Exploring the categorical nature of colour perception: Insights from artificial networks.

The electromagnetic spectrum of light from a rainbow is a continuous signal, yet we perceive it vividly in several distinct colour categories. The origins and underlying mechanisms of this phenomenon remain partly unexplained. We investigate categorical colour perception in artificial neural networks (ANNs) using the odd-one-out paradigm. In the first experiment, we compared unimodal vision networks (e.g., ImageNet object recognition) to multimodal vision-language models (e.g., CLIP text-image matching). Our results show that vision networks predict a significant portion of human data (approximately 80%), while vision-language models account for the remaining unexplained data, even in non-linguistic experiments. These findings suggest that categorical colour perception is a language-independent representation, though it is partly shaped by linguistic colour terms during its development. In the second experiment, we explored how the visual task influences the colour categories of an ANN by examining twenty-four Taskonomy networks. Our results indicate that human-like colour categories are task-dependent, predominantly emerging in semantic and 3D tasks, with a notable absence in low-level tasks. To explain this difference, we analysed kernel responses before the winner-takes-all stage, observing that networks with mismatching colour categories may still align in underlying continuous representations. Our findings quantify the dual influence of visual signals and linguistic factors in categorical colour perception and demonstrate the task-dependent nature of this phenomenon, suggesting that categorical colour perception emerges to facilitate certain visual tasks.

Ptychographic phase retrieval via a deep-learning-assisted iterative algorithm.

Ptychography is a powerful computational imaging technique with microscopic imaging capability and adaptability to various specimens. To obtain an imaging result, it requires a phase-retrieval algorithm whose performance directly determines the imaging quality. Recently, deep neural network (DNN)-based phase retrieval has been proposed to improve the imaging quality from the ordinary model-based iterative algorithms. However, the DNN-based methods have some limitations because of the sensitivity to changes in experimental conditions and the difficulty of collecting enough measured specimen images for training the DNN. To overcome these limitations, a ptychographic phase-retrieval algorithm that combines model-based and DNN-based approaches is proposed. This method exploits a DNN-based denoiser to assist an iterative algorithm like ePIE in finding better reconstruction images. This combination of DNN and iterative algorithms allows the measurement model to be explicitly incorporated into the DNN-based approach, improving its robustness to changes in experimental conditions. Furthermore, to circumvent the difficulty of collecting the training data, it is proposed that the DNN-based denoiser be trained without using actual measured specimen images but using a formula-driven supervised approach that systemically generates synthetic images. In experiments using simulation based on a hard X-ray ptychographic measurement system, the imaging capability of the proposed method was evaluated by comparing it with ePIE and rPIE. These results demonstrated that the proposed method was able to reconstruct higher-spatial-resolution images with half the number of iterations required by ePIE and rPIE, even for data with low illumination intensity. Also, the proposed method was shown to be robust to its hyperparameters. In addition, the proposed method was applied to ptychographic datasets of a Simens star chart and ink toner particles measured at SPring-8 BL24XU, which confirmed that it can successfully reconstruct images from measurement scans with a lower overlap ratio of the illumination regions than is required by ePIE and rPIE.

MRI-based deep learning for differentiating between bipolar and major depressive disorders.

Mood disorders, particularly bipolar disorder (BD) and major depressive disorder (MDD), manifest changes in brain structure that can be detected using structural magnetic resonance imaging (MRI). Although structural MRI is a promising diagnostic tool, prevailing diagnostic criteria for BD and MDD are predominantly subjective, sometimes leading to misdiagnosis. This challenge is compounded by a limited understanding of the underlying causes of these disorders. In response, we present SE-ResNet, a Residual Network (ResNet)-based framework designed to discriminate between BD, MDD, and healthy controls (HC) using structural MRI data. Our approach extends the traditional Squeeze-and-Excitation (SE) layer by incorporating a dedicated branch for spatial attention map generation, equipped with soft-pooling, a 7 × 7 convolution, and a sigmoid function, intended to detect complex spatial patterns. The fusion of channel and spatial attention maps through element-wise addition aims to enhance the model's ability to discriminate features. Unlike conventional methods that use max-pooling for downsampling, our methodology employs soft-pooling, which aims to preserve a richer representation of input features and reduce data loss. When evaluated on a proprietary dataset comprising 303 subjects, the SE-ResNet achieved an accuracy of 85.8 %, a recall of 85.7 %, a precision of 85.9 %, and an F1 score of 85.8 %. These performance metrics suggest that the SE-ResNet framework has potential as a tool for detecting psychiatric disorders using structural MRI data.

