FPGA Accelerator for Gradient Boosting Decision Trees

Depth-of-interaction encoding techniques for pixelated PET detectors enabled by machine learning methods and fast waveform digitization

Objective. Pixelated detectors with single-ended readout are routinely used by commercial positron emission tomography scanners owing to their good energy and timing resolution and optimized manufacturing, but they typically do not provide depth-of-interaction (DOI) information, which can help improve the performance of systems with higher resolution and smaller ring diameter. This work aims to develop a technique for multi-level DOI classification that does not require modifications to the detector designs.Approach. We leveraged high-speed (5 Gs s-1) waveform sampling electronics with the Domino Ring Sampler (DRS4) and machine learning (ML) methods to extract DOI information from the entire scintillation waveforms of pixelated crystals. We evaluated different grouping schemes for multi-level DOI classification by analyzing the DOI positioning profile and DOI positioning error. We examined trade-offs among crystal configurations, detector timing performance, and DOI classification accuracy. We also investigated the impact of different ML algorithms and input features-extracted from scintillation waveforms-on model accuracy.Main results. The DOI positioning profile and positioning error suggest that 2- or 3-level binning was effective for 20 mm long crystals. 2-level discrete DOI models achieved 95% class-wise accuracy and 83% overall accuracy in positioning events into the correct DOI level and 3-level up to 90% class-wise accuracy for long and narrow crystals (2 × 2 × 20 mm3). Long short-term memory networks trained with time-frequency moments were twice as efficient in training time while maintaining equal or better accuracy compared to those trained with waveforms. Classical ML algorithms exhibit comparable accuracy while consuming one order less training time than deep learning models.Significance. This work demonstrates a proof-of-concept approach for obtaining DOI information from commercially available pixelated detectors without altering the detector design thereby avoiding potential degradation in detector timing performance. It provides an alternative solution for multi-level DOI classification, potentially inspiring future scanner designs.

Improving the Timing Resolution of Positron Emission Tomography Detectors Using Boosted Learning-A Residual Physics Approach

Artificial intelligence (AI) is entering medical imaging, mainly enhancing image reconstruction. Nevertheless, improvements throughout the entire processing, from signal detection to computation, potentially offer significant benefits. This work presents a novel and versatile approach to detector optimization using machine learning (ML) and residual physics. We apply the concept to positron emission tomography (PET), intending to improve the coincidence time resolution (CTR). PET visualizes metabolic processes in the body by detecting photons with scintillation detectors. Improved CTR performance offers the advantage of reducing radioactive dose exposure for patients. Modern PET detectors with sophisticated concepts and read-out topologies represent complex physical and electronic systems requiring dedicated calibration techniques. Traditional methods primarily depend on analytical formulations successfully describing the main detector characteristics. However, when accounting for higher-order effects, additional complexities arise matching theoretical models to experimental reality. Our work addresses this challenge by combining traditional calibration with AI and residual physics, presenting a highly promising approach. We present a residual physics-based strategy using gradient tree boosting and physics-guided data generation. The explainable AI framework SHapley Additive exPlanations (SHAPs) was used to identify known physical effects with learned patterns. In addition, the models were tested against basic physical laws. We were able to improve the CTR significantly (more than 20%) for clinically relevant detectors of 19 mm height, reaching CTRs of 185 ps (450-550 keV).

Improving fraud detection with semi-supervised topic modeling and keyword integration

Fraud detection through auditors' manual review of accounting and financial records has traditionally relied on human experience and intuition. However, replicating this task using technological tools has represented a challenge for information security researchers. Natural language processing techniques, such as topic modeling, have been explored to extract information and categorize large sets of documents. Topic modeling, such as latent Dirichlet allocation (LDA) or non-negative matrix factorization (NMF), has recently gained popularity for discovering thematic structures in text collections. However, unsupervised topic modeling may not always produce the best results for specific tasks, such as fraud detection. Therefore, in the present work, we propose to use semi-supervised topic modeling, which allows the incorporation of specific knowledge of the study domain through the use of keywords to learn latent topics related to fraud. By leveraging relevant keywords, our proposed approach aims to identify patterns related to the vertices of the fraud triangle theory, providing more consistent and interpretable results for fraud detection. The model's performance was evaluated by training with several datasets and testing it with another one that did not intervene in its training. The results showed efficient performance averages with a 7% increase in performance compared to a previous job. Overall, the study emphasizes the importance of deepening the analysis of fraud behaviors and proposing strategies to identify them proactively.

