Small Training Set Research Articles

Application of Prompt Learning Models in Identifying the Collaborative Problem Solving Skills in an Online Task

Collaborative problem solving (CPS) competence is considered one of the essential 21st-century skills. To facilitate the assessment and learning of CPS competence, researchers have proposed a series of frameworks to conceptualize CPS and explored ways to make sense of the complex processes involved in collaborative problem solving. However, encoding explicit behaviors into subskills within the frameworks of CPS skills is still a challenging task. Traditional studies have relied on manual coding to decipher behavioral data for CPS, but such coding methods can be very time-consuming and cannot support real-time analyses. Scholars have begun to explore approaches for constructing automatic coding models. Nevertheless, the existing models built using machine learning or deep learning techniques depend on a large amount of training data and have relatively low accuracy. To address these problems, this paper proposes a prompt-based learning pre-trained model. The model can achieve high performance even with limited training data. In this study, three experiments were conducted, and the results showed that our model not only produced the highest accuracy, macro F1 score, and kappa values on large training sets, but also performed the best on small training sets of the CPS behavioral data. The application of the proposed prompt-based learning pre-trained model contributes to the CPS skills coding task and can also be used for other CSCW coding tasks to replace manual coding.

MEF-AlloSite: an accurate and robust Multimodel Ensemble Feature selection for the Allosteric Site identification model

A crucial mechanism for controlling the actions of proteins is allostery. Allosteric modulators have the potential to provide many benefits compared to orthosteric ligands, such as increased selectivity and saturability of their effect. The identification of new allosteric sites presents prospects for the creation of innovative medications and enhances our comprehension of fundamental biological mechanisms. Allosteric sites are increasingly found in different protein families through various techniques, such as machine learning applications, which opens up possibilities for creating completely novel medications with a diverse variety of chemical structures. Machine learning methods, such as PASSer, exhibit limited efficacy in accurately finding allosteric binding sites when relying solely on 3D structural information.Scientific ContributionPrior to conducting feature selection for allosteric binding site identification, integration of supporting amino-acid–based information to 3D structural knowledge is advantageous. This approach can enhance performance by ensuring accuracy and robustness. Therefore, we have developed an accurate and robust model called Multimodel Ensemble Feature Selection for Allosteric Site Identification (MEF-AlloSite) after collecting 9460 relevant and diverse features from the literature to characterise pockets. The model employs an accurate and robust multimodal feature selection technique for the small training set size of only 90 proteins to improve predictive performance. This state-of-the-art technique increased the performance in allosteric binding site identification by selecting promising features from 9460 features. Also, the relationship between selected features and allosteric binding sites enlightened the understanding of complex allostery for proteins by analysing selected features. MEF-AlloSite and state-of-the-art allosteric site identification methods such as PASSer2.0 and PASSerRank have been tested on three test cases 51 times with a different split of the training set. The Student’s t test and Cohen’s D value have been used to evaluate the average precision and ROC AUC score distribution. On three test cases, most of the p-values (<0.05\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$< 0.05$$\\end{document}) and the majority of Cohen’s D values (>0.5\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$> 0.5$$\\end{document}) showed that MEF-AlloSite’s 1–6% higher mean of average precision and ROC AUC than state-of-the-art allosteric site identification methods are statistically significant.

Robust quantum dots charge autotuning using neural network uncertainty

Abstract This study presents a machine-learning-based procedure to automate the charge tuning of semiconductor spin qubits with minimal human intervention, addressing one of the significant challenges in scaling up quantum dot technologies. This method exploits artificial neural networks to identify noisy transition lines in stability diagrams, guiding a robust exploration strategy leveraging neural networks' uncertainty estimations. Tested across three distinct offline experimental datasets representing different single quantum dot technologies, the approach achieves over 99% tuning success rate in optimal cases, where more than 10% of the success is directly attributable to uncertainty exploitation. The challenging constraints of small training sets containing high diagram-to-diagram variability allowed us to evaluate the capabilities and limits of the proposed procedure.&amp;#xD;&amp;#xD;