Uncertainty quantification in multi-class image classification using chest X-ray images of COVID-19 and pneumonia.

This paper investigates uncertainty quantification (UQ) techniques in multi-class classification of chest X-ray images (COVID-19, Pneumonia, and Normal). We evaluate Bayesian Neural Networks (BNN) and the Deep Neural Network with UQ (DNN with UQ) techniques, including Monte Carlo dropout, Ensemble Bayesian Neural Network (EBNN), Ensemble Monte Carlo (EMC) dropout, across different evaluation metrics. Our analysis reveals that DNN with UQ, especially EBNN and EMC dropout, consistently outperform BNNs. For example, in Class 0 vs. All, EBNN achieved a UAcc of 92.6%, UAUC-ROC of 95.0%, and a Brier Score of 0.157, significantly surpassing BNN's performance. Similarly, EMC Dropout excelled in Class 1 vs. All with a UAcc of 83.5%, UAUC-ROC of 95.8%, and a Brier Score of 0.165. These advanced models demonstrated higher accuracy, better discriaminative capability, and more accurate probabilistic predictions. Our findings highlight the efficacy of DNN with UQ in enhancing model reliability and interpretability, making them highly suitable for critical healthcare applications like chest X-ray imageQ6 classification.

ASCHOPLEX: A generalizable approach for the automatic segmentation of choroid plexus.

BACKGROUND: The Choroid Plexus (ChP) plays a vital role in brain homeostasis, serving as part of the Blood-Cerebrospinal Fluid Barrier, contributing to brain clearance pathways and being the main source of cerebrospinal fluid. Since the involvement of ChP in neurological and psychiatric disorders is not entirely established and currently under investigation, accurate and reproducible segmentation of this brain structure on large cohorts remains challenging. This paper presents ASCHOPLEX, a deep-learning tool for the automated segmentation of human ChP from structural MRI data that integrates existing software architectures like 3D UNet, UNETR, and DynUNet to deliver accurate ChP volume estimates. METHODS: Here we trained ASCHOPLEX on 128 T1-w MRI images comprising both controls and patients with Multiple Sclerosis. ASCHOPLEX's performances were evaluated using traditional segmentation metrics; manual segmentation by experts served as ground truth. To overcome the generalizability problem that affects data-driven approaches, an additional fine-tuning procedure (ASCHOPLEXtune) was implemented on 77 T1-w PET/MRI images of both controls and depressed patients. RESULTS: ASCHOPLEX showed superior performance compared to commonly used methods like FreeSurfer and Gaussian Mixture Model both in terms of Dice Coefficient (ASCHOPLEX 0.80, ASCHOPLEXtune 0.78) and estimated ChP volume error (ASCHOPLEX 9.22%, ASCHOPLEXtune 9.23%). CONCLUSION: These results highlight the high accuracy, reliability, and reproducibility of ASCHOPLEX ChP segmentations.

Improving synthetic media generation and detection using generative adversarial networks.

Synthetic images ar---e created using computer graphics modeling and artificial intelligence techniques, referred to as deepfakes. They modify human features by using generative models and deep learning algorithms, posing risks violations of social media regulations and spread false information. To address these concerns, the study proposed an improved generative adversarial network (GAN) model which improves accuracy while differentiating between real and fake images focusing on data augmentation and label smoothing strategies for GAN training. The study utilizes a dataset containing human faces and employs DCGAN (deep convolutional generative adversarial network) as the base model. In comparison with the traditional GANs, the proposed GAN outperform in terms of frequently used metrics i.e., Fréchet Inception Distance (FID) and accuracy. The model effectiveness is demonstrated through evaluation on the Flickr-Faces Nvidia dataset and Fakefaces d--ataset, achieving an FID score of 55.67, an accuracy of 98.82%, and an F1-score of 0.99 in detection. This study optimizes the model parameters to achieve optimal parameter settings. This study fine-tune the model parameters to reach optimal settings, thereby reducing risks in synthetic image generation. The article introduces an effective framework for both image manipulation and detection.

