Discriminant Model Research Articles

Inverse design of nanophotonic devices enabled by optimization algorithms and deep learning: recent achievements and future prospects

Abstract Nanophotonics, which explores significant light–matter interactions at the nanoscale, has facilitated significant advancements across numerous research fields. A key objective in this area is the design of ultra-compact, high-performance nanophotonic devices to pave the way for next-generation photonics. While conventional brute-force, intuition-based forward design methods have produced successful nanophotonic solutions over the past several decades, recent developments in optimization methods and artificial intelligence offer new potential to expand these capabilities. In this review, we delve into the latest progress in the inverse design of nanophotonic devices, where AI and optimization methods are leveraged to automate and enhance the design process. We discuss representative methods commonly employed in nanophotonic design, including various meta-heuristic algorithms such as trajectory-based, evolutionary, and swarm-based approaches, in addition to adjoint-based optimization. Furthermore, we explore state-of-the-art deep learning techniques, involving discriminative models, generative models, and reinforcement learning. We also introduce and categorize several notable inverse-designed nanophotonic devices and their respective design methodologies. Additionally, we summarize the open-source inverse design tools and commercial foundries. Finally, we provide our perspectives on the current challenges of inverse design, while offering insights into future directions that could further advance this rapidly evolving field.

The brain-gut microbiota network (BGMN) is correlated with symptom severity and neurocognition in patients with schizophrenia.

The association between the human brain and gut microbiota, known as the "brain-gut-microbiota axis", is involved in the neuropathological mechanisms of schizophrenia (SZ); however, its association patterns and correlations with symptom severity and neurocognition are still largely unknown. In this study, 43 SZ patients and 55 normal controls (NCs) were included, and resting-state functional magnetic resonance imaging (rs-fMRI) and gut microbiota data were acquired for each participant. First, the brain features of brain images and functional brain networks were computed from rs-fMRI data; the gut features of gut microbiota abundance and the gut microbiota network were computed from gut microbiota data. Second, we propose a novel methodology to construct an individual brain-gut microbiota network (BGMN) for each participant by combining the brain and gut features via multiple strategies. Third, discriminative models between SZ patients and NCs were built using the connectivity matrices of the BGMN as input features. Moreover, the correlations between the most discriminative features and the scores of symptom severity and neurocognition were analyzed in SZ patients. The results showed that the best discriminative model between SZ patients and NCs was achieved using the connectivity matrices of the BGMN when all the brain and gut features were integrated, with an accuracy of 0.90 and an area under the curve value of 0.97. The most discriminative features were related primarily to the genera Faecalibacterium and Collinsella, in which the genus Faecalibacterium was linked to the visual system and subcortical cortices and the genus Collinsella was linked to the default network and subcortical cortices. Furthermore, parts of the most discriminative features were significantly correlated with the scores of neurocognition in the SZ patients. The methodology for constructing individual BGMNs proposed in this study can help us reveal the associations between the brain and gut microbiota and understand the neuropathology of SZ.

Machine Learning-Based Tectonic Discrimination Using Basalt Element Geochemical Data: Insights into the Carboniferous–Permian Tectonic Regime of Western Tianshan Orogen

Identifying the tectonic setting of rocks is essential for gaining insights into the geological contexts in which these rocks were formed, aiding in tectonic plate reconstruction and enhancing our comprehensive understanding of the Earth’s history. The application of machine learning algorithms helps identify complex patterns and relationships between big data that may be overlooked by binary or ternary tectonomagmatic discrimination diagrams based on basalt compositions. In this study, three machine learning algorithms, i.e., Support Vector Machine (SVM), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost), were employed to classify the basalts from seven diverse settings, including intraplate basalts, island arc basalts, ocean island basalts, mid-ocean ridge basalts, back-arc basin basalts, oceanic flood basalts, and continental flood basalts. Specifically, for altered and fresh basalt samples, we utilized 22 immobile elements and 35 major and trace elements, respectively, to construct discrimination models. The results indicate that XGBoost demonstrates the best performance in discriminating basalts into seven tectonic settings, achieving accuracies of 85% and 89% for the altered and fresh basalt samples, respectively. A key innovation of our newly developed tectonic discrimination model is the establishment of tailored models for altered and fresh basalts. Moreover, by omitting isotopic features during model construction, the new models offer broader applicability in predicting a wider range of basalt samples in practical scenarios. The classification models were applied to investigate the Carboniferous to Permian evolution in the Western Tianshan Orogen (WTO), revealing that the subduction of Tianshan Ocean ceased at the end of Carboniferous and the WTO evolved into a post-collisional orogenesis during the Permian.

