Approximate Computation Research Articles

Integrating control-theoretic predictive deep learning for resource-efficient computing systems

Abstract Deep learning (DL) powers numerous applications on ubiquitous edge devices, but its high resource demands pose a challenge. Approximate computing is often proposed to alleviate this, yet such calculation usually suffers from a fixed level of accuracy loss. We propose a novel control-theoretic approach for predictive DL inference on resource constrained devices. Our system dynamically adjusts approximation levels based on a trade-off between resource utilisation and accuracy, considering future demands. Extensive experiments across diverse domains - human activity recognition, acoustic scene profiling, and computer vision - with various neural network architectures and approximation techniques, demonstrate that our approach achieves up to 50% energy savings while maintaining the desired inference accuracy and incurring minimal runtime overhead. Furthermore, we showcase our method in a real-world deployment on low-power edge devices and confirm its superiority over current state-of-the-art solutions.

Approximate Computing Survey, Part II: Application-Specific & Architectural Approximation Techniques and Applications

The challenging deployment of compute-intensive applications from domains such as Artificial Intelligence (AI) and Digital Signal Processing (DSP), forces the community of computing systems to explore new design approaches. Approximate Computing appears as an emerging solution, allowing to tune the quality of results in the design of a system in order to improve the energy efficiency and/or performance. This radical paradigm shift has attracted interest from both academia and industry, resulting in significant research on approximation techniques and methodologies at different design layers (from system down to integrated circuits). Motivated by the wide appeal of Approximate Computing over the last 10 years, we conduct a two-part survey to cover key aspects (e.g., terminology and applications) and review the state-of-the art approximation techniques from all layers of the traditional computing stack. Part II of the survey classifies and presents the technical details of application-specific and architectural approximation techniques, which both target the design of resource-efficient processors/accelerators and systems. Moreover, it reports a quantitative analysis of the techniques and a detailed analysis of the application spectrum of Approximate Computing, and finally, it discusses open challenges and future directions.

Genomic exploration of the journey of Plasmodium vivax in Latin America.

Plasmodium vivax is the predominant malaria parasite in Latin America. Its colonization history in the region is rich and complex, and is still highly debated, especially about its origin(s). Our study employed cutting-edge population genomic techniques to analyze whole genome variation from 620 P. vivax isolates, including 107 newly sequenced samples from West Africa, Middle East, and Latin America. This sampling represents nearly all potential source populations worldwide currently available. Analyses of the genetic structure, diversity, ancestry, coalescent-based inferences, including demographic scenario testing using Approximate Bayesian Computation, have revealed a more complex evolutionary history than previously envisioned. Indeed, our analyses suggest that the current American P. vivax populations predominantly stemmed from a now-extinct European lineage, with the potential contribution also from unsampled populations, most likely of West African origin. We also found evidence that P. vivax arrived in Latin America in multiple waves, initially during early European contact and later through post-colonial human migration waves in the late 19th-century. This study provides a fresh perspective on P. vivax's intricate evolutionary journey and brings insights into the possible contribution of West African P. vivax populations to the colonization history of Latin America.

Estimating evolutionary and demographic parameters via ARG-derived IBD.

Inference of evolutionary and demographic parameters from a sample of genome sequences often proceeds by first inferring identical-by-descent (IBD) genome segments. By exploiting efficient data encoding based on the ancestral recombination graph (ARG), we obtain three major advantages over current approaches: (i) no need to impose a length threshold on IBD segments, (ii) IBD can be defined without the hard-to-verify requirement of no recombination, and (iii) computation time can be reduced with little loss of statistical efficiency using only the IBD segments from a set of sequence pairs that scales linearly with sample size. We first demonstrate powerful inferences when true IBD information is available from simulated data. For IBD inferred from real data, we propose an approximate Bayesian computation inference algorithm and use it to show that even poorly-inferred short IBD segments can improve estimation. Our mutation-rate estimator achieves precision similar to a previously-published method despite a 4 000-fold reduction in data used for inference, and we identify significant differences between human populations. Computational cost limits model complexity in our approach, but we are able to incorporate unknown nuisance parameters and model misspecification, still finding improved parameter inference.