Model and Method for Providing Resilience to Resource-Constrained AI-System.

Artificial intelligence technologies are becoming increasingly prevalent in resource-constrained, safety-critical embedded systems. Numerous methods exist to enhance the resilience of AI systems against disruptive influences. However, when resources are limited, ensuring cost-effective resilience becomes crucial. A promising approach for reducing the resource consumption of AI systems during test-time involves applying the concepts and methods of dynamic neural networks. Nevertheless, the resilience of dynamic neural networks against various disturbances remains underexplored. This paper proposes a model architecture and training method that integrate dynamic neural networks with a focus on resilience. Compared to conventional training methods, the proposed approach yields a 24% increase in the resilience of convolutional networks and a 19.7% increase in the resilience of visual transformers under fault injections. Additionally, it results in a 16.9% increase in the resilience of convolutional network ResNet-110 and a 21.6% increase in the resilience of visual transformer DeiT-S under adversarial attacks, while saving more than 30% of computational resources. Meta-training the neural network model improves resilience to task changes by an average of 22%, while achieving the same level of resource savings.

Predicting Adherence to Computer-Based Cognitive Training Programs Among Older Adults: Study of Domain Adaptation and Deep Learning.

Background: Cognitive impairment and dementia pose a significant challenge to the aging population, impacting the well-being, quality of life, and autonomy of affected individuals. As the population ages, this will place enormous strain on health care and economic systems. While computerized cognitive training programs have demonstrated some promise in addressing cognitive decline, adherence to these interventions can be challenging. Objective: The objective of this study is to improve the accuracy of predicting adherence lapses to ultimately develop tailored adherence support systems to promote engagement with cognitive training among older adults. Methods: Data from 2 previously conducted cognitive training intervention studies were used to forecast adherence levels among older participants. Deep convolutional neural networks were used to leverage their feature learning capabilities and predict adherence patterns based on past behavior. Domain adaptation (DA) was used to address the challenge of limited training data for each participant, by using data from other participants with similar playing patterns. Time series data were converted into image format using Gramian angular fields, to facilitate clustering of participants during DA. To the best of our knowledge, this is the first effort to use DA techniques to predict older adults' daily adherence to cognitive training programs. Results: Our results demonstrated the promise and potential of deep neural networks and DA for predicting adherence lapses. In all 3 studies, using 2 independent datasets, DA consistently produced the best accuracy values. Conclusions: Our findings highlight that deep learning and DA techniques can aid in the development of adherence support systems for computerized cognitive training, as well as for other interventions aimed at improving health, cognition, and well-being. These techniques can improve engagement and maximize the benefits of such interventions, ultimately enhancing the quality of life of individuals at risk for cognitive impairments. This research informs the development of more effective interventions, benefiting individuals and society by improving conditions associated with aging.

Deep neural network models for cell type prediction based on single-cell Hi-C data.

BACKGROUND: Cell type prediction is crucial to cell type identification of genomics, cancer diagnosis and drug development, and it can solve the time-consuming and difficult problem of cell classification in biological experiments. Therefore, a computational method is urgently needed to classify and predict cell types using single-cell Hi-C data. In previous studies, there is a lack of convenient and accurate method to predict cell types based on single-cell Hi-C data. Deep neural networks can form complex representations of single-cell Hi-C data and make it possible to handle the multidimensional and sparse biological datasets. RESULTS: We compare the performance of SCANN with existing methods and analyze the model by using five different evaluation metrics. When using only ML1 and ML3 datasets, the ARI and NMI values of SCANN increase by 14% and 11% over those of scHiCluster respectively. However, when using all six libraries of data, the ARI and NMI values of SCANN increase by 63% and 88% over those of scHiCluster respectively. These findings show that SCANN is highly accurate in predicting the type of independent cell samples using single-cell Hi-C data. CONCLUSIONS: SCANN enhances the training speed and requires fewer resources for predicting cell types. In addition, when the number of cells in different cell types was extremely unbalanced, SCANN has higher stability and flexibility in solving cell classification and cell type prediction using the single-cell Hi-C data. This predication method can assist biologists to study the differences in the chromosome structure of cells between different cell types.