GSCAT-UNET: Enhanced U-Net model with spatial-channel attention gate and three-level attention for oil spill detection using SAR data.

Marine pollution due to oil spills presents major risks to coastal areas and aquatic life, leading to serious environmental health concerns. Oil Spill detection using SAR data has transitioned from traditional segmentation to a variety of machine learning & deep learning models like UNET proving its efficiency for the task. This research paper proposes a GSCAT-UNET model for efficient oil spill detection and discrimination from lookalikes. The GSCAT-UNET is an advanced UNET architecture comprising of Spatial-Channel Attention Gates(SCAG), Three Level Attention Module(TLM) and Global Feature Module(GFM) for global level oil spill feature enhancement leading to effective oil spill detection and discrimination from lookalikes. Sentinel-1 Dual-Pol SAR dataset of 1112 images and respective labeled images (5 classes) including confirmed oil spills and lookalikes is used to demonstrate the efficacy of the GSCAT-UNET model. The GSCAT-UNET model significantly enhances segmentation accuracy and robustness for oil spill detection with 5% higher accuracy and 29% higher IoU i.e. 93.7% compared to the UNET segmentation model, addressing the challenges of SAR data complexities and imbalanced datasets. The strong performance of the GSCAT-UNET model demonstrates its potential as a critical tool for disaster response and environmental monitoring.

A Future Picture: A Review of Current Generative Adversarial Neural Networks in Vitreoretinal Pathologies and Their Future Potentials

Machine learning has transformed ophthalmology, particularly in predictive and discriminatory models for vitreoretinal pathologies. However, generative modeling, especially generative adversarial networks (GANs), remains underexplored. GANs consist of two neural networks—the generator and discriminator—that work in opposition to synthesize highly realistic images. These synthetic images can enhance diagnostic accuracy, expand the capabilities of imaging technologies, and predict treatment responses. GANs have already been applied to fundus imaging, optical coherence tomography (OCT), and fluorescein autofluorescence (FA). Despite their potential, GANs face challenges in reliability and accuracy. This review explores GAN architecture, their advantages over other deep learning models, and their clinical applications in retinal disease diagnosis and treatment monitoring. Furthermore, we discuss the limitations of current GAN models and propose novel applications combining GANs with OCT, OCT-angiography, fluorescein angiography, fundus imaging, electroretinograms, visual fields, and indocyanine green angiography.

Study on the safety distance determination model for natural gas wells in Ordos Basin subsidence areas of coal mining

To optimize the safe distance between coal mining faces and natural gas wells, with the Ordos Basin as the engineering background for the intersectional development of natural gas and coal resources, a Gaussian function equation for subsidence curves under complete mining was proposed based on the probability integral method, and a model was constructed to determine the coordinated mining safety distance between coal mining faces and natural gas wells. Based on the distributed optical fiber monitoring principle of the BOTDR technology, on-site monitoring of the simulated monitoring well GX1 within the safety distance was carried out. The results indicated that the maximum strain change values of the outer casing and inner production casing of the simulated monitoring well GX1 were 836 με and 766 με, respectively, far less than the maximum ultimate tensile strain value of the overlying rock at the same depth position. Furthermore, the casing structure possessed ample resistance capability, ensuring no noticeable deformation occurred. Through comparison, it was found that the discriminant model reduced the reserved safety distance between the coal mining working face and the natural gas well by 58% compared with the currently used avoidance method of the surface subsidence boundary, liberated coal resources, and achieved a "win–win" situation of the coordinated mining of natural gas resources and coal resources. It can reasonably adjust the coordinated mining problem without a natural gas-coal co-mining mechanism, provide a new method for the layout of natural gas wells in gas-coal overlapping areas, and provide new ideas for the coordinated mining of coal-related associated resources.