Calendar ageing modelling using machine learning: an experimental investigation on lithium ion battery chemistries

Background: The phenomenon of calendar ageing continues to have an impact on battery systems worldwide by causing them to have undesirable operation life and performance. Predicting the degradation in the capacity can identify whether this phenomenon is occurring for a cell and pave the way for placing mechanisms that can circumvent this behaviour. Methods: In this study, the machine learning algorithms, Extreme Gradient Boosting (XGBoost) and artificial neural network (ANN) have been used to predict the calendar ageing data belonging to six types of cell chemistries namely, Lithium Cobalt Oxide, Lithium Iron Phosphate, Lithium Manganese Oxide, Lithium Titanium Oxide, Nickle Cobalt Aluminum Oxide and Nickle Manganese Cobalt Oxide. Results: Prediction results with overall Mean Absolute Percentage Error of 0.0126 have been obtained for XGBoost algorithm. Among these results, Nickle Cobalt Aluminum Oxide and Nickle Manganese Cobalt Oxide type cell chemistries stand out with their mean absolute percentage errors of 0.0035 and 0.0057 respectively. Also, algorithm fitting performance is relatively better for these chemistries at 100% state of charge and 60°C temperature compared to ANN results. ANN algorithm predicts with mean absolute error of approximately 0.0472 overall and 0.0238 and 0.03825 for Nickle Cobalt Aluminum Oxide and Nickle Manganese Cobalt Oxide. The fitting performance of ANN for Nickle Manganese Cobalt Oxide at 100% state of charge and 60°C temperature is especially poor compared to XGBoost. Conclusions: For an electric vehicle battery calendar ageing prediction application, XGBoost can establish itself as the primary choice more easily compared to ANN. The reason is XGBoost's error rates and fitting performance are more usable for such application especially for Nickel Cobalt Aluminum Oxide and Nickel Manganese Cobalt Oxide chemistries, which are amongst the most demanded cell chemistries for electric vehicle battery packs.

Spatiotemporal PM2.5 estimations in China from 2015 to 2020 using an improved gradient boosting decision tree

Fine particulate matter (PM_2.5) with spatiotemporal continuity can provide important basis for the assessment of adverse effects on human health. In recent years, researchers have done a lot of work on the surface PM_2.5 simulation. However, due to the limitations of data and models, it is difficult to accurately evaluate the spatial and temporal PM_2.5 variations on a fine scale. In this study, we adopted the multi-angle implementation of atmospheric correction (MAIAC) aerosol products, and proposed a spatiotemporal model based on the gradient boosting decision tree (GBDT) algorithm to retrieve PM_2.5 concentration across China from 2015 to 2020 at 1-km resolution. Our model achieved excellent performance, with overall CV-R² of 0.92, and annual CV-R² of 0.90-0.93. In addition, the model can also be used for evaluation on different time scales. Compared with previous studies, the model developed in our study performed better and more stable, which showed the highest accuracies in PM_2.5 estimation works at 1-km resolution. During the study period, the overall national PM_2.5 pollution showed a downward trend, with the annual mean concentration dropping from 42.42 μg/m³ to 27.91 μg/m³. The largest decrease occurred in Beijing-Tianjin-Hebei (BTH), with a trend of -5.17 μg/m³/yr, while it remains the most polluted region. The area meeting the secondary national air quality standard (<35 μg/m³) increased from ∼34% to ∼79%. These results indicate that the atmospheric environment has improved significantly. Moreover, different regions have different time nodes for the start of the continuous standard-met day during the year, and the duration is different as well. Overall, this study can provide reliable large-scale PM_2.5 estimations.

Universal Reconfigurable Hardware Accelerator for Sparse Machine Learning Predictive Models

This study presents a universal reconfigurable hardware accelerator for efficient processing of sparse decision trees, artificial neural networks and support vector machines. The main idea is to develop a hardware accelerator that will be able to directly process sparse machine learning models, resulting in shorter inference times and lower power consumption compared to existing solutions. To the author’s best knowledge, this is the first hardware accelerator of this type. Additionally, this is the first accelerator that is capable of processing sparse machine learning models of different types. Besides the hardware accelerator itself, algorithms for induction of sparse decision trees, pruning of support vector machines and artificial neural networks are presented. Such sparse machine learning classifiers are attractive since they require significantly less memory resources for storing model parameters. This results in reduced data movement between the accelerator and the DRAM memory, as well as a reduced number of operations required to process input instances, leading to faster and more energy-efficient processing. This could be of a significant interest in edge-based applications, with severely constrained memory, computation resources and power consumption. The performance of algorithms and the developed hardware accelerator are demonstrated using standard benchmark datasets from the UCI Machine Learning Repository database. The results of the experimental study reveal that the proposed algorithms and presented hardware accelerator are superior when compared to some of the existing solutions. Throughput is increased up to 2 times for decision trees, 2.3 times for support vector machines and 38 times for artificial neural networks. When the processing latency is considered, maximum performance improvement is even higher: up to a 4.4 times reduction for decision trees, a 84.1 times reduction for support vector machines and a 22.2 times reduction for artificial neural networks. Finally, since it is capable of supporting sparse classifiers, the usage of the proposed hardware accelerator leads to a significant reduction in energy spent on DRAM data transfers and a reduction of 50.16% for decision trees, 93.65% for support vector machines and as much as 93.75% for artificial neural networks, respectively.