT Routing Attack Dataset (IRAD), which comprises Hello Flood (HF), Decreased Rank (DR), and Version Number (VN), is used in this technique to increase the detection efficiency. The attacks on IoT threaten the security of the IoT system since they mainly focus on essential elements of RPL. The components include control messages, routing topologies, repair procedures, and resources within sensor networks. Binary classification approaches have been used to assess the training efficiency of the proposed model. The training step includes the implementation of machine learning algorithms, including logistic regression (LR), K-nearest neighbors (KNN), support vector machine (SVM), and naive Bayes (NB). The comparative analysis illustrates that this study, with SVM and KNN classifiers, exhibits the highest accuracy during training and achieves the most efficient runtime performance. The proposed system demonstrates exceptional performance, achieving a prediction precision of 97.50%, an accuracy of 95%, a recall rate of 98.33%, and an F1 score of 97.01%. It outperforms the current leading research in this field, with its classification results, scalability, and enhanced privacy.

Q-Seg: Quantum Annealing-Based Unsupervised Image Segmentation.

We present Q-Seg, a novel unsupervised image segmentation method based on quantum annealing, tailored for existing quantum hardware. We formulate the pixelwise segmentation problem, which assimilates spectral and spatial information of the image, as a graph-cut optimization task. Our method efficiently leverages the interconnected qubit topology of the D-wave advantage device, offering superior scalability over existing quantum approaches and outperforming several tested state-of-the-art classical methods. Empirical evaluations on synthetic datasets have shown that Q-Seg has better runtime performance than the state-of-the-art classical optimizer Gurobi. The method has also been tested on earth observation image segmentation, a critical area with noisy and unreliable annotations. In the era of noisy intermediate-scale quantum, Q-Seg emerges as a reliable contender for real-world applications in comparison to advanced techniques like Segment Anything. Consequently, Q-Seg offers a promising solution using available quantum hardware, especially in situations constrained by limited labeled data and the need for efficient computational runtime.

Approximating problems in abstract argumentation with graph convolutional networks

In this article, we present a novel approximation approach for abstract argumentation using a customized Graph Convolutional Network (GCN) architecture and a tailored training method. Our approach demonstrates promising results in approximating abstract argumentation tasks across various semantics, setting a new state of the art for performance on certain tasks. We provide a detailed analysis of approximation and runtime performance and propose a new scheme for evaluation. By advancing the state of the art for approximating the acceptability status of abstract arguments, we make theoretical and empirical advances in understanding the limits and opportunities for approximation in this field. Our approach shows potential for creating both general purpose and task-specific approximators and offers insights into the performance differences across benchmarks and semantics.

Longitudinal registration of T1-weighted breast MRI: A registration algorithm (FLIRE) and clinical application

PurposeMRI is commonly used to aid breast cancer diagnosis and treatment evaluation. For patients with breast cancer, neoadjuvant chemotherapy aims to reduce the tumor size and extent of surgery necessary. The current clinical standard to measure breast tumor response on MRI uses the longest tumor diameter. Radiologists also account for other tissue properties including tumor contrast or pharmacokinetics in their assessment. Accurate longitudinal image registration of breast tissue is critical to properly compare response to treatment at different timepoints. MethodsIn this study, a deformable Fast Longitudinal Image Registration (FLIRE) algorithm was optimized for breast tissue. FLIRE was then compared to the publicly available software packages with high accuracy (DRAMMS) and fast runtime (Elastix). Patients included in the study received longitudinal T1-weighted MRI without fat saturation at two to six timepoints as part of asymptomatic screening (n = 27) or throughout neoadjuvant chemotherapy treatment (n = 32). T1-weighted images were registered to the first timepoint with each algorithm. ResultsAlignment and runtime performance were compared using two-way repeated measure ANOVAs (P < 0.05). Across all patients, Pearson's correlation coefficient across the entire image volume was slightly higher with statistical significance and had less variance for FLIRE (0.98 ± 0.01 stdev) compared to DRAMMS (0.97 ± 0.03 stdev) and Elastix (0.95 ± 0.03 stdev). Additionally, FLIRE runtime (10.0 mins) was 9.0 times faster than DRAMMS (89.6 mins) and 1.5 times faster than Elastix (14.5 mins) on a Linux workstation. ConclusionFLIRE demonstrates promise for time-sensitive clinical applications due to its accuracy, robustness across patients and timepoints, and speed.