Optimizing contact tracing for avian influenza in poultry flocks.

Contact tracing is commonly used to manage infectious diseases of both humans and animals. It aims to detect early and control potentially infected individuals or farms that had contact with infectious cases. Because it is very resource-intensive, contact tracing is usually performed on a pre-defined time window, based on previous knowledge of the duration of the incubation period. However, pre-defined time windows may not be always relevant, reducing the efficiency of contact tracing. In this study, we estimated the day when farms were first infected with highly pathogenic avian influenza viruses, a devastating pathogen causing severe socio-economic damage in domestic poultry. The estimation was performed by fitting a stochastic mechanistic model to observed daily mortality data from 63 infected poultry farms in France and The Netherlands, using approximate Bayesian computation. Independent of the poultry species or country, the estimates of the time of first infection ranged between 3.4 (95% credible interval-CrI: 2.6, 4.6) and 19.9 (95% CrI: 11.9, 31.3) days prior to the last observation. We developed an online application to provide real-time support to policymakers by estimating realistic ranges of dates of first infection to inform contact tracing and improve its efficiency.

Stochastic Calibration of Automated Vehicle Car-Following Control: An Approximate Bayesian Computation Approach

Stochastic Calibration of Automated Vehicle Car-Following Control: An Approximate Bayesian Computation Approach

A novel class of zipper fractal Bézier curves and its graphics applications

In this article, we present a novel generalization of a Bézier curve that can be non-smooth. We name it the zipper fractal Bézier curve. This new curve is constructed using a class of zipper α-fractal polynomials corresponding to Bernstein basis polynomials. We establish sufficient conditions on the parameters to ensure these zipper α-fractal polynomials exhibit properties such as non-negativity, partition of unity, linear precision, and symmetry. Utilizing these properties, we show that the zipper fractal Bézier curve can achieve endpoint interpolation, symmetry, the convex-hull property, and geometric invariance. Using the zipper fractal Bézier curves, we create attractive deltoid-like, astroid-like, exoid-like, and six-leaf flower designs. Our findings have applications in various fields, including approximation theory, CAGD, computer graphics, art, and design.

Stratified distance space improves the efficiency of sequential samplers for approximate Bayesian computation

Stratified distance space improves the efficiency of sequential samplers for approximate Bayesian computation

A Low-Power High-Accuracy Approximate Multiplier Using High-Order Approximate Compressors

To address the need to reduce power consumption, approximate multipliers have emerged as a potential solution for fault-tolerant applications. In this work, we present a new 8x8 approximate multiplier that focuses on minimizing performance while maintaining a high degree of accuracy. The design features two key features: firstly, based on their importance, different weights are handled by the compressors with different levels of precision, allowing for a trade-off between energy efficiency and minimum error. Second, higher order approximation compressors such as 8:2 compressors are used for intermediate weights to simplify the drive chain logic. This is, to our knowledge, the first design to successfully integrate higher-order approximate compressors into an approximate multiplier. Compared to a precision multiplier such as the Dadda tree multiplier, experimental results show that the proposed design offers significant energy savings while maintaining high accuracy. Key Words:- Approximate computing; Arithmetic circuits; Logic design; Low-power design; Partial Product reduction

Fitness landscapes of human microsatellites.

Advances in DNA sequencing technology and computation now enable genome-wide scans for natural selection to be conducted on unprecedented scales. By examining patterns of sequence variation among individuals, biologists are identifying genes and variants that affect fitness. Despite this progress, most population genetic methods for characterizing selection assume that variants mutate in a simple manner and at a low rate. Because these assumptions are violated by repetitive sequences, selection remains uncharacterized for an appreciable percentage of the genome. To meet this challenge, we focus on microsatellites, repetitive variants that mutate orders of magnitude faster than single nucleotide variants, can harbor substantial variation, and are known to influence biological function in some cases. We introduce four general models of natural selection that are each characterized by just two parameters, are easily simulated, and are specifically designed for microsatellites. Using a random forests approach to approximate Bayesian computation, we fit these models to carefully chosen microsatellites genotyped in 200 humans from a diverse collection of eight populations. Altogether, we reconstruct detailed fitness landscapes for 43 microsatellites we classify as targets of selection. Microsatellite fitness surfaces are diverse, including a range of selection strengths, contributions from dominance, and variation in the number and size of optimal alleles. Microsatellites that are subject to selection include loci known to cause trinucleotide expansion disorders and modulate gene expression, as well as intergenic loci with no obvious function. The heterogeneity in fitness landscapes we report suggests that genome-scale analyses like those used to assess selection targeting single nucleotide variants run the risk of oversimplifying the evolutionary dynamics of microsatellites. Moreover, our fitness landscapes provide a valuable visualization of the selective dynamics navigated by microsatellites.