Cross-lingual dependency parsing for a language with a unique script

Abstract Syntactic parsing is one of the areas in Natural Language Processing. The development of large-scale multilingual language models has enabled cross-lingual parsing approaches, which allows us to develop parsers for languages that do not have treebanks available. However, these approaches rely on the assumption that languages share orthographic representations and lexical entries. In this article, we investigate methods for developing a dependency parser for Xibe, a low-resource language that is written in a unique script. We first investigate lexicalized monolingual dependency parsing experiments to examine the effectiveness of word, part-of-speech, and character embeddings as well as pre-trained language models. Results show that character embeddings can significantly improve performance, while pre-trained language models decrease performance since they do not recognize the Xibe script. We also train delexicalized monolingual models, which yield competitive results to the best lexicalized model. Since the monolingual models are trained on a very small training set, we also investigate lexicalized and delexicalized cross-lingual models. We use six closely related languages as source language, which cover a wide range of scripts. In this setting, the delexicalized models achieve higher performance than lexicalized models. A final experiment shows that we can increase performance of the cross-lingual model by combining source languages and selecting the most similar sentences to Xibe as training set. However, all cross-lingual parsing results are still considerably lower than the monolingual model. We attribute the low performance of cross-lingual methods to syntactic and annotation differences as well as to the impoverished input of Universal Dependency Part-of-Speech tags that the delexicalized model has access to.

Basis Set Extrapolation from the Vanishing Counterpoise Correction Condition.

Basis set extrapolations are typically rationalized either from analytical arguments involving the partial-wave or principal expansions of the correlation energy in helium-like systems or from fitting extrapolation parameters to reference energetics for a small(ish) training set. Seeking to avoid both, we explore a third alternative: extracting extrapolation parameters from the requirement that the BSSE (basis set superposition error) should vanish at the complete basis set limit. We find this to be a viable approach provided that the underlying basis sets are not too small and reasonably well balanced. For basis sets not augmented by diffuse functions, BSSE minimization and energy fitting yield quite similar parameters.

Deep-learning-based image reconstruction with limited data: generating synthetic raw data using deep learning.

Deep learning has shown great promise for fast reconstruction of accelerated MRI acquisitions by learning from large amounts of raw data. However, raw data is not always available in sufficient quantities. This study investigates synthetic data generation to complement small datasets and improve reconstruction quality. An adversarial auto-encoder was trained to generate phase and coil sensitivity maps from magnitude images, which were combined into synthetic raw data. On a fourfold accelerated MR reconstruction task, deep-learning-based reconstruction networks were trained with varying amounts of training data (20 to 160 scans). Test set performance was compared between baseline experiments and experiments that incorporated synthetic training data. Training with synthetic raw data showed decreasing reconstruction errors with increasing amounts of training data, but importantly this was magnitude-only data, rather than real raw data. For small training sets, training with synthetic data decreased the mean absolute error (MAE) by up to 7.5%, whereas for larger training sets the MAE increased by up to 2.6%. Synthetic raw data generation improved reconstruction quality in scenarios with limited training data. A major advantage of synthetic data generation is that it allows for the reuse of magnitude-only datasets, which are more readily available than raw datasets.

Neuromorphic overparameterisation and few-shot learning in multilayer physical neural networks

Physical neuromorphic computing, exploiting the complex dynamics of physical systems, has seen rapid advancements in sophistication and performance. Physical reservoir computing, a subset of neuromorphic computing, faces limitations due to its reliance on single systems. This constrains output dimensionality and dynamic range, limiting performance to a narrow range of tasks. Here, we engineer a suite of nanomagnetic array physical reservoirs and interconnect them in parallel and series to create a multilayer neural network architecture. The output of one reservoir is recorded, scaled and virtually fed as input to the next reservoir. This networked approach increases output dimensionality, internal dynamics and computational performance. We demonstrate that a physical neuromorphic system can achieve an overparameterised state, facilitating meta-learning on small training sets and yielding strong performance across a wide range of tasks. Our approach’s efficacy is further demonstrated through few-shot learning, where the system rapidly adapts to new tasks.

