Trustworthy Model Research Articles

Enhanced detection of surface deformations in LPBF using deep convolutional neural networks and transfer learning from a porosity model

Our previous research papers have shown the potential of deep-learning models for real-time detection and control of porosity defects in 3D printing, specifically in the laser powder bed fusion (LPBF) process. Extending these models to identify other defects like surface deformation poses a challenge due to the scarcity of available data. This study introduces the use of Transfer Learning (TL) to train models on limited data for high accuracy in detecting surface deformations, marking the first attempt to apply a model trained on one defect type to another. Our approach demonstrates the power of transfer learning in adapting a model known for porosity detection in LPBF to identify surface deformations with high accuracy (94%), matching the performance of the best existing models but with significantly less complexity. This results in faster training and evaluation, ideal for real-time systems with limited computing capabilities. We further employed Gradient-weighted Class Activation Mapping (Grad-CAM) to visualize the model’s decision-making, highlighting the areas influencing defect detection. This step is vital for developing a trustworthy model, showcasing the effectiveness of our approach in broadening the model’s applicability while ensuring reliability and efficiency.

Enhancing Transmission Efficiency in Non-IID Federated Learning by Integrating Quantization into an Existing Trustworthiness Model

Federated Learning (FL) is a distributed machine learning paradigm that enables multiple clients to collaboratively train a model without sharing their data, thus preserving privacy. In this approach, data is not transmitted between the client and the base station. Instead, a model is sent to each device, where it is trained locally, and then retransmitted back to the base station. Each iteration of this process is known as a communication round. However, FL faces challenges, especially when data is not independently and identically distributed (Non-IID). Non-IID data means that the data across clients can vary significantly in distribution, leading to situations where certain classes or features are overrepresented in some clients and underrepresented in others. This lack of uniformity quickly leads to biased model updates and reduced performance. To address this, previous studies have introduced trustworthiness metrics to ensure more reliable model aggregation, minimizing accuracy losses associated with Non-IID data. Another significant challenge is the high transmission load, as model updates between clients and the base station are resource-intensive. Model parameters are transmitted between clients and base stations, which can strain communication channels and slow down the entire process, especially when dealing with larger models and datasets. This communication overhead is a bottleneck that limits the scalability of FL, particularly in resource-constrained environments. Our research addresses these challenges by integrating quantization into the trustworthiness model specifically in a Non-IID scenario. Quantization reduces the precision of model parameters to minimize data transmission, resulting in more efficient communication. We applied different levels of quantization intensity to a model trained on the CIFAR-10 dataset and found that certain methods can significantly reduce transmission overhead without substantial accuracy loss. Our findings suggest that combining quantization with trustworthiness metrics can significantly enhance the efficiency and potentially improve the adoption of Federated Learning in resource-constrained environments.

AutoPeptideML: a study on how to build more trustworthy peptide bioactivity predictors.

Automated machine learning (AutoML) solutions can bridge the gap between new computational advances and their real-world applications by enabling experimental scientists to build their own custom models. We examine different steps in the development life-cycle of peptide bioactivity binary predictors and identify key steps where automation cannot only result in a more accessible method, but also more robust and interpretable evaluation leading to more trustworthy models. We present a new automated method for drawing negative peptides that achieves better balance between specificity and generalization than current alternatives. We study the effect of homology-based partitioning for generating the training and testing data subsets and demonstrate that model performance is overestimated when no such homology correction is used, which indicates that prior studies may have overestimated their performance when applied to new peptide sequences. We also conduct a systematic analysis of different protein language models as peptide representation methods and find that they can serve as better descriptors than a naive alternative, but that there is no significant difference across models with different sizes or algorithms. Finally, we demonstrate that an ensemble of optimized traditional machine learning algorithms can compete with more complex neural network models, while being more computationally efficient. We integrate these findings into AutoPeptideML, an easy-to-use AutoML tool to allow researchers without a computational background to build new predictive models for peptide bioactivity in a matter of minutes. Source code, documentation, and data are available at https://github.com/IBM/AutoPeptideML and a dedicated web-server at http://peptide.ucd.ie/AutoPeptideML. A static version of the software to ensure the reproduction of the results is available at https://zenodo.org/records/13363975.