Impact-force identification using deep learning and Bayesian inference with application on pipeline structures

Structures like bridges and pipelines are vulnerable to impacts from various sources, such as falling debris, over-height vehicles, or floating objects, which can compromise their integrity. Traditional inspection methods are costly and complex and can lead to economic losses due to shutdowns. This research introduces a probabilistic and more efficient approach by combining deep learning and Bayesian inference techniques to accurately measure and analyze these impacts, aiming to enhance the longevity and safety of these structures while mitigating potential critical failures. In this methodology, a novel two-stage approach is implemented to resolve the inverse problem of identifying impact forces on structures. Initially, a convolutional neural network (CNN) classifier for each of the four sensors is employed to determine the impacting material (aluminum, rubber, plastic). This classification stage utilizes ground truth data to identify the nature of the impact accurately. Following this, the preclassified data, categorized by the actual impacting materials, is directed into one of three 5-layer artificial neural networks (ANNs), each designated for a different impacting material. The ANNs act as surrogate models for Bayesian inference to determine impact force and position, resulting in 12 specialized models, each corresponding to a specific combination of sensor and tip type. The ANNs were evaluated using mean squared error, showing precise predictions of the pipeline’s acceleration frequency signals, while the CNN classifier achieved over 99% F1 scores. Using ANN and CNN results, the approximate Bayesian computation with subset simulation technique showed over 90% precision with 7% uncertainty in inferring impact force and 92% precision with 2% uncertainty in determining the impact position along the pipe. Data fusion, combining sensor responses, further improved precision and reduced uncertainty, achieving over 92% precision with 8% uncertainty for impact force and over 98% precision with 1.8% uncertainty for depth. These results demonstrate the method’s reliability and effectiveness in accurately identifying impact forces and locations, even with a single distant sensor.

A Survey on Approximate Computing and Storage Techniques Applied to Video Coding Systems

A Survey on Approximate Computing and Storage Techniques Applied to Video Coding Systems

Analysis of Block Adaptive Type-II Progressive Hybrid Censoring with Weibull Distribution

The estimation of unknown model parameters and reliability characteristics is considered under a block adaptive progressive hybrid censoring scheme, where data are observed from a Weibull model. This censoring scheme enhances experimental efficiency by conducting experiments across different testing facilities. Point and interval estimates for parameters and reliability assessments are derived using both classical and Bayesian approaches. The existence and uniqueness of maximum likelihood estimates are established. Consequently, reliability performance and differences across different testing facilities are analyzed. In addition, a Metropolis–Hastings sampling algorithm is developed to approximate complex posterior computations. Approximate confidence intervals and highest posterior density credible intervals are obtained for the parametric functions. The performance of all estimators is evaluated through an extensive simulation study, and observations are discussed. A cancer dataset is analyzed to illustrate the findings under the block adaptive censoring scheme.

Recent Range Expansion and Genomic Admixture in a Kleptoparasitic Spider, Argyrodes lanyuensis: A Case of Adaptive Introgression on Small, Isolated Islands of the Taiwan-Philippine Transition Zone?