Detection of reactor neutrinos using gadolinium-doped plastic scintillators enhanced by machine learning techniques

Abstract In this study, we present a detailed approach for detecting reactor neutrinos by employing gadolinium-doped plastic scintillators combined with machine learning techniques. The proposed detector design was simulated using the Geant4 framework, featuring segmented modules of plastic scintillators doped with gadolinium to enhance neutron capture efficiency. Inverse beta decay (IBD) events generated by ERNIE and background events produced by cosmic ray simulations were used to train and test an Extreme Gradient Boosting (XGBoost) model for signal-background discrimination. The model demonstrated high discrimination accuracy for prompt IBD events but encountered challenges with delayed neutron background discrimination due to similarities in neutron capture. Performance comparisons with traditional cut-based analyses highlighted the improved accuracy achieved through machine learning, particularly when utilizing additional event features. This work establishes the potential of gadolinium-doped plastic scintillators and machine learning in enhancing neutrino detection and lays the groundwork for future experimental validation and detector optimization.

Unleashing the Potential of Pre-Trained Diffusion Models for Generalizable Person Re-Identification.

Domain-generalizable re-identification (DG Re-ID) aims to train a model on one or more source domains and evaluate its performance on unseen target domains, a task that has attracted growing attention due to its practical relevance. While numerous methods have been proposed, most rely on discriminative or contrastive learning frameworks to learn generalizable feature representations. However, these approaches often fail to mitigate shortcut learning, leading to suboptimal performance. In this work, we propose a novel method called diffusion model-assisted representation learning with a correlation-aware conditioning scheme (DCAC) to enhance DG Re-ID. Our method integrates a discriminative and contrastive Re-ID model with a pre-trained diffusion model through a correlation-aware conditioning scheme. By incorporating ID classification probabilities generated from the Re-ID model with a set of learnable ID-wise prompts, the conditioning scheme injects dark knowledge that captures ID correlations to guide the diffusion process. Simultaneously, feedback from the diffusion model is back-propagated through the conditioning scheme to the Re-ID model, effectively improving the generalization capability of Re-ID features. Extensive experiments on both single-source and multi-source DG Re-ID tasks demonstrate that our method achieves state-of-the-art performance. Comprehensive ablation studies further validate the effectiveness of the proposed approach, providing insights into its robustness.

A dataset for evaluating clinical research claims in large language models

Large language models (LLMs) have the potential to enhance the verification of health claims. However, issues with hallucination and comprehension of logical statements require these models to be closely scrutinized in healthcare applications. We introduce CliniFact, a scientific claim dataset created from hypothesis testing results in clinical research, covering 992 unique interventions for 22 disease categories. The dataset used study arms and interventions, primary outcome measures, and results from clinical trials to derive and label clinical research claims. These claims were then linked to supporting information describing clinical trial results in scientific publications. CliniFact contains 1,970 instances from 992 unique clinical trials related to 1,540 unique publications. When evaluating LLMs against CliniFact, discriminative models, such as BioBERT with an accuracy of 80.2%, outperformed generative counterparts, such as Llama3-70B, which reached 53.6% accuracy (p-value < 0.001). Our results demonstrate the potential of CliniFact as a benchmark for evaluating LLM performance in clinical research claim verification.

Progress in quality assessment of Italian saffron

Saffron (Crocus sativus L.) is the most expensive spice in the World, and Italy is among the major European producers. The study aims to improve value and recognition of high quality saffron by proposing a subdivision within the first quality category, according to ISO 3632 standards. The analysis of 125 saffron samples, collected from different Italian regions in 2021–2022 harvesting seasons, revealed that 95% of the samples met the first quality criteria, following ISO guidelines. Consequently, for the first quality samples, a differentiation into “premium”, “superior” and “high-quality” subcategories was proposed. Along with traditional methods, FT-NIR spectroscopy combined with chemometrics was employed for a comprehensive quality assessment. Discriminant and class-modelling approaches were developed to predict saffron quality subcategory. Best results were obtained by linear discriminant analysis models with accuracy in calibration (92.7%), cross-validation (86.4%) and prediction (87.0%). However, SIMCA modelling resulted more appropriate for class-modelling, confirming that none of the “premium” samples were misclassified as “high-quality” and vice-versa. The results support the inclusion of subcategories within ISO 3632 standards, thus refining the classification of saffron quality. Furthermore, the study emphasises the effectiveness of FT-NIR spectroscopy as a valuable tool for saffron quality assessment, with potential implications for industry standards and practices.