Advanced sampling discovers apparently similar ankle models with distinct internal load states under minimal parameter modification

Creating valid and trustworthy models is a key issue in biomedical engineering that affects the quality of life of both patients and healthy individuals in various scientific and industrial domains. This however is a difficult task due to the complex nature of biomechanical joints. In this study, a sampling strategy combining Genetic Algorithm and clustering is proposed to investigate biomechanical joints. A computational model of a human ankle joint with 43 input parameters serves as an illustrative case for the procedure. The Genetic Algorithm is used to efficiently search for distinct variants of the model with similar output, while clustering helps to quantify the obtained results. The search is performed in a close vicinity to the original model, mimicking subjective decisions in parameter acquisition. The method reveals twelve distinct clusters in the model parameter set, all resulting in the same angular displacements. These clusters correspond to three unique internal load states for the model, confirming the complex nature of the ankle. The proposed approach is general and could be applied to study other models in mechanical engineering and robotics.

Mass spectrometry-complemented molecular modeling predicts the interaction interface for a camelid single-domain antibody targeting the Plasmodium falciparum circumsporozoite protein’s C-terminal domain

BackgroundBioanalytical methods that enable rapid and high-detail characterization of binding specificities and strengths of protein complexes with low sample consumption are highly desired. The interaction between a camelid single domain antibody (sdAbCSP1) and its target antigen (PfCSP-Cext) was selected as a model system to provide proof-of-principle for the here described methodology. Research design and methodsThe structure of the sdAbCSP1 – PfCSP-Cext complex was modeled using AlphaFold2. The recombinantly expressed proteins, sdAbCSP1, PfCSP-Cext, and the sdAbCSP1 – PfCSP-Cext complex, were subjected to limited proteolysis and mass spectrometric peptide analysis. ITEM MS (Intact Transition Epitope Mapping Mass Spectrometry) and ITC (Isothermal Titration Calorimetry) were applied to determine stoichiometry and binding strength. ResultsThe paratope of sdAbCSP1 mainly consists of its CDR3 (aa100–118). PfCSP-Cext’s epitope is assembled from its α-helix (aa40–52) and opposing loop (aa83–90). PfCSP-Cext’s GluC cleavage sites E46 and E58 were shielded by complex formation, confirming the predicted epitope. Likewise, sdAbCSP1’s tryptic cleavage sites R105 and R108 were shielded by complex formation, confirming the predicted paratope. ITEM MS determined the 1:1 stoichiometry and the high complex binding strength, exemplified by the gas phase dissociation reaction enthalpy of 50.2 kJ/mol. The in-solution complex dissociation constant is 5 × 10-10 M. ConclusionsCombining AlphaFold2 modeling with mass spectrometry/limited proteolysis generated a trustworthy model for the sdAbCSP1 – PfCSP-Cext complex interaction interface.

Learning trustworthy model from noisy labels based on rough set for surface defect detection

In surface defect detection, some regions remain ambiguous and cannot be distinctly classified as abnormal or normal. This challenge is exacerbated by subjective factors, including workers’ emotional fluctuations and judgment variability, resulting in noisy labels that lead to false positives and missed detections. Current methods depend on additional labels, such as clean and multi-labels, which are both time-consuming and labor-intensive. To address this, we utilize Rough Set theory and Bayesian neural networks to learn a trustworthy model from noisy labels for Surface Defect Detection. Our approach features a novel pixel-level representation of suspicious areas using lower and upper approximations, and a novel loss function that emphasizes both precision and recall. The Pluggable Spatially Bayesian Module (PSBM) we developed enhances probabilistic segmentation, effectively capturing uncertainty without requiring extra labels or architectural modifications. Additionally, we have devised a ‘defect discrimination confidence’ metric to better quantify uncertainty and assist in product quality grading. Without the need for extra labeling, our method significantly outperforms state-of-the-art techniques across three types of datasets and enhances seven types of classic networks as a pluggable module, without compromising real-time computing performance. For further details and implementation, our code is accessible at https://github.com/ntongzhi/RoughSet-BNNs.