Self-supervised contrastive learning improves machine learning discrimination of full thickness macular holes from epiretinal membranes in retinal OCT scans.

There is a growing interest in using computer-assisted models for the detection of macular conditions using optical coherence tomography (OCT) data. As the quantity of clinical scan data of specific conditions is limited, these models are typically developed by fine-tuning a generalized network to classify specific macular conditions of interest. Full thickness macular holes (FTMH) present a condition requiring urgent surgical repair to prevent vision loss. Other works on automated FTMH classification have tended to use supervised ImageNet pre-trained networks with good results but leave room for improvement. In this paper, we develop a model for FTMH classification using OCT B-scans around the central foveal region to pre-train a naïve network using contrastive self-supervised learning. We found that self-supervised pre-trained networks outperform ImageNet pre-trained networks despite a small training set size (284 eyes total, 51 FTMH+ eyes, 3 B-scans from each eye). On three replicate data splits, 3D spatial contrast pre-training yields a model with an average F1-score of 1.0 on holdout data (50 eyes total, 10 FTMH+), compared to an average F1-score of 0.831 for FTMH detection by ImageNet pre-trained models. These results demonstrate that even limited data may be applied toward self-supervised pre-training to substantially improve performance for FTMH classification, indicating applicability toward other OCT-based problems.

Leveraging Pretrained Transformers for Efficient Segmentation and Lesion Detection in Cone-Beam Computed Tomography Scans

IntroductionCone-beam computed tomography (CBCT) is widely used to detect jaw lesions, although CBCT interpretation is time-consuming and challenging. Artificial intelligence for CBCT segmentation may improve lesion detection accuracy. However, consistent automated lesion detection remains difficult, especially with limited training data. This study aimed to assess the applicability of pretrained transformer-based architectures for semantic segmentation of CBCT volumes when applied to periapical lesion detection. MethodsCBCT volumes (n = 138) were collected and annotated by expert clinicians using 5 labels – "lesion," "restorative material," "bone," "tooth structure," and "background." U-Net (convolutional neural network-based) and Swin-UNETR (transformer-based) models, pretrained (Swin-UNETR-PRETRAIN), and from scratch (Swin-UNETR-SCRATCH), were trained with subsets of the annotated CBCTs. These models were then evaluated for semantic segmentation performance using the Sørensen–Dice coefficient (DICE), lesion detection performance using sensitivity and specificity, and training sample size requirements by comparing models trained with 20, 40, 60, or 103 samples. ResultsTrained with 103 samples, Swin-UNETR-PRETRAIN achieved a DICE of 0.8512 for "lesion," 0.8282 for "restorative materials," 0.9178 for "bone," 0.9029 for "tooth structure," and 0.9901 for "background." “Lesion” DICE was statistically similar between Swin-UNETR-PRETRAIN trained with 103 and 60 images (P > .05), with the latter achieving 1.00 sensitivity and 0.94 specificity in lesion detection. With small training sets, Swin-UNETR-PRETRAIN outperformed Swin-UNETR-SCRATCH in DICE over all labels (P < .001 [n = 20], P < .001 [n = 40]), and U-Net in lesion detection specificity (P = .006 [n = 20], P = .031 [n = 40]). ConclusionsTransformer-based Swin-UNETR architectures allowed for excellent semantic segmentation and periapical lesion detection. Pretrained, it may provide an alternative with smaller training datasets compared to classic U-Net architectures.

