Pruning Incremental Linear Model Trees with Approximate Lookahead

Identifying genetic variants associated with complex disease in high-dimensional data is a challenging problem, and complicated etiologies such as gene-gene interactions are often ignored in analyses. The data-mining method Random Forests (RF) can handle high-dimensions; however, in high-dimensional data, RF is not an effective filter for identifying risk factors associated with the disease trait via complex genetic models such as gene-gene interactions without strong marginal components. Here we propose an extension called Weighted Random Forests (wRF), which incorporates tree-level weights to emphasize more accurate trees in prediction and calculation of variable importance. We demonstrate through simulation and application to data from a genetic study of addiction that wRF can outperform RF in high-dimensional data, although the improvements are modest and limited to situations with effect sizes that are larger than what is realistic in genetics of complex disease. Thus, the current implementation of wRF is unlikely to improve detection of relevant predictors in high-dimensional genetic data, but may be applicable in other situations where larger effect sizes are anticipated.

Random forests (RFs) have been widely used as a powerful classification method. However, with the randomization in both bagging samples and feature selection, the trees in the forest tend to select uninformative features for node splitting. This makes RFs have poor accuracy when working with high-dimensional data. Besides that, RFs have bias in the feature selection process where multivalued features are favored. Aiming at debiasing feature selection in RFs, we propose a new RF algorithm, called xRF, to select good features in learning RFs for high-dimensional data. We first remove the uninformative features using p-value assessment, and the subset of unbiased features is then selected based on some statistical measures. This feature subset is then partitioned into two subsets. A feature weighting sampling technique is used to sample features from these two subsets for building trees. This approach enables one to generate more accurate trees, while allowing one to reduce dimensionality and the amount of data needed for learning RFs. An extensive set of experiments has been conducted on 47 high-dimensional real-world datasets including image datasets. The experimental results have shown that RFs with the proposed approach outperformed the existing random forests in increasing the accuracy and the AUC measures.

Subject of the research – visual object tracking using various image hashing algorithms for real-time tracking tasks. The goal of this study is to evaluate the tracking success and processing speed of existing and new hashing algorithms for object tracking and to identify the most suitable algorithms to be used under limited computational resources. The objectives of the research include: developing and implementing object tracking based on the aHash, dHash, pHash, mHash, LHash, and LDHash algorithms; comparing the processing speed and accuracy of these methods on the video sequences "OccludedFace2," "David," and "Sylvester"; determining the tracking success rate (TSR) and frames per second (FPS) metrics for each algorithm; analyzing the impact of the search window size, search strategy, and type of hashing on tracking quality, and providing recommendations for their use. The study also explores the trade-off between accuracy and processing speed for each algorithm considering the constraints of limited computational resources. The methods of this study involve testing and evaluating the accuracy and speed of image hashing algorithms on different test video sequences, as well as the use of metrics to determine object similarity using the Hamming distance. The results demonstrate that the aHash and mHash algorithms demonstrate the best accuracy indicators for all hash window sizes, aHash has a higher processing speed, and mHash offers better robustness to changes in lighting and object position. The dHash and pHash algorithms were less effective than the aHash and mHash algorithms due to their sensitivity to changes in scale and rotation. However, perceptual hashing-based methods, such as pHash, are more robust to contrast and blurring. Conclusions. The best hashing algorithms for real-time object-tracking tasks are aHash and mHash. This study underscores the significance of selecting suitable hashing algorithms and search strategies tailored to specific application scenarios and offers possibilities for further optimization.

Tropical forest biomass is a crucial component of global carbon emission estimations. However, calibration and validation of such estimates require accurate and effective methods to estimate in situ above‐ground biomass (AGB). Present methods rely on allometric models that are highly uncertain for large tropical trees. Terrestrial laser scanning (TLS) tree modelling has demonstrated to be more accurate than these models to infer forest AGB. Nevertheless, applying TLS methods on tropical large trees is still challenging. We propose a method to estimate AGB of large tropical trees by three‐dimensional (3D) tree modelling of TLS point clouds. Twenty‐nine plots were scanned with a TLS in three study sites (Peru, Indonesia and Guyana). We identified the largest tree per plot (mean diameter at breast height of 73.5 cm), extracted its point cloud and calculated its volume by 3D modelling its structure using quantitative structure models (QSM) and converted to AGB using species‐specific wood density. We also estimated AGB using pantropical and local allometric models. To assess the accuracy of our and allometric methods, we harvest the trees and took destructive measurements. AGB estimates by the TLS–QSM method showed the best agreement in comparison to destructive harvest measurements (28.37% coefficient of variation of root mean square error [CV‐RMSE] and concordance correlation coefficient [CCC] of 0.95), outperforming the pantropical allometric models tested (35.6%–54.95% CV‐RMSE and CCC of 0.89–0.73). TLS–QSM showed also the lowest bias (overall underestimation of 3.7%) and stability across tree size range, contrasting with the allometric models that showed a systematic bias (overall underestimation ranging 15.2%–35.7%) increasing linearly with tree size. The TLS–QSM method also provided accurate tree wood volume estimates (CV RMSE of 23.7%) with no systematic bias regardless the tree structural characteristics. Our TLS–QSM method accounts for individual tree biophysical structure more effectively than allometric models, providing more accurate and less biased AGB estimates for large tropical trees, independently of their morphology. This non‐destructive method can be further used for testing and calibrating new allometric models, reducing the current under‐representation of large trees in and enhancing present and past estimates of forest biomass and carbon emissions from tropical forests.