Revisiting the classic ABI model of trustworthiness: interactive effects of trustworthiness components on trust in mixed-motive social exchange contexts

ABSTRACT Trust is a core precondition for positive outcomes in multiple types of social exchange interactions, and has been successfully predicted by the perceived ability, benevolence, and integrity of a counterpart (i.e. the ABI model; Mayer et al. 1995. However, the specific interplay of these ABI components remains largely unclear, especially in mixed-motive social exchange settings. We propose there is an interplay of the ABI components, expecting a main effect of benevolence moderated by ability and integrity perceptions. In doing so, we refer to basic rationales for interaction effects from existing motivational and social psychological theories such as Expectancy Theory (e.g. Vroom, 1964) or the Stereotype Content Model (Fiske et al., 2002). We tested our predictions in two preregistered experimental studies (N = 1.058) using negotiation vignettes (Study 1) and a multi-round decision game (Study 2). As expected, counterpart benevolence consistently predicted trust and this main effect was moderated by counterpart integrity in Study 1 and by ability in Study 2. Our findings challenge the common main effect assumption for the trustworthiness – trust relationship, provide insights into the mechanisms of trust formation, and are relevant for trust building as well as long-term business relationships.

AI-based 3D-QSAR model of FDA-approved repurposed drugs for inhibiting sclerostin

BACKGROUND: Wnt activation promotes bone formation and prevents bone loss. The Wnt pathway antagonist sclerostin and additional anti-sclerostin antibodies were discovered as a result of the development of the monoclonal antibody romosozumab. These monoclonal antibodies greatly increase the risk of cardiac arrest. Three-dimensional quantitative structure-activity relationships (3D-QSAR) predicts biological activities of ligands based on their three-dimensional features by employing powerful chemometric investigations such as artificial neural networks (ANNs) and partial least squares (PLS). OBJECTIVE: In this study, ligand-receptor interactions were investigated using 3D-QSAR Comparative molecular field analysis (CoMFA). Estimates of steric and electrostatic characteristics in CoMFA are made using Lennard-Jones and Coulomb potentials. METHODS: To identify the conditions necessary for the activity of these molecules, fifty Food and Drug Administration (FDA)-approved medications were chosen for 3D-QSAR investigations and done by CoMFA. For QSAR analysis, there are numerous tools available. This study employed Open 3D-QSAR for analysis due to its simplicity of use and capacity to produce trustworthy results. Four tools were used for the analysis on this platform: Py-MolEdit, Py-ConfSearch, and Py-CoMFA. RESULTS: Maps that were generated were used to determine the screen’s r2 (Coefficient of Multiple Determinations) value and q2 (correlation coefficient). These numbers must be fewer than 1, suggesting a good, trustworthy model. Cross-validated (q2) 0.532 and conventional (r2) correlation values of 0.969 made the CoMFA model statistically significant. The model showed that hydroxamic acid inhibitors are significantly more sensitive to the steric field than the electrostatic field (70%) (30%). This hypothesis states that steric (43.1%), electrostatic (26.4%), and hydrophobic (20.3%) qualities were important in the design of sclerostin inhibitors. CONCLUSION: With 3D-QSAR and CoMFA, statistically meaningful models were constructed to predict ligand inhibitory effects. The test set demonstrated the model’s robustness. This research may aid in the development of more effective sclerostin inhibitors that are synthesised using FDA-approved medications.