Deep learning-based automated liver contouring using a small sample of radiotherapy planning computed tomography images

IntroductionNo study has yet investigated the minimum amount of data required for deep learning-based liver contouring. Therefore, this study aimed to investigate the feasibility of automated liver contouring using limited data. MethodsRadiotherapy planning Computed tomography (CT) images were subjected to various preprocessing methods, such as denoising and windowing. Segmentation was conducted using the modified Attention U-Net and Residual U-Net networks. Two different modified networks were trained separately for different training sizes. For each architecture, the model trained with the training set size that achieved the highest dice similarity coefficient (DSC) score was selected for further evaluation. Two unseen external datasets with different distributions from the training set were also used to examine the generalizability of the proposed method. ResultsThe modified Residual U-Net and Attention U-Net networks achieved average DSCs of 97.62% and 96.48%, respectively, on the test set, using 62 training cases. The average Hausdorff distances (AHDs) for the modified Residual U-Net and Attention U-Net networks were 0.57 mm and 0.71 mm, respectively. Also, the modified Residual U-Net and Attention U-Net networks were tested on two unseen external datasets, achieving DSCs of 95.35% and 95.82% for data from another center and 95.16% and 94.93% for the AbdomenCT-1K dataset, respectively. ConclusionThis study demonstrates that deep learning models can accurately segment livers using a small training set. The method, utilizing simple preprocessing and modified network architectures, shows strong performance on unseen datasets, indicating its generalizability. Implications for practiceThis promising result suggests its potential for automated liver contouring in radiotherapy planning.

Exploring galaxy properties of eCALIFA with contrastive learning

Contrastive learning (CL) has emerged as a potent tool for building meaningful latent representations of galaxy properties across a broad spectrum of wavelengths, ranging from optical and infrared to radio frequencies. These latent representations facilitate a variety of downstream tasks, including galaxy classification, similarity searches in extensive datasets, and parameter estimation, which is why they are often referred to as foundation models for galaxies. In this study, we employ CL on the latest extended data release from the Calar Alto Legacy Integral Field Area (CALIFA) survey, which encompasses a total of 895 galaxies with enhanced spatial resolution that reaches the limits imposed by natural seeing (FWHMPSF ∼ 1.5). We demonstrate that CL can be effectively applied to Integral Field Unit (IFU) surveys, even with relatively small training sets, to construct meaningful embedding where galaxies are well separated based on their physical properties. We discover that the strongest correlations in the embedding space are observed with the equivalent width of Hα, galaxy morphology, stellar metallicity, luminosity-weighted age, stellar surface mass density, the [NII]/Hα ratio, and stellar mass, in descending order of correlation strength. Additionally, we illustrate the feasibility of unsupervised separation of galaxy populations along the star formation main sequence, successfully identifying the blue cloud and the red sequence in a two-cluster scenario, and the green valley population in a three-cluster scenario. Our findings indicate that galaxy luminosity profiles have minimal impact on the construction of the embedding space, suggesting that morphology and spectral features play a more significant role in distinguishing between galaxy populations. Moreover, we explore the use of CL for detecting variations in galaxy population distributions across different large-scale structures, including voids, clusters, and filaments and walls. Nonetheless, we acknowledge the limitations of the CL framework and our specific training set in detecting subtle differences in galaxy properties, such as the presence of an AGN or other minor scale variations that exceed the scope of primary parameters such as the stellar mass or morphology. Conclusively, we propose that CL can serve as an embedding function for the development of larger models capable of integrating data from multiple datasets, thereby advancing the construction of more comprehensive foundation models for galaxies.