PAFE: A lightweight visualization-based fast malware classification method.

With the development of automated malware toolkits, cybersecurity faces evolving threats. Although visualization-based malware analysis has proven to be an effective method, existing approaches struggle with challenging malware samples due to alterations in the texture features of binary images during the visualization preprocessing stage, resulting in poor performance. Furthermore, to enhance classification accuracy, existing methods sacrifice prediction time by designing deeper neural network architectures. This paper proposes PAFE, a lightweight and visualization-based rapid malware classification method. It addresses the issue of texture feature variations in preprocessing through pixel-filling techniques and applies data augmentation to overcome the challenges of class imbalance in small sample datasets. PAFE combines multi-scale feature fusion and a channel attention mechanism, enhancing feature expression through modular design. Extensive experimental results demonstrate that PAFE outperforms the current state-of-the-art methods in both efficiency and effectiveness for malware variant classification, achieving an accuracy rate of 99.25 % with a prediction time of 10.04 ms.

Bridging auditory perception and natural language processing with semantically informed deep neural networks.

Sound recognition is effortless for humans but poses a significant challenge for artificial hearing systems. Deep neural networks (DNNs), especially convolutional neural networks (CNNs), have recently surpassed traditional machine learning in sound classification. However, current DNNs map sounds to labels using binary categorical variables, neglecting the semantic relations between labels. Cognitive neuroscience research suggests that human listeners exploit such semantic information besides acoustic cues. Hence, our hypothesis is that incorporating semantic information improves DNN's sound recognition performance, emulating human behaviour. In our approach, sound recognition is framed as a regression problem, with CNNs trained to map spectrograms to continuous semantic representations from NLP models (Word2Vec, BERT, and CLAP text encoder). Two DNN types were trained: semDNN with continuous embeddings and catDNN with categorical labels, both with a dataset extracted from a collection of 388,211 sounds enriched with semantic descriptions. Evaluations across four external datasets, confirmed the superiority of semantic labeling from semDNN compared to catDNN, preserving higher-level relations. Importantly, an analysis of human similarity ratings for natural sounds, showed that semDNN approximated human listener behaviour better than catDNN, other DNNs, and NLP models. Our work contributes to understanding the role of semantics in sound recognition, bridging the gap between artificial systems and human auditory perception.

Near-optimal deep neural network approximation for Korobov functions with respect to Lp and H1 norms.

This paper derives the optimal rate of approximation for Korobov functions with deep neural networks in the high dimensional hypercube with respect to Lp-norms and H1-norm. Our approximation bounds are non-asymptotic in both the width and depth of the networks. The obtained approximation rates demonstrate a remarkable super-convergence feature, improving the existing convergence rates of neural networks that are continuous function approximators. Finally, using a VC-dimension argument, we show that the established rates are near-optimal.

Intermediate-grained kernel elements pruning with structured sparsity.

Neural network pruning provides a promising prospect for the deployment of neural networks on embedded or mobile devices with limited resources. Although current structured strategies are unconstrained by specific hardware architecture in the phase of forward inference, the decline in classification accuracy of structured methods is beyond the tolerance at the level of general pruning rate. This inspires us to develop a technique that satisfies high pruning rate with a small decline in accuracy and has the general nature of structured pruning. In this paper, we propose a new pruning method, namely KEP (Kernel Elements Pruning), to compress deep convolutional neural networks by exploring the significance of elements in each kernel plane and removing unimportant elements. In this method, we apply a controllable regularization penalty to constrain unimportant elements by adding a prior knowledge mask and obtain a compact model. In the calculation procedure of forward inference, we introduce a sparse convolution operation which is different from the sliding window to eliminate invalid zero calculations and verify the effectiveness of the operation for further deployment on FPGA. A massive variety of experiments demonstrate the effectiveness of KEP on two datasets: CIFAR-10 and ImageNet. Specially, with few indexes of non-zero weights introduced, KEP has a significant improvement over the latest structured methods in terms of parameter and float-point operation (FLOPs) reduction, and performs well on large datasets.