Adaptive introgression involves the acquisition of advantageous genetic variants through hybridisation, which are subsequently favoured by natural selection due to their association with beneficial traits. Here, we analysed speciation patterns of the kleptoparasitic spider, Argyrodes lanyuensis, through genomic analyses and tested for possible genetic evidence of adaptive introgression at the Taiwan-Philippines transition zone. Our study used highly polymorphic SNPs to demonstrate that speciation occurred when the Hualien (on Taiwan Island + Green Island) and Orchid Island + Philippine lineages separated during the early to mid-Pleistocene. The best colonisation model suggested by approximate Bayesian computation and random forests and biogeographical analyses supported an inference of a bottleneck during speciation, an interpretation reinforced by observation of lower FST values and reduced genetic diversity of the Orchid Island + Philippines lineage. We also found the highest support for the occurrence of introgression on the youngest island (Green Island) of the Taiwan-Philippines transition zone based on the ABBA-BABA test. Our study highlights the inference of two noteworthy species (Hualien + Green Island and Orchid Island + Philippines) based on our species delimitation tests, with gene flow between Green Island and Orchid Island that indicates introgression. The potential adaptive alleles in Green Island population, which are under balancing selection, provide initial evidence of possible rare case of adaptive introgression. This could represent an evolutionary response to a newly formed niche (or novel geographical context) lying between the tropical climate of the Philippines and the subtropical climate of Hualien, Taiwan.

FPGA‐Based Resource‐Optimal Approximate Multiplier for Error‐Resilient Applications

ABSTRACTArithmetic units inspired by approximate computations have seen a significant development in error‐resilient applications, wherein accuracy can be traded off for enhanced performance. Most of the existing literature pertaining to approximate computations targets ASIC platforms. In this paper, we focus on exploiting the features of approximate computation to design efficient digital hardware for FPGA platforms. Specifically, we propose an FPGA implementation of an approximate multiplier unit based on the CORDIC algorithm. Contemporary FPGA‐based approximate multiplier implementations report a lot of compromise in accuracy and a relatively higher implementation cost in terms of utilized resources, timing, and energy. We conduct a detailed Pareto analysis to determine the number of optimal computing stages for the proposed CORDIC‐based approximate multiplier that justifies the accuracy‐performance trade‐offs. More importantly, we focus on the optimal logic distribution of the proposed multiplier circuit by restructuring the top‐level Boolean network and translating it into a circuit netlist that can be efficiently mapped onto the inherent FPGA fabric of LUTs and Carry4 primitives. Our CORDIC‐based implementations significantly improve the accuracy metrics while maintaining a suitable performance trade‐off. The efficacy of our proposed multiplier is tested using two image‐processing applications, namely, image blending and image smoothening. The obtained results show a substantial improvement over the existing state‐of‐the‐art approximate multipliers.

Calibration verification for stochastic agent-based disease spread models.

Accurate disease spread modeling is crucial for identifying the severity of outbreaks and planning effective mitigation efforts. To be reliable when applied to new outbreaks, model calibration techniques must be robust. However, current methods frequently forgo calibration verification (a stand-alone process evaluating the calibration procedure) and instead use overall model validation (a process comparing calibrated model results to data) to check calibration processes, which may conceal errors in calibration. In this work, we develop a stochastic agent-based disease spread model to act as a testing environment as we test two calibration methods using simulation-based calibration, which is a synthetic data calibration verification method. The first calibration method is a Bayesian inference approach using an empirically-constructed likelihood and Markov chain Monte Carlo (MCMC) sampling, while the second method is a likelihood-free approach using approximate Bayesian computation (ABC). Simulation-based calibration suggests that there are challenges with the empirical likelihood calculation used in the first calibration method in this context. These issues are alleviated in the ABC approach. Despite these challenges, we note that the first calibration method performs well in a synthetic data model validation test similar to those common in disease spread modeling literature. We conclude that stand-alone calibration verification using synthetic data may benefit epidemiological researchers in identifying model calibration challenges that may be difficult to identify with other commonly used model validation techniques.

Inferring Long-Term and Short-Term Determinants of Genetic Diversity in Honey Bees: Beekeeping Impact and Conservation Strategies.

