The rapid growth of machine learning (ML) technologies has raised significant concerns about their environmental impact, particularly regarding energy consumption and carbon emissions. This comprehensive review examines the intersection of ML and sustainability, synthesizing research from 2014 to 2024 to provide a holistic view of sustainable ML practices. This systematic review, encompassing over 200 peer-reviewed publications, reveals a growing emphasis on quantifying and mitigating the environmental footprint of ML systems. Key findings include: (1) a 300% increase in sustainable ML research since 2020; (2) the emergence of specialized carbon footprint quantification tools for ML; and (3) promising advancements in energy-efficient algorithms and green computing infrastructure. This research identifies critical challenges, including the lack of standardized sustainability metrics and the need for more robust life-cycle assessments of ML systems. The review also highlights the potential of transfer learning, federated learning, and hardware innovations in reducing ML's environmental impact. The analysis culminates in a novel framework for implementing sustainable practices in ML projects and a detailed roadmap for future research. This work provides researchers, practitioners, and policymakers with crucial insights to drive the development of more environmentally responsible ML technologies, ultimately contributing to global sustainability goals.

The demand for indoor localization services in the Internet of Things (IoT) has been increasing dramatically during the last decade. Many indoor localization systems adopt Wi-Fi fingerprinting with received signal strength indicators (RSSIs) as a source of sensors to localize an object because it is cost effective and can give high accuracy. However, the fluctuation of wireless signals resulting from environmental uncertainties leads to considerable variations in RSSIs, which poses a challenge to accurate localization on a single floor, not to mention multifloor or even 3-D localization. Most existing multifloor methods employ a sequential approach where a different algorithm is tailored for each step in the sequence to determine the floor and then the location of an object. In this article, we formulate the indoor localization problem as a Markov decision process rather than a typical classification or regression problem. A deep reinforcement learning method is used to bisect the search space in a hierarchy from the entire building down to a prespecified distance scale to the object position. This approach significantly reduces the time complexity of the searching fromO(N3) toO(logN), where N indicates the localization resolution. The proposed method tackles environmental dynamics with Wi-Fi fingerprinting for 3-D continuous space. The experimental results demonstrate the high accuracy, efficiency, and robustness of the proposed approach.

The location-based services for Internet of Things (IoTs) have attracted extensive research effort during the last decades. Wi-Fi fingerprinting with received signal strength indicator (RSSI) has been widely adopted in vast indoor localization systems due to its relatively low cost and the potency for high accuracy. However, the fluctuation of wireless signal resulting from environment uncertainties leads to considerable variations on RSSIs, which poses grand challenges to the fingerprint-based indoor localization regarding positioning accuracy. In this paper, we propose a top-down searching method using a deep reinforcement learning agent to tackle environment dynamics in indoor positioning with Wi-Fi fingerprints. Our model learns an action policy that is capable to localize 75% of the targets in an area of 25000m2within 0.55m.

Network motif detection is the search for statistically overrepresented subgraphs present in a larger target network. They are thought to represent key structure and control mechanisms. Although the problem is exponential in nature, several algorithms and tools have been developed for efficiently detecting network motifs. This work analyzes 11 network motif detection tools and algorithms. Detailed comparisons and insightful directions for using these tools and algorithms are discussed. Key aspects of network motif detection are investigated. Network motif types and common network motifs as well as their biological functions are discussed. Applications of network motifs are also presented. Finally, the challenges, future improvements and future research directions for network motif detection are also discussed.

UNLABELLED: LASAGNA-Search 2.0 is an integrated webtool for transcription factor (TF) binding site search and visualization. The tool is based on the LASAGNA (Length-Aware Site Alignment Guided by Nucleotide Association) algorithm. It eliminates manual TF model collection and promoter sequence retrieval. Search results can be visualized locally or in the University of California Santa Cruz Genome Browser. Gene regulatory network inference based on the search results offers another way of visualization. A list of TFs and target genes is all a user needs to start using the tool. LASAGNA-Search 2.0 currently offers 1792 TF models and supports 15 species for automatic promoter retrieval and visualization in the University of California Santa Cruz Genome Browser. It is a user-friendly tool designed for non-bioinformaticians and is suitable for research and teaching. We describe important changes made since the initial release. AVAILABILITY AND IMPLEMENTATION: LASAGNA-Search 2.0 is freely available without registration at http://biogrid.engr.uconn.edu/lasagna_search/.