Unmanned aerial vehicle (UAV)-based remote sensing (RS) possesses the significant advantage of being able to efficiently collect images for precision agricultural applications. Although numerous methods have been proposed to monitor crop nitrogen (N) status in recent decades, just how to utilize an appropriate modeling algorithm to estimate crop leaf N content (LNC) remains poorly understood, especially based on UAV multispectral imagery. A comparative assessment of different modeling algorithms (i.e., simple and non-parametric modeling algorithms alongside the physical model retrieval method) for winter wheat LNC estimation is presented in this study. Experiments were conducted over two consecutive years and involved different winter wheat varieties, N rates, and planting densities. A five-band multispectral camera (i.e., 490 nm, 550 nm, 671 nm, 700 nm, and 800 nm) was mounted on a UAV to acquire canopy images across five critical growth stages. The results of this study showed that the best-performing vegetation index (VI) was the modified renormalized difference VI (RDVI), which had a determination coefficient (R2) of 0.73 and a root mean square error (RMSE) of 0.38. This method was also characterized by a high processing speed (0.03 s) for model calibration and validation. Among the 13 non-parametric modeling algorithms evaluated here, the random forest (RF) approach performed best, characterized by R2 and RMSE values of 0.79 and 0.33, respectively. This method also had the advantage of full optical spectrum utilization and enabled flexible, non-linear fitting with a fast processing speed (2.3 s). Compared to the other two methods assessed here, the use of a look up table (LUT)-based radiative transfer model (RTM) remained challenging with regard to LNC estimation because of low prediction accuracy (i.e., an R2 value of 0.62 and an RMSE value of 0.46) and slow processing speed. The RF approach is a fast and accurate technique for N estimation based on UAV multispectral imagery.

Since 2020, the SARS-CoV-2 virus has infected billions of people and spread to 185 countries. The virus spreads by making new copies of its genome inside human cells and exploits the cells’ machinery to synthesise viral proteins it needs to infect further cells. Each time the virus copies its genetic material there’s a chance that the replication process introduces an error to the genetic sequence. Over time, these mutations accumulate which can give rise to new variants with different properties. These new variants, originating from a common ancestor, may spread faster or be able to evade immune systems that have learnt to recognise previous variants. To understand where new variants of SARS-CoV-2 come from and how related they are to each other, scientists build family trees called ‘phylogenetic trees’ based on similarities in the genetic sequences of different variants of the virus. Looking at these trees researchers can track how a variant spreads geographically, and also attempt to identify new worrying variants that might lead to a new wave of infections. The scale of the COVID-19 pandemic together with the global effort by clinicians and researchers to sequence SARS-CoV-2 genetic material means a library of over 13 million SARS-CoV-2 genomes now exists, making it the largest such collection for any organism. Although phylogenetic trees of viruses have been studied for a long time, exploring the SARS-CoV-2 library presents technical and practical challenges due to its sheer size. Sanderson has developed an open-source web tool called Taxonium that allows users to explore phylogenetic trees with millions of sequences. With help from collaborators at the University of California, Santa Cruz, Sanderson built a website called Cov2Tree, that uses the Taxonium platform to allow immediate access to an expansive tree of all publicly available SARS-CoV-2 sequences. Cov2Tree enables users to visualise all SARS-CoV-2 genomes in a birds-eye view akin to a ‘Google Earth for virus sequences’ where anyone can zoom in on a related family of viruses down to the level of individual sequences. This can be used to compare variants and follow geographic spread. Using Taxonium, scientists can explore how virus sequences are related to each other. They can also see the individual mutations that have occurred at each branch of the tree, and can search for sequences based on mutation, geographical location, or other factors. Interestingly, a trend appearing in the SARS-CoV-2 phylogenetic tree is the emergence of identical mutations at different branches of the tree without a common origin. These mutations may be a result of convergent evolution, a phenomenon that occurs when a mutation appears independently in different variants as it confers an advantage to the virus making such mutations more likely to persist. This means that scientists may be able to expect certain mutations to appear in more distantly related variants if they have appeared independently in several different variants already. Overall, Taxonium is an important tool for monitoring SARS-CoV-2 genomes, but it also has broader applications. The tool can be used to browse phylogenetic trees of other viruses and organisms. Furthermore, the Taxonium website offers a way to browse a tree of life, with images and links to Wikipedia. The SARS-CoV-2 library might be the largest now, but in the future even bigger datasets will likely be available, highlighting the importance of tools like Taxonium.