Bees are vital pollinators in natural and agricultural landscapes around the globe, playing a key role in maintaining flowering plant biodiversity and ensuring food security. Among the honey bee species, the Western honey bee (Apis mellifera) is particularly significant, not only for its extensive crop pollination services but also for producing economically valuable products such as honey. Here, we analyzed whole-genome sequence data from four Apis species to explore how honey bee evolution has shaped current diversity patterns. Using Approximate Bayesian Computation, we first reconstructed the demographic history of A. mellifera in Europe, finding support for postglacial secondary contacts, therefore predating human-mediated transfers linked to modern beekeeping. However, our analysis of recent demographic changes reveals significant bottlenecks due to beekeeping practices, which have notably affected genetic diversity. Black honey bee populations from conservatories, particularly those on islands, exhibit considerable genetic loss, highlighting the need to evaluate the long-term effectiveness of current conservation strategies. Additionally, we observed a high degree of conservation in the genomic landscapes of nucleotide diversity across the four species, despite a divergence gradient spanning over 15 million years, consistent with a long-term conservation of the recombination landscapes. Taken together, our results provide the most comprehensive assessment of diversity patterns in honey bees to date and offer insights into the optimal management of resources to ensure the long-term persistence of honey bees and their invaluable pollination services.

Breaking Uncertainty Barriers: Approximate Bayesian Computation Advances in Rainfall–Runoff Modeling

Breaking Uncertainty Barriers: Approximate Bayesian Computation Advances in Rainfall–Runoff Modeling

Optimization of Delay and Area for Approximate Radix-8 Booth Multiplier

This paper investigates the optimization of Radix-8 Booth Multipliers, which are essential for efficient arithmetic operations in modern digital systems, particularly in applications such as digital signal processing, telecommunications, and image processing where rapid and accurate calculations are crucial. The study aims to enhance performance by focusing on reducing both delay and area while ensuring that acceptable accuracy levels are maintained for error-tolerant applications. To achieve these optimization goals, we compare three methodologies: the Carry Save Adder (CSA), the Kogge Stone Adder (KSA), and the Carry Look-Ahead Adder (CLA), each of which presents unique advantages and trade-offs in terms of speed, area utilization, and power consumption. The results of our analysis demonstrate that the Kogge Stone Adder provides the best overall performance in terms of speed and area efficiency, making it the most suitable choice for optimizing Radix-8 Booth Multipliers, particularly in scenarios where high performance and efficient resource use are critical. By emphasizing these findings, this study contributes valuable insights into the design of more efficient multipliers that can meet the increasing demands of contemporary digital applications. Key Words: Radix-8 Booth Multiplier, approximate computing, delay optimization, area optimization, Carry Save Adder (CSA), Kogge Stone Adder (KSA), Carry Look-Ahead Adder (CLA), digital systems, arithmetic operations, performance enhancement, error-tolerant applications, partial product generation, hardware description languages (HDLs), digital signal processing, machine learning, approximation techniques, simulation and testing, power consumption, ASIC design.

Approximate Bayesian computation supports a high incidence of chromosomal mosaicism in blastocyst-stage human embryos.

Chromosome mis-segregation is common in human meiosis and mitosis, and the resulting aneuploidies are the leading cause of pregnancy loss. Preimplantation genetic testing for aneuploidy (PGT-A) seeks to prioritize chromosomally normal embryos for transfer based on genetic analysis of a biopsy of approximately five trophectoderm cells from blastocyst-stage in vitro fertilized (IVF) embryos. While modern PGT-A platforms classify these biopsies as aneuploid, euploid, or mosaic (possessing a mixture of normal and aneuploid cells), the underlying incidences of aneuploid, euploid, and mosaic embryos and the rates of meiotic and mitotic error that produced them remain largely unknown. To address this knowledge gap, we paired a recent method for embryo simulation with approximate Bayesian computation (ABC) to infer rates of meiotic and mitotic error that best explain published PGT-A data. By simulating from these posterior distributions, we also evaluated the chromosomal status of entire embryos. For a published clinical sample, we estimated a 39-43% probability of meiotic error per meiosis, as well as a 1.0-3.0% probability of mitotic error per mitosis, depending on assumptions about spatial clustering of aneuploid cells within mosaic embryos. In addition, our analyses suggest that less than 1% of blastocysts are fully euploid, and that many embryos possess low-level mosaic clones that are not captured during biopsy. These broad conclusions were relatively insensitive to potential misclassification of mosaic biopsies. Together, our work helps overcome the limitations of embryo biopsies to estimate the fundamental rates of cell division errors that are the main causes of human pregnancy loss.