SkinNet-14: a deep learning framework for accurate skin cancer classification using low-resolution dermoscopy images with optimized training time

The increasing incidence of skin cancer necessitates advancements in early detection methods, where deep learning can be beneficial. This study introduces SkinNet-14, a novel deep learning model designed to classify skin cancer types using low-resolution dermoscopy images. Unlike existing models that require high-resolution images and extensive training times, SkinNet-14 leverages a modified compact convolutional transformer (CCT) architecture to effectively process 32 × 32 pixel images, significantly reducing the computational load and training duration. The framework employs several image preprocessing and augmentation strategies to enhance input image quality and balance the dataset to address class imbalances in medical datasets. The model was tested on three distinct datasets—HAM10000, ISIC and PAD—demonstrating high performance with accuracies of 97.85%, 96.00% and 98.14%, respectively, while significantly reducing the training time to 2–8 s per epoch. Compared to traditional transfer learning models, SkinNet-14 not only improves accuracy but also ensures stability even with smaller training sets. This research addresses a critical gap in automated skin cancer detection, specifically in contexts with limited resources, and highlights the capabilities of transformer-based models that are efficient in medical image analysis.

Using reservoir computing to solve FPK equations for stochastic dynamical systems under Gaussian or Non-Gaussian excitation

This paper presents a new approach that uses the Reservoir Computing Algorithm to solve Fokker-Planck-Kolmogorov (FPK) equation excited by both Gaussian white noise and non-Gaussian noise. Unlike typical numerical methods, this methodology does not necessitate spatial reconstruction or numerical supplementation. The novelty of this paper lies in the modifications made to the conventional Reservoir Computing algorithm. We altered the approach for calculating values of the input weight matrix and incorporated autoregressive techniques in the reservoir layer. In addition, we applied data normalization to the training data before training the algorithm to avoid a zero solution. The efficacy of this approach was verified through multiple arithmetic examples, showcasing its practicality and efficiency in solving FPK equations. Moreover, the Reservoir Computing-FPK algorithm is capable of solving high-dimensional and fractional-order FPK equations with a smaller training set than earlier algorithms. Finally, we analyzed how values of the input weight matrix and regularization parameter affected the performance of the algorithm. The findings suggest that the careful selection of hyperparameters can greatly improve the performance of the Reservoir Computing algorithm.

Novel Domain Knowledge-Encoding Algorithm Enables Label-Efficient Deep Learning for Cardiac CT Segmentation to Guide Atrial Fibrillation Treatment in a Pilot Dataset.

Background: Segmenting computed tomography (CT) is crucial in various clinical applications, such as tailoring personalized cardiac ablation for managing cardiac arrhythmias. Automating segmentation through machine learning (ML) is hindered by the necessity for large, labeled training data, which can be challenging to obtain. This article proposes a novel approach for automated, robust labeling using domain knowledge to achieve high-performance segmentation by ML from a small training set. The approach, the domain knowledge-encoding (DOKEN) algorithm, reduces the reliance on large training datasets by encoding cardiac geometry while automatically labeling the training set. The method was validated in a hold-out dataset of CT results from an atrial fibrillation (AF) ablation study. Methods: The DOKEN algorithm parses left atrial (LA) structures, extracts "anatomical knowledge" by leveraging digital LA models (available publicly), and then applies this knowledge to achieve high ML segmentation performance with a small number of training samples. The DOKEN-labeled training set was used to train a nnU-Net deep neural network (DNN) model for segmenting cardiac CT in N = 20 patients. Subsequently, the method was tested in a hold-out set with N = 100 patients (five times larger than training set) who underwent AF ablation. Results: The DOKEN algorithm integrated with the nn-Unet model achieved high segmentation performance with few training samples, with a training to test ratio of 1:5. The Dice score of the DOKEN-enhanced model was 96.7% (IQR: 95.3% to 97.7%), with a median error in surface distance of boundaries of 1.51 mm (IQR: 0.72 to 3.12) and a mean centroid-boundary distance of 1.16 mm (95% CI: -4.57 to 6.89), similar to expert results (r = 0.99; p < 0.001). In digital hearts, the novel DOKEN approach segmented the LA structures with a mean difference for the centroid-boundary distances of -0.27 mm (95% CI: -3.87 to 3.33; r = 0.99; p < 0.0001). Conclusions: The proposed novel domain knowledge-encoding algorithm was able to perform the segmentation of six substructures of the LA, reducing the need for large training data sets. The combination of domain knowledge encoding and a machine learning approach could reduce the dependence of ML on large training datasets and could potentially be applied to AF ablation procedures and extended in the future to other imaging, 3D printing, and data science applications.