System for the analysis and visualization of large 3D anatomical trees

Taking a byte out of big data

Non-invasive real-time prediction of inner knee temperatures during therapeutic cooling

FPGA-oriented HW/SW implementation of ECG beat detection and classification algorithm

In this paper, the implementation of a new pupil detection system based on artificial intelligence techniques suitable for real-time and real-word applications is presented. The proposed AI-based pupil detection system uses a classifier implemented with slim-type neural networks, with its classes being defined according to the possible positions of the pupil within the eye image. In order to reduce the complexity of the neural network, a new parallel architecture is used in which two independent classifiers deliver the pupil center coordinates. The training, testing, and validation of the proposed system were performed using almost 40,000 eye images with a resolution of 320 × 240 pixels and coming from 20 different databases, with a high degree of generality. The experimental results show a detection rate of 96.29% at five pixels with a standard deviation of 3.38 pixels for all eye images from all databases and a processing speed of 100 frames/s. These results indicate both high accuracy and high processing speed, and they allow us to use the proposed solution for different real-time applications in variable and non-uniform lighting conditions, in fields such as assistive technology to communicate with neuromotor-disabled patients by using eye typing, in computer gaming, and in the automotive industry for increasing traffic safety by monitoring the driver’s cognitive state.

Development of the CCD sensor calls for a rapid star image processing speed and the basic work is to extract star points correctly. The typical histogram of a star image has a two-peak value structure and majority of the pixels belong to the background. In this paper, the noise of the image is removed with the adaptive Wiener filter. The background and the star points are distinguished using a window transform on the pixel gray value, all pixels whose gray value below the threshold are set to zero, where the threshold is determined with the optimal iterative algorithm or the Otsu algorithm. We locate the star points with an improved cross projection algorithm. Firstly, preprocess the star image into a binary image with the threshold determined above. Then calculate the horizontal and vertical projections of the binary image. In each projection direction, extract the domain bound of the star points with the proposed binarization differential extremum method which is different from the traditional marker method. Finally, the precise positions of the star points are calculated with the weighted centroid algorithm. Experiments on simulated and real star images show that the improved algorithm has a high processing speed (~300ms) and good centroid locating accuracy.

This paper uses joint algorithm and architecture design to enable high coding efficiency in conjunction with high processing speed and low area cost. Specifically, it presents several optimizations that can be performed on Context Adaptive Binary Arithmetic Coding (CABAC), a form of entropy coding used in H.264/AVC, to achieve the throughput necessary for real-time low power high definition video coding. The combination of syntax element partitions and interleaved entropy slices, referred to as Massively Parallel CABAC, increases the number of binary symbols that can be processed in a cycle. Subinterval reordering is used to reduce the cycle time required to process each binary symbol. Under common conditions using the JM12.0 software, the Massively Parallel CABAC, increases the bins per cycle by 2.7 to 32.8× at a cost of 0.25 to 6.84% coding loss compared with sequential single slice H.264/AVC CABAC. It also provides a 2× reduction in area cost, and reduces memory bandwidth. Subinterval reordering reduces the critical path delay by 14 to 22%, while modifications to context selection reduces the memory requirement by 67%. This work demonstrates that accounting for implementation cost during video coding algorithms design can enable higher processing speed and reduce hardware cost, while still delivering high coding efficiency in the next generation video coding standard.

In general, an object detector performs a series of processes finding regions estimated as objects in the image and specifying them as a predefined class. However, in the case of detecting only a single class such as a pedestrian detector or a license plate detector, it may not classify the estimated regions. Based on this idea, we developed a fast and lightweight single-class object detector with high processing speed while minimizing the loss of accuracy by excluding the class prediction layer from the YOLOX network and training the regression layer to respond only to the target class. As can be seen from the experimental results, the proposed method shows similar precision while having a faster inference speed compared to the original YOLOX model. It can be applied to mobile applications that lack resources but require a fast processing speed.

Stress relaxation behavior of Fe–Co–Si–B–Nb metallic glassy alloys in their supercooled-liquid state