Modeling Sequences as Star Graphs to Address Over-Smoothing in Self-Attentive Sequential Recommendation

Self-attention (SA) mechanisms have been widely used in developing sequential recommendation (SR) methods, and demonstrated state-of-the-art performance. However, in this article, we show that self-attentive SR methods substantially suffer from the over-smoothing issue that item embeddings within a sequence become increasingly similar across attention blocks. As widely demonstrated in the literature, this issue could lead to a loss of information in individual items, and significantly degrade models’ scalability and performance. To address the over-smoothing issue, in this article, we view items within a sequence constituting a star graph and develop a method, denoted as \(\mathop{\mathtt{MSSG}}\limits\) , for SR. Different from existing self-attentive methods, \(\mathop{\mathtt{MSSG}}\limits\) introduces an additional internal node to specifically capture the global information within the sequence, and does not require information propagation among items. This design fundamentally addresses the over-smoothing issue and enables \(\mathop{\mathtt{MSSG}}\limits\) a linear time complexity with respect to the sequence length. We compare \(\mathop{\mathtt{MSSG}}\limits\) with eleven state-of-the-art baseline methods on six public benchmark datasets. Our experimental results demonstrate that \(\mathop{\mathtt{MSSG}}\limits\) significantly outperforms the baseline methods, with an improvement of as much as 10.10%. Our analysis shows the superior scalability of \(\mathop{\mathtt{MSSG}}\limits\) over the state-of-the-art self-attentive methods. Our complexity analysis and runtime performance comparison together show that \(\mathop{\mathtt{MSSG}}\limits\) is both theoretically and practically more efficient than self-attentive methods. Our analysis of the attention weights learned in SA-based methods indicates that on sparse recommendation data, modeling dependencies in all item pairs using the SA mechanism yields limited information gain, and thus, might not benefit the recommendation performance. Our source code and data are publicly accessible through GitHub .

Intelligent algorithm selection for efficient update predictions in social media feeds

Efficiently synchronizing data with social media feeds while minimizing unnecessary requests presents a significant challenge in various fields. This paper investigates prediction algorithms for determining the optimal update intervals for Facebook and Twitter (now X) feeds, focusing on metrics such as delay (the time between a post’s publication and its retrieval) and requests per post. Variations in update intervals result in different algorithms producing varying results, making the selection of the most suitable algorithm for each feed crucial yet time-intensive. To address this, we propose three strategies for algorithm selection: baseline (applying a single algorithm to all feeds), optimum (identifying the best algorithm for each individual feed), and classification (selecting algorithms through a classification process based on each feed’s unique update patterns and context). Our strategies leverage various prediction algorithms, including static and adaptive algorithms, and inhomogeneous Poisson processes. We evaluate these strategies using real-world data from Facebook and Twitter, thoroughly assessing their performance in terms of delay and request efficiency. The findings demonstrate that the strategy Optimum effectively identifies the best algorithms for each feed, ensuring the highest prediction quality, though at a considerable computational cost. On the other hand, the strategy Classification offers superior runtime performance required to select algorithms. This research highlights the trade-offs between delay and request efficiency and presents a comprehensive solution for optimizing update predictions in social media feeds.

Optimization Applications as Quantum Performance Benchmarks

Combinatorial optimization is anticipated to be one of the primary use cases for quantum computation in the coming years. The Quantum Approximate Optimization Algorithm and Quantum Annealing can potentially demonstrate significant run-time performance benefits over current state-of-the-art solutions. Inspired by existing methods to characterize classical optimization algorithms, we analyze the solution quality obtained by solving Max-cut problems using gate-model quantum devices and a quantum annealing device. This is used to guide the development of an advanced benchmarking framework for quantum computers designed to evaluate the trade-off between run-time execution performance and the solution quality for iterative hybrid quantum-classical applications. The framework generates performance profiles through compelling visualizations that show performance progression as a function of time for various problem sizes and illustrates algorithm limitations uncovered by the benchmarking approach. As an illustration, we explore the factors that influence quantum computing system throughput, using results obtained through execution on various quantum simulators and quantum hardware systems.