Alchemy in Nature: The Role of Lawsonia inermis Extract Choice in Crafting Potent Anticancer Metal Nanoparticles.

Phyto-nanotechnology provides an eco-friendly approach for synthesizing biocompatible metal nanoparticles (NPs) with therapeutic potential. Lawsonia inermis (LI) has been historically valued for its diverse medicinal applications, especially its exceptional biological potency against various skin diseases, attributed to its rich abundance of bioactive compounds. Therefore, herein, plant-based iron and zinc NPs were biofabricated via sustainable and simple methods, using crude extracts of the aerial parts of LI as reducing, coating, and stabilizing agents. Since the extraction method affects the type of extracted phytocompounds, two extraction approaches─aqueous and hydro-alcoholic─were applied to determine the influence of the extraction route on the physicochemical and biological properties of the formed NPs. These properties were characterized via various analytical techniques and assays. The UV-Vis spectra revealed absorption bands ranging from 265 to 270 nm, while FT-IR confirmed the successful coating of the NPs with the extracts' phytochemicals, validating the biofabrication of the proposed NPs. The alcoholic-based NPs displayed higher total phenolic content, total flavonoid content, and antioxidant effect compared to their aqueous-based counterparts, reaching up to 55.13 μg of GAE/1 mg of dry weight (DW), 30.48 μg of QU/1 mg of DW, and IC50 of 46.02 μg/mL, respectively. All tested samples, except for Fe NPs, displayed significant cytotoxic effects against skin cancer, resulting in a cell viability as low as 1% at 1000 μg/mL. QTOF-LC/MS/MS analyses of LI extracts revealed tentative identification of more than 100 metabolites with phenolic compounds representing the largest share. Orthogonal Projections to Latent Structures Discriminant Analysis modeling revealed a clear separation between both extracts, with more than 40 marker compounds. The results indicated that both extracts were effective for the green synthesis of Fe and Zn NPs for biomedical applications, with the alcoholic extract of LI as a superior coating candidate and the aqueous extract as a stronger reducing agent. This work showcases the influence of extraction protocols on physicochemical and biological characteristics of the resulting nanoparticles.

Evaluation of Peri-Implantitis through Fourier-Transform Infrared Spectroscopy on Saliva.

Peri-implantitis is characterized as a pathological change in the tissues around dental implants. Fourier-transform infrared spectroscopy (FTIR) provides molecular information from optical phenomena observed by the vibration of molecules, which is used in biological studies to characterize changes and serves as a form of diagnosis. this case-control study evaluated the peri-implant disease by using FTIR spectroscopy with attenuated total reflectance in the fingerprint region. 38 saliva samples were evaluated, 17 from the control group and 21 from the peri-implantitis group. Clinical data such as plaque index (PI), gingival index, probing depth (PS), and attachment level were assessed. The results of clinical parameters showed a statistical difference between the two groups regarding an excess of the PI. In the FTIR-ATR analysis, the main components revealed vibrational modes of fatty acids, histidine, lipid esters, nucleic acids, and tryptophan, with the main molecules contributing to spectral discrimination. The five-component partial least-squares discriminant analysis classification model had an accuracy of 81%, showing differences between healthy and diseased implants. the FTIR spectroscopy provides important molecular characteristics of the samples and the results in association with clinical data show the effectiveness of using this tool for diagnosing the disease.

Identification of high-risk oral leukoplakia (OLK) using combined Raman spectroscopic analysis of brush biopsy and saliva samples: A proof of concept study.