Remote Assessment of Eczema Severity via AI-powered Skin Image Analytics: A Systematic Review.

Various studies have been published on the remote assessment of eczema severity from digital camera images. Successful deployment of an accurate and robust AI-powered tool for such purposes can aid the formulation of eczema treatment plans and assist in patient monitoring. This review aims to provide an overview of the quality of published studies on this topic and to identify challenges and suggestions to improve the robustness and reliability of existing tools. We identified 25 articles from the Scopus database that aimed to assess eczema severity automatically from digital camera images by eczema area detection (n=13), which is important for prior delineation of the most relevant clinical features, and/or severity prediction (n=12). Deep learning methods (n=14) were more commonly used in recent years over conventional machine learning (n=11). A set of 20 pre-defined criteria were used for critical appraisal in this study. Study quality was hindered in many cases due to dataset challenges, with only 28% of studies reporting patient age range and 16% reporting skin phototype range. Furthermore, 52% of studies utilised solely non-public datasets and only 17% provided open-source access to code repositories, making validation of experimental results a significant challenge. In terms of algorithm design, attempts to improve model accuracy and process automation are widely reported. However, there remains limited implementation of methods for explicitly improving model trustworthiness and robustness. There is a need for a high-quality dataset with a sufficient number of bias-free images and consistent labels, as well as improved image analytics methods, to enhance the state of remote eczema severity assessment algorithms. Improving the interpretability and explainability of developed tools will further improve long-term reliability and trustworthiness.

Synthesis of higher-B0 CEST Z-spectra from lower-B0 data via deep learning and singular value decomposition.

Chemical exchange saturation transfer (CEST) MRI at 3 T suffers from low specificity due to overlapping CEST effects from multiple metabolites, while higher field strengths (B0) allow for better separation of Z-spectral "peaks," aiding signal interpretation and quantification. However, data acquisition at higher B0 is restricted by equipment access, field inhomogeneity and safety issues. Herein, we aim to synthesize higher-B0 Z-spectra from readily available data acquired with 3 T clinical scanners using a deep learning framework. Trained with simulation data using models based on Bloch-McConnell equations, this framework comprised two deep neural networks (DNNs) and a singular value decomposition (SVD) module. The first DNN identified B0 shifts in Z-spectra and aligned them to correct frequencies. After B0 correction, the lower-B0 Z-spectra were streamlined to the second DNN, casting into the key feature representations of higher-B0 Z-spectra, obtained through SVD truncation. Finally, the complete higher-B0 Z-spectra were recovered from inverse SVD, given the low-rank property of Z-spectra. This study constructed and validated two models, a phosphocreatine (PCr) model and a pseudo-in-vivo one. Each experimental dataset, including PCr phantoms, egg white phantoms, and in vivo rat brains, was sequentially acquired on a 3 T human and a 9.4 T animal scanner. Results demonstrated that the synthetic 9.4 T Z-spectra were almost identical to the experimental ground truth, showing low RMSE (0.11% ± 0.0013% for seven PCr tubes, 1.8% ± 0.2% for three egg white tubes, and 0.79% ± 0.54% for three rat slices) and high R2 (>0.99). The synthesized amide and NOE contrast maps, calculated using the Lorentzian difference, were also well matched with the experiments. Additionally, the synthesis model exhibited robustness to B0 inhomogeneities, noise, and other acquisition imperfections. In conclusion, the proposed framework enables synthesis of higher-B0 Z-spectra from lower-B0 ones, which may facilitate CEST MRI quantification and applications.

Face Omron Ring: Proactive defense against face forgery with identity awareness.