Progressive transfer learning for advancing machine learning-based reduced-order modeling

To maximize knowledge transfer and improve the data requirement for data-driven machine learning (ML) modeling, a progressive transfer learning for reduced-order modeling (p-ROM) framework is proposed. A key concept of p-ROM is to selectively transfer knowledge from previously trained ML models and effectively develop a new ML model(s) for unseen tasks by optimizing information gates in hidden layers. The p-ROM framework is designed to work with any type of data-driven ROMs. For demonstration purposes, we evaluate the p-ROM with specific Barlow Twins ROMs (p-BT-ROMs) to highlight how progress learning can apply to multiple topological and physical problems with an emphasis on a small training set regime. The proposed p-BT-ROM framework has been tested using multiple examples, including transport, flow, and solid mechanics, to illustrate the importance of progressive knowledge transfer and its impact on model accuracy with reduced training samples. In both similar and different topologies, p-BT-ROM achieves improved model accuracy with much less training data. For instance, p-BT-ROM with four-parent (i.e., pre-trained models) outperforms the no-parent counterpart trained on data nine times larger. The p-ROM framework is poised to significantly enhance the capabilities of ML-based ROM approaches for scientific and engineering applications by mitigating data scarcity through progressively transferring knowledge.

A simple scheme to amplify inter-class discrepancy for improving few-shot fine-grained image classification

Few-shot image classification is a challenging topic in pattern recognition and computer vision. Few-shot fine-grained image classification is even more challenging, due to not only the few shots of labelled samples but also the subtle differences to distinguish subcategories in fine-grained images. A recent method called task discrepancy maximisation (TDM) can be embedded into the feature map reconstruction network (FRN) to generate discriminative features, by preserving the appearance details through reconstructing the query image and then assigning higher weights to more discriminative channels, producing the state-of-the-art performance for few-shot fine-grained image classification. However, due to the small inter-class discrepancy in fine-grained images and the small training set in few-shot learning, the training of FRN+TDM can result in excessively flexible boundaries between subcategories and hence overfitting. To resolve this problem, we propose a simple scheme to amplify inter-class discrepancy and thus improve FRN+TDM. To achieve this aim, instead of developing new modules, our scheme only involves two simple amendments to FRN+TDM: relaxing the inter-class score in TDM, and adding a centre loss to FRN. Extensive experiments on five benchmark datasets showcase that, although embarrassingly simple, our scheme is quite effective to improve the performance of few-shot fine-grained image classification. The code is available at https://github.com/Airgods/AFRN.git.

Fire detection and risk assessment via Support Vector Regression with Flattening-Samples Based Augmented Regularization

We propose a Hybrid Support Vector Regression (SVR) with Flattening-Samples Based Augmented Regularization (Hybrid FSR-SVR) architecture for multi-sensor fire detection and forest fire risk assessment. The Hybrid FSR-SVR is a lightweight architecture built upon the novel Flattening-Samples Based Augmented Regularization (FSR) approach and temporal trends of environmental variables. The FSR approach augments l2 norm based smoothing term into an l1-l2 combination, facilitating the integration of l1 regularization into the SVR method, thereby enhancing generalization with minimal computational load. We evaluate the performance of Hybrid FSR-SVR using two distinct datasets covering indoor and forest fires, benchmarking against 15 machine learning models, including state-of-the-art techniques, such as Recurrent Trend Predictive Neural Network (rTPNN), Long-Short Term Memory (LSTM), Multi-Layer Perceptron (MLP), Gated Recurrent Unit (GRU), and Gradient Boosting. Our findings demonstrate that Hybrid FSR-SVR effectively assesses the risk of forest fire, enabling early preventive measures. Notably, it achieves a remarkable accuracy of 0.95 for forest fire detection and ranks third with 0.88 accuracy for indoor fire detection. Importantly, it exhibits computation times significantly lower – by 1 to 2 orders of magnitude – than the majority of compared techniques. The superior generalization ability of Hybrid FSR-SVR, facilitated by flattening-samples based augmented regularization, allows for high detection performance even with smaller training sets.