Runtime performance of a GAMESS quantum chemistry application offloaded to GPUs

SummaryComputational chemistry is at the forefront of solving urgent societal problems, such as polymer upcycling and carbon capture. The complexity of modeling these processes at appropriate length and time scales is mainly manifested in the number and types of chemical species involved in the reactions and may require models of several thousand atoms and large basis sets to accurately capture the chemical complexity and heterogeneity in the physical and chemical processes. The quantum chemistry package General Atomic and Molecular Electronic Structure System (GAMESS) has a wide array of methods that can efficiently and accurately treat complex chemical systems. In this work, we have used the GAMESS Effective Fragment Molecule Orbital (EFMO) method for electronic structure calculation of a challenging mesoporous silica nanoparticle (MSN) model surrounded by about 4700 water molecules to investigate the strong scaling and GPU offloading on hybrid CPU‐GPU nodes. Experiments were performed on the Perlmutter platform at the National Energy Research Scientific Computing Center. Good strong scaling and load balancing have been observed on up to 88 hybrid nodes for different settings of the execution parameters for the calculation considered here. When GPUs are oversubscribed by offloading work from multiple CPU processes, using the NVIDIA multi‐process service (MPS) has consistently reduced time to solution and energy consumed. Additionally, for some configuration parameter settings, oversubscription with MPS improved performance by up to 5.8% over the case without oversubscription.

The Application of Big Data in the Management of Ideological and Political Education in the Development of Education Network

Based on a brief analysis of the current situation of university education management and research on intelligent algorithms, this article constructs a university education management system based on big data. For the clustering and prediction modules in higher education management, corresponding algorithms are used for optimization design. A fuzzy clustering algorithm based on entropy weight is proposed to address the shortcomings of the C-means clustering algorithm. This algorithm adds weighting coefficients on the basis of improvement and continuously updates the clustering centers. The prediction module uses Apriori algorithm to map and compress the target transaction database, reduces the number of candidate item sets through pruning process, and designs simulation experiments to measure the performance of the algorithm. The simulation results show that the clustering results of this algorithm are closer to the actual clustering situation, with shorter running time and better algorithm performance.

Efficient all-electron hybrid density functionals for atomistic simulations beyond 10 000 atoms.

Hybrid density functional approximations (DFAs) offer compelling accuracy for abinitio electronic-structure simulations of molecules, nanosystems, and bulk materials, addressing some deficiencies of computationally cheaper, frequently used semilocal DFAs. However, the computational bottleneck of hybrid DFAs is the evaluation of the non-local exact exchange contribution, which is the limiting factor for the application of the method for large-scale simulations. In this work, we present a drastically optimized resolution-of-identity-based real-space implementation of the exact exchange evaluation for both non-periodic and periodic boundary conditions in the all-electron code FHI-aims, targeting high-performance central processing unit (CPU) compute clusters. The introduction of several new refined message passing interface (MPI) parallelization layers and shared memory arrays according to the MPI-3 standard were the key components of the optimization. We demonstrate significant improvements of memory and performance efficiency, scalability, and workload distribution, extending the reach of hybrid DFAs to simulation sizes beyond ten thousand atoms. In addition, we also compare the runtime performance of the PBE, HSE06, and PBE0 functionals. As a necessary byproduct of this work, other code parts in FHI-aims have been optimized as well, e.g., the computation of the Hartree potential and the evaluation of the force and stress components. We benchmark the performance and scaling of the hybrid DFA-based simulations for a broad range of chemical systems, including hybrid organic-inorganic perovskites, organic crystals, and ice crystals with up to 30 576 atoms (101 920 electrons described by 244 608 basis functions).