BACKGROUND: Scientists routinely scan DNA sequences for transcription factor (TF) binding sites (TFBSs). Most of the available tools rely on position-specific scoring matrices (PSSMs) constructed from aligned binding sites. Because of the resolutions of assays used to obtain TFBSs, databases such as TRANSFAC, ORegAnno and PAZAR store unaligned variable-length DNA segments containing binding sites of a TF. These DNA segments need to be aligned to build a PSSM. While the TRANSFAC database provides scoring matrices for TFs, nearly 78% of the TFs in the public release do not have matrices available. As work on TFBS alignment algorithms has been limited, it is highly desirable to have an alignment algorithm tailored to TFBSs. RESULTS: We designed a novel algorithm named LASAGNA, which is aware of the lengths of input TFBSs and utilizes position dependence. Results on 189 TFs of 5 species in the TRANSFAC database showed that our method significantly outperformed ClustalW2 and MEME. We further compared a PSSM method dependent on LASAGNA to an alignment-free TFBS search method. Results on 89 TFs whose binding sites can be located in genomes showed that our method is significantly more precise at fixed recall rates. Finally, we described LASAGNA-ChIP, a more sophisticated version for ChIP (Chromatin immunoprecipitation) experiments. Under the one-per-sequence model, it showed comparable performance with MEME in discovering motifs in ChIP-seq peak sequences. CONCLUSIONS: We conclude that the LASAGNA algorithm is simple and effective in aligning variable-length binding sites. It has been integrated into a user-friendly webtool for TFBS search and visualization called LASAGNA-Search. The tool currently stores precomputed PSSM models for 189 TFs and 133 TFs built from TFBSs in the TRANSFAC Public database (release 7.0) and the ORegAnno database (08Nov10 dump), respectively. The webtool is available at http://biogrid.engr.uconn.edu/lasagna_search/.

Network motifs are statistically overrepresented sub-structures (sub-graphs) in a network, and have been recognized as 'the simple building blocks of complex networks'. Study of biological network motifs may reveal answers to many important biological questions. The main difficulty in detecting larger network motifs in biological networks lies in the facts that the number of possible sub-graphs increases exponentially with the network or motif size (node counts, in general), and that no known polynomial-time algorithm exists in deciding if two graphs are topologically equivalent. This article discusses the biological significance of network motifs, the motivation behind solving the motif-finding problem, and strategies to solve the various aspects of this problem. A simple classification scheme is designed to analyze the strengths and weaknesses of several existing algorithms. Experimental results derived from a few comparative studies in the literature are discussed, with conclusions that lead to future research directions.

BACKGROUND: Computational approaches to transcription factor binding site identification have been actively researched in the past decade. Learning from known binding sites, new binding sites of a transcription factor in unannotated sequences can be identified. A number of search methods have been introduced over the years. However, one can rarely find one single method that performs the best on all the transcription factors. Instead, to identify the best method for a particular transcription factor, one usually has to compare a handful of methods. Hence, it is highly desirable for a method to perform automatic optimization for individual transcription factors. RESULTS: We proposed to search for transcription factor binding sites in vector spaces. This framework allows us to identify the best method for each individual transcription factor. We further introduced two novel methods, the negative-to-positive vector (NPV) and optimal discriminating vector (ODV) methods, to construct query vectors to search for binding sites in vector spaces. Extensive cross-validation experiments showed that the proposed methods significantly outperformed the ungapped likelihood under positional background method, a state-of-the-art method, and the widely-used position-specific scoring matrix method. We further demonstrated that motif subtypes of a TF can be readily identified in this framework and two variants called the k NPV and k ODV methods benefited significantly from motif subtype identification. Finally, independent validation on ChIP-seq data showed that the ODV and NPV methods significantly outperformed the other compared methods. CONCLUSIONS: We conclude that the proposed framework is highly flexible. It enables the two novel methods to automatically identify a TF-specific subspace to search for binding sites. Implementations are available as source code at: http://biogrid.engr.uconn.edu/tfbs_search/.

Classification of mitochondrial DNA (mtDNA) into their respective haplogroups allows the addressing of various anthropologic and forensic issues. Unique to mtDNA is its abundance and non-recombining uni-parental mode of inheritance; consequently, mutations are the only changes observed in the genetic material. These individual mutations are classified into their cladistic haplogroups allowing the tracing of different genetic branch points in human (and other organisms) evolution. Due to the large number of samples, it becomes necessary to automate the classification process. Using 5-fold cross-validation, we investigated two classification techniques on the consented database of 21 141 samples published by the Genographic project. The support vector machines (SVM) algorithm achieved a macro-accuracy of 88.06% and micro-accuracy of 96.59%, while the random forest (RF) algorithm achieved a macro-accuracy of 87.35% and micro-accuracy of 96.19%. In addition to being faster and more memory-economic in making predictions, SVM and RF are better than or comparable to the nearest-neighbor method employed by the Genographic project in terms of prediction accuracy.

BACKGROUND: Handling genotype data typed at hundreds of thousands of loci is very time-consuming and it is no exception for population structure inference. Therefore, we propose to apply PCA to the genotype data of a population, select the significant principal components using the Tracy-Widom distribution, and assign the individuals to one or more subpopulations using generic clustering algorithms. RESULTS: We investigated K-means, soft K-means and spectral clustering and made comparison to STRUCTURE, a model-based algorithm specifically designed for population structure inference. Moreover, we investigated methods for predicting the number of subpopulations in a population. The results on four simulated datasets and two real datasets indicate that our approach performs comparably well to STRUCTURE. For the simulated datasets, STRUCTURE and soft K-means with BIC produced identical predictions on the number of subpopulations. We also showed that, for real dataset, BIC is a better index than likelihood in predicting the number of subpopulations. CONCLUSION: Our approach has the advantage of being fast and scalable, while STRUCTURE is very time-consuming because of the nature of MCMC in parameter estimation. Therefore, we suggest choosing the proper algorithm based on the application of population structure inference.