Resource-constrained narrowband optoelectronic oscillator-based reservoir computing for classification of modulated signals.

We experimentally investigate the performance of narrowband optoelectronic oscillator (OEO) reservoir computers using the standard 10th-order nonlinear autoregressive-moving-average (NARMA10) task. Because comparing results from differently parameterized photonic time-delay systems can be difficult, we introduce a new, to the best of our knowledge, metric that accounts for system size, computational accuracy, and training effort overhead in order to provide an "at-a-glance" method to holistically determine a reservoir computer's performance. We then demonstrate the first experimental effort of narrowband OEO-based reservoir computing for the RADIOML dataset, which consists of recognizing and classifying IQ-modulated radio signals including analog and digital modulations. Our results indicate that narrowband OEOs are capable of achieving reasonable accuracies with exceptionally small training sets, thereby paving the way to real-time machine learning for radio frequency signals.

Application of a Machine Learning-Based Classification Approach for Developing Host Protein Diagnostic Models for Infectious Disease

In recent years, infectious disease diagnosis has increasingly turned to host-centered approaches as a complement to pathogen-directed ones. The former, however, typically requires the interpretation of complex multiple biomarker datasets to arrive at an informative diagnostic outcome. This report describes a machine learning (ML)-based classification workflow that is intended as a template for researchers seeking to apply ML approaches for developing host-based infectious disease biomarker classifiers. As an example, we built a classification model that could accurately distinguish between three disease etiology classes: bacterial, viral, and normal in human sera using host protein biomarkers of known diagnostic utility. After collecting protein data from known disease samples, we trained a series of increasingly complex Auto-ML models until arriving at an optimized classifier that could differentiate viral, bacterial, and non-disease samples. Even when limited to a relatively small training set size, the model had robust diagnostic characteristics and performed well when faced with a blinded sample set. We present here a flexible approach for applying an Auto-ML-based workflow for the identification of host biomarker classifiers with diagnostic utility for infectious disease, and which can readily be adapted for multiple biomarker classes and disease states.

Evolutionary Strategies AI Addresses Multiple Technical Challenges in Deep Learning Deployment: Proof-of-Principle Demonstration for Neuroblastoma Brain Metastasis Detection.

Two significant obstacles hinder the advancement of Radiology AI. The first is the challenge of overfitting, where small training data sets can result in unreliable outcomes. The second challenge is the need for more generalizability, the lack of which creates difficulties in implementing the technology across various institutions and practices. A recent innovation, deep neuroevolution (DNE), has been introduced to tackle the overfitting issue by training on small data sets and producing accurate predictions. However, the generalizability of DNE has yet to be proven. This paper strives to overcome this barrier by demonstrating that DNE can achieve satisfactory results in diverse external validation sets. The main innovation of the work is thus showing that DNE can generalize to varied outside data. Our example use case is predicting brain metastasis from neuroblastoma, emphasizing the importance of AI with limited data sets. Despite image collection and labeling advancements, rare diseases will always constrain data availability. We optimized a convolutional neural network (CNN) with DNE to demonstrate generalizability. We trained the CNN with 60 MRI images and tested it on a separate diverse collection of images from over 50 institutions. For comparison, we also trained with the more traditional stochastic gradient descent (SGD) method, with the two variants of (1) training from scratch and (2) transfer learning. Our results show that DNE demonstrates excellent generalizability with 97% accuracy on the heterogeneous testing set, while neither form of SGD could reach 60% accuracy. DNE's ability to generalize from small training sets to external and diverse testing sets suggests that it or similar approaches may play an integral role in improving the clinical performance of AI.