In the era of Artificial Intelligence Generated Content (AIGC), face forgery models pose significant security threats. These models have caused widespread negative impacts through the creation of forged products targeting public figures, national leaders, and other Persons-of-interest (POI). To address this, we propose the Face Omron Ring (FOR) to proactively protect the POI from face forgery. Specifically, by introducing FOR into a target face forgery model, the model will proactively refuse to forge any face image of protected identities without compromising the forgery capability for unprotected ones. We conduct extensive experiments on 4 face forgery models, StarGAN, AGGAN, AttGAN, and HiSD on the widely used large-scale face image datasets CelebA, CelebA-HQ, and PubFig83. Our results demonstrate that the proposed method can effectively protect 5000 different identities with a 100% protection success rate, for each of which only about 100 face images are needed. Our method also shows great robustness against multiple image processing attacks, such as JPEG, cropping, noise addition, and blurring. Compared to existing proactive defense methods, our method offers identity-centric protection for any image of the protected identity without requiring any special preprocessing, resulting in improved scalability and security. We hope that this work can provide a solution for responsible AIGC companies in regulating the use of face forgery models.

Hyperparameter optimization: Classics, acceleration, online, multi-objective, and tools.

Hyperparameter optimization (HPO) has been well-developed and evolved into a well-established research topic over the decades. With the success and wide application of deep learning, HPO has garnered increased attention, particularly within the realm of machine learning model training and inference. The primary objective is to mitigate the challenges associated with manual hyperparameter tuning, which can be ad-hoc, reliant on human expertise, and consequently hinders reproducibility while inflating deployment costs. Recognizing the growing significance of HPO, this paper surveyed classical HPO methods, approaches for accelerating the optimization process, HPO in an online setting (dynamic algorithm configuration, DAC), and when there is more than one objective to optimize (multi-objective HPO). Acceleration strategies were categorized into multi-fidelity, bandit-based, and early stopping; DAC algorithms encompassed gradient-based, population-based, and reinforcement learning-based methods; multi-objective HPO can be approached via scalarization, metaheuristics, and model-based algorithms tailored for multi-objective situation. A tabulated overview of popular frameworks and tools for HPO was provided, catering to the interests of practitioners.

Proton-Driven Dynamic Behavior of Nanoconfined Water in Hydrophilic MXene Sheets.

Liquid water under nanoscale confinement has attracted intensive attention due to its pivotal role in understanding various phenomena across many scientific fields. MXenes serve an ideal paradigm for investigating the dynamic behaviors of nanoconfined water in a hydrophilic environment. Combining deep neural networks and an active learning scheme, here we elucidate the proton-driven dynamics of water molecules confined between V2CTx sheets using molecular dynamics simulation. Firstly, we have found that the Eigen and Zundel cations can inhibit water-induced oxidation by adjusting the orientation of water molecules, thus proposing a general antioxidant strategy. Besides, we also identified a hexagonal ice phase with abnormal bonding rules at room temperature, rather than only at ultralow temperatures as other studies reported, and further captured the proton-induced water phase transition. This highlighted the importance of protons in the maintaining stable crystal phase and phase transition of water. Furthermore, we discussed the conversions of different water structures and water diffusivity with changing proton concentrations in detail. The results provide useful guidance in practical applications of MXenes including developing antioxidant strategies, identifying novel 2D water phases and optimizing energy storage and conversion.

Analyzing and identifying predictable time range for stress prediction based on chaos theory and deep learning.

Propose: Stress is a common problem globally. Prediction of stress in advance could help people take effective measures to manage stress before bad consequences occur. Considering the chaotic features of human psychological states, in this study, we integrate deep learning and chaos theory to address the stress prediction problem. Methods: Based on chaos theory, we embed one's seemingly disordered stress sequence into a high dimensional phase space so as to reveal the underlying dynamics and patterns of the stress system, and meanwhile are able to identify the stress predictable time range. We then conduct deep learning with a two-layer (dimension and temporal) attention mechanism to simulate the nonlinear state of the embedded stress sequence for stress prediction. Results: We validate the effectiveness of the proposed method on the public available Tesserae dataset. The experimental results show that the proposed method outperforms the pure deep learning method and Chaos method in both 2-label and 3-label stress prediction. Conclusion: Integrating deep learning and chaos theory for stress prediction is effective, and can improve the prediction accuracy over 2% and 8% more than those of the deep learning and the Chaos method respectively. Implications and further possible improvements are also discussed at the end of the paper.