The gold standard method of diagnosis of oral leukoplakia (OLK) is a tissue biopsy followed by histological examination. Raman spectroscopic studies of cytological brush biopsy and saliva samples have previously been shown to differentiate low (no and mild dysplasia) and high risk (moderate and severe dysplasia) OLKs, discriminant models of cellular samples achieving higher specificity, whereas those based on saliva samples achieved higher sensitivity. The current study combines the spectral data sets of cell and saliva samples in an attempt to improve the overall efficiency of the discriminating models. Raman spectral data from cellular (nucleus and cytoplasm) and saliva samples, collected from patients with OLK (n=12), was analysed as a concatenated or fused dataset and as data blocks in a multiblock analysis. The concatenated data was subjected to partial least squares-discriminant analysis (PLS-DA) to discriminate high and low grade dysplasia. Finally, multi-block analysis was performed using sequential orthogonalised PLS-DA, by which each set of data blocks was combined sequentially to provide maximum discrimination. For the concatenated dataset of cells and saliva, 87% sensitivity and 76% specificity were achieved, while in the case of the multi-block analysis, 97% sensitivity and 100% specificity were achieved. It is concluded that multiblock analysis provides maximum sensitivity and specificity using both cell and saliva datasets, compared to fused datasets. This study demonstrates that Raman spectroscopy of minimally invasive brush biopsy and saliva samples could have a role in differentiating high and low-risk OLKs.

Distinguishing the botanical origins of rare honey through untargeted metabolomics and machine learning interpreting flavonoid profiles.

Distinguishing the botanic origins of monofloral honey is the foremost concern in ensuring its authentication. In this work, an innovative, green, and comprehensive approach was developed to distinguish the botanic origins of four types of rare honey, and the strategy involved in the following aspects: Based on theoretical design, suitable natural deep eutectic solvent (NADES) was screened to extract flavonoids from honey samples; after NADES extracts were directly analyzed by high-resolution mass spectrometry, the discrimination models of monofloral honey were established by untargeted metabolomics combined with machine learning. Based on the comparison of various models, the Random Forest algorithm had higher prediction accuracy for four types of monofloral honey, and characteristic compounds for each rare monofloral honey were screened based on SHapley Additive exPlanations values. This work provides a new perspective on the use of AI technology and green chemistry to control the quality of honey.

Prevalence of keratoconus and keratoconus suspect, and their characteristics on corneal tomography in a population-based study.

Keratoconus (KC) is a progressive corneal disorder resulting in severe visual impairment. We aimed to determine the prevalence and corneal tomographic characteristics of KC and keratoconus suspect (KCS) in a population-based study, and to construct discrimination models with or without corneal tomography. A total of 1,544 eyes (822 participants aged ≥35 years) were evaluated using data from the Yamagata Study (2015-2017). Systemic and ophthalmological examinations including corneal tomography with swept-source anterior segment optical coherence tomography (AS-OCT) were conducted to determine the prevalence and corneal tomographic characteristics of KC and KCS. In addition, data on 766 eyes were used to construct discrimination models with or without corneal tomography. In results, KC was diagnosed in six (0.85%) participants, and KCS was diagnosed in 27 (1.46%) participants. The values including corneal power, keratometric cylinder, corneal central and thinnest thickness, corneal asymmetry, higher-order irregularity, and their inter-eye differences were associated with KC and KCS. The areas under the receiver operating characteristic curves for the three multivariate discrimination models (without corneal tomography, with corneal tomography, and without corneal tomography + inter-eye difference models) for participants with KC or KCS were 0.848, 1.000, and 0.930, respectively. When corneal tomography is unavailable, inter-eye differences in corneal parameters may be useful screening tools for KC and KCS.

Tracing the change of the volatile compounds of soy sauce at different fermentation times by PTR-TOF-MS, E-nose and GC–MS

Proton-transfer reaction time-of-flight mass spectrometry (PTR-TOF-MS), combined with electronic nose (E-nose), was first used to track the change of volatile organic compounds (VOCs) in soy sauce in this study. The results showed that 163 VOCs with different mass numbers were identified. Based on the differences in VOCs, the entire fermentation cycle was divided into four stages (0D and 15D; 30D-75D; 90D; 105D120D). The key stage of aroma formation was on day 90, when the VOC content reached its peak. It was clear that the key feature differential component was nonanal and phenethyl acetate by partial least squares discriminant analysis model, which showed significant correlation with the fermentation time. Moreover, the Quadratic Discriminant Analysis and Linear Discriminant Analysis models can accurately predict the fermentation cycle of soy sauce. The results of study demonstrated the potential of PTR-TOF-MS for comprehensive online monitoring and distinguishing of VOCs change during soy sauce fermentation.