The 11 papers in this special section focus on biomedical computations on the grid.

BACKGROUND: DNA Microarray technology is an innovative methodology in experimental molecular biology, which has produced huge amounts of valuable data in the profile of gene expression. Many clustering algorithms have been proposed to analyze gene expression data, but little guidance is available to help choose among them. The evaluation of feasible and applicable clustering algorithms is becoming an important issue in today's bioinformatics research. RESULTS: In this paper we first experimentally study three major clustering algorithms: Hierarchical Clustering (HC), Self-Organizing Map (SOM), and Self Organizing Tree Algorithm (SOTA) using Yeast Saccharomyces cerevisiae gene expression data, and compare their performance. We then introduce Cluster Diff, a new data mining tool, to conduct the similarity analysis of clusters generated by different algorithms. The performance study shows that SOTA is more efficient than SOM while HC is the least efficient. The results of similarity analysis show that when given a target cluster, the Cluster Diff can efficiently determine the closest match from a set of clusters. Therefore, it is an effective approach for evaluating different clustering algorithms. CONCLUSION: HC methods allow a visual, convenient representation of genes. However, they are neither robust nor efficient. The SOM is more robust against noise. A disadvantage of SOM is that the number of clusters has to be fixed beforehand. The SOTA combines the advantages of both hierarchical and SOM clustering. It allows a visual representation of the clusters and their structure and is not sensitive to noises. The SOTA is also more flexible than the other two clustering methods. By using our data mining tool, Cluster Diff, it is possible to analyze the similarity of clusters generated by different algorithms and thereby enable comparisons of different clustering methods.

Grid computing can be characterized as a distributed infrastructure that is a collection of computing resources within or across locations that are aggregated to act as a unified processing resource. In some of the anticipated future grid applications, the same data will be transmitted to multiple sites. It is widely accepted that this can be, in theory, best achieved using reliable multicast protocols. This paper addresses the use of reliable multicast in grid computing, identifying applications and requirements, and discussing how these requirements are met by the main protocols being standardized by the Internet engineering task force (IETF). The emphasis of the study is on high-performance computing and communication, and on I-N reliable multicast using the NACK oriented reliable multicast (NORM) protocol family. The two existing implementations are evaluated and compared with TCP, in scenarios that require I-N transmission of data with a deadline. Unlike other work, our conclusions are backed by experimental evaluation of existing protocols in production environments. Therefore, the results shown portray the range of performance and cost that will be obtained if grid applications are to embrace reliable multicast. The evaluation shows that NORM protocols can greatly reduce the network overhead incurred in I-N transfers using TCP, but currently fail to satisfy the high-throughput requirements of the grid.

This paper describes a new parallel semantic network system, the TROJAN. Unlike the synchronous and static marker passing algorithm previously used for parallel semantic network design, the TROJAN operates asynchronously, supporting knowledge sharing, dynamic load balancing and duplicate checking. Current implementation of the TROJAN focuses on path-based knowledge inferences, using ANSI C and the MPICH-G2 with flex lexical analyzer and the yacc parser generator. The performance tests have been carried out on a local Grid, consisting of three heterogeneous systems: a SUN Cluster, an SGI Origin 3800, and a Dell Pentium cluster. The experiments demonstrate promising speedups.

This paper presents a low communication overhead parallel algorithm for pattern matching in biological sequences. Given such a sequence of length n and a pattern of length m, we conclude an algorithm with five computation/communication phases, each requiring O(n) computation time and only O(p) message units. The low communication overhead of the algorithm is essential to achieving reasonable speedups on clusters, where the interprocessor communication latency is usually higher Previous parallel implementations use straightforward domain decomposition based on existing sequential algorithms and rely on parallel machines with low-latency interconnection network and fast hardware support for processor synchronization.

This paper answers an open question in the design of complimentary metal-oxide semiconductor VLSI circuits. The question asks whether a polynomial-time algorithm can decide if a given planar graph has a plane embedding ${\cal E}$ such that ${\cal E}$ has an Euler trail P = e 1 e 2 ... e m and its dual graph has an Euler trail $P^*=e^*_1 e^*_2 \ldots e^*_m$, where $e^*_i$ is the dual edge of e i for i=1,2,...,m. This paper answers this question in the affirmative by presenting a linear-time algorithm.

The paper answers an open question in the design of complimentary metal-oxide semiconductor (CMOS) VLSI circuits. It asks whether a polynomial-time algorithm can decide if a given planar graph has a plane embedding /spl epsiv/ such that /spl epsiv/ has a Euler trail P=e/sub 1/e/sub 2/...e/sub m/ and its dual graph has a Euler trail P*=e/sub 1/*e/sub 2/*...e/sub m/* where e/sub i/* is the dual edge of e/sub i/ for i=1, 2, ..., m. The paper answers this question in the affirmative by presenting a linear-time algorithm.