Immunoglobulin G N-glycan markers of mild cognitive impairment in a Chinese population with cerebrovascular stenosis: A case-control study

BackgroundImmunoglobulin G (IgG) N-glycans have been shown to regulate the inflammatory response in the context of disease. In recent years, it has been found to be associated with several neurodegenerative disorders. In this study, we examined the relationship between IgG N-glycans and mild cognitive impairment (MCI) in a high-risk population for MCI, specifically patients with cerebrovascular stenosis. MethodsIn a case-control study, we investigated IgG N-glycans and cytokines in MCI and non-MCI patients in a population with cerebrovascular stenosis. A multifactorial logistic regression analysis was employed to investigate the potential association between IgG N-glycoprotein and MCI, with familial error rates being corrected for using the Benjamin-Hochberg method. To construct discriminatory models, logistic stepwise regression was employed and evaluated for their diagnostic efficacy. ResultsA statistically significant difference was found in eight of the IgG-GPs between the two groups. Three IgG-GPs were correlated with MCI, with an overall false discovery rate <0.05. Specifically, IgG-GP7 (non-sialylated glycan) was positively correlated with MCI, while IgG-GP14 (digalactosylated glycans) and IgG-GP18 (bis-sialylated glycan) were negatively correlated with MCI. The model constructed by combining IgG N-glycans (IgG-GP7, IgG-GP14, IgG-GP18) and cytokines (IL-1β, IL-10, BDNF and VEGF) demonstrated the highest diagnostic efficacy [AUC: 0.939, 95 % CI: (0.910–0.967)]. DiscussionIn the present study, we observed that agalactosylation and no-sialylation play a role in the progression of MCI by influencing the pro-inflammatory impact of IgG. The integration of IgG N-glycan and cytokines into a discriminative model demonstrated strong diagnostic efficacy, suggesting its potential use as a screening tool for early prediction of MCI in patients with cerebrovascular stenosis.

Guided Conditional Diffusion Classifier (ConDiff) for Enhanced Prediction of Infection in Diabetic Foot Ulcers.

Goal: To accurately detect infections in Diabetic Foot Ulcers (DFUs) using photographs taken at the Point of Care (POC). Achieving high performance is critical for preventing complications and amputations, as well as minimizing unnecessary emergency department visits and referrals. Methods: This paper proposes the Guided Conditional Diffusion Classifier (ConDiff). This novel deep-learning framework combines guided image synthesis with a denoising diffusion model and distance-based classification. The process involves (1) generating guided conditional synthetic images by injecting Gaussian noise to a guide (input) image, followed by denoising the noise-perturbed image through a reverse diffusion process, conditioned on infection status and (2) classifying infections based on the minimum Euclidean distance between synthesized images and the original guide image in embedding space. Results: ConDiff demonstrated superior performance with an average accuracy of 81% that outperformed state-of-the-art (SOTA) models by at least 3%. It also achieved the highest sensitivity of 85.4%, which is crucial in clinical domains while significantly improving specificity to 74.4%, surpassing the best SOTA model. Conclusions: ConDiff not only improves the diagnosis of DFU infections but also pioneers the use of generative discriminative models for detailed medical image analysis, offering a promising approach for improving patient outcomes.

A discriminative model for scale, translation and rotation invariant face recognition.

A discriminative model for scale, translation and rotation invariant face recognition.

Can the Discriminative Lexicon Model account for the family size effect in auditory word recognition?

Abstract Words with larger morphological families elicit shorter response times (RTs) in lexical decision experiments (e.g., Bertram et al. 2000). One possible account for this family size (FS) effect draws on the Discriminative Lexicon Model (DLM; Chuang &amp; Baayen 2021), positing that morphological family members strengthen relationships between forms and meanings. While it has been shown that the DLM successfully explains FS effects in reading (Mulder et al. 2014), we investigated whether it does so in listening too. We trained the computational model LDL-AURIS (Shafaei-Bajestan et al. 2023), which implements the DLM, on Dutch and show that a measure derived from LDL-AURIS accounts for variance in auditory lexical decision RTs in Dutch, and also partially accounts for the same variance in the RTs as the auditory FS effect. Future research should investigate whether some other measure derived from the DLM can fully explain FS effects in listening.