Applying Trust Patterns to Model Complex Trustworthiness in the Internet of Things

Key aspects of communities of the Internet of Things (IoT) smart objects presenting social aspects are represented by trust and reputation relationships between the objects. Several trustworthiness models have been presented in the literature in the context of multi-smart object community that could be adopted in the IoT scenario; however, most of these approaches represent the different dimensions of trust using scalar measures, then integrating these measures in a global trustworthiness value. In this paper, we discuss the limitation of this approach in the IoT context, highlighting the necessity of modeling complex trust relationships that cannot be captured by a vector-based model, and we propose a new trust model in which the trust perceived by an object with respect to another object is modeled by a directed, weighted graph whose vertices are trust dimensions and whose arcs represent relationships between trust dimensions. By using this new model, we provide the IoT community with the possibility of representing also situations in which an object does not know a trust dimension, e.g., reliability, but it is able to derive it from another one, e.g., honesty. The introduced model can represent any trust structure of the type illustrated above, in which several trust dimensions are mutually dependent.

IMPA-Net: Interpretable Multi-Part Attention Network for Trustworthy Brain Tumor Classification from MRI.

Deep learning (DL) networks have shown attractive performance in medical image processing tasks such as brain tumor classification. However, they are often criticized as mysterious "black boxes". The opaqueness of the model and the reasoning process make it difficult for health workers to decide whether to trust the prediction outcomes. In this study, we develop an interpretable multi-part attention network (IMPA-Net) for brain tumor classification to enhance the interpretability and trustworthiness of classification outcomes. The proposed model not only predicts the tumor grade but also provides a global explanation for the model interpretability and a local explanation as justification for the proffered prediction. Global explanation is represented as a group of feature patterns that the model learns to distinguish high-grade glioma (HGG) and low-grade glioma (LGG) classes. Local explanation interprets the reasoning process of an individual prediction by calculating the similarity between the prototypical parts of the image and a group of pre-learned task-related features. Experiments conducted on the BraTS2017 dataset demonstrate that IMPA-Net is a verifiable model for the classification task. A percentage of 86% of feature patterns were assessed by two radiologists to be valid for representing task-relevant medical features. The model shows a classification accuracy of 92.12%, of which 81.17% were evaluated as trustworthy based on local explanations. Our interpretable model is a trustworthy model that can be used for decision aids for glioma classification. Compared with black-box CNNs, it allows health workers and patients to understand the reasoning process and trust the prediction outcomes.

Initial development of perceptions of ability and intent factors of (un)trustworthiness in short‐term teams

SummaryIndividual perceptions of team trustworthiness are critical antecedents to the development of relationships within teams. Yet, clarity is needed regarding the focus of these individual perceptions and how they change, particularly at the outset of team interactions. Two distinct research streams have emerged regarding trustworthiness. One proposes a difference between perceived ability and intent, whereas the other proposes a difference between confident positive and negative expectations of others. Regardless of their structure, theory suggests that perceptions of team trustworthiness ought to develop across performance episodes as individuals identify more with their team and have more information about how they are performing. We explore these relationships across three distinct performance episodes in newly formed short‐term teams. Our results support a four‐factor model of trustworthiness within the context of short‐term, lab‐based teams. Further, perceptions regarding teammates' abilities tended to change quickly and curvilinearly while those regarding teammates' intents changed linearly. Positive team performance signals bolstered individual perceptions of team trustworthiness in ability and intent but not perceived team untrustworthiness. Negative team performance signals bolstered individual perceptions of team untrustworthiness in ability and intent while harming both types of individual perceptions of team trustworthiness. This study supports propositions from the model of trust over time and lays the groundwork for a comprehensive approach toward trust research within teams.

Manufacturing the Illusion of Epistemic Trustworthiness

Abstract There are epistemic manipulators in the world. These people are actively attempting to sacrifice epistemic goods for personal gain. In doing so, manipulators have led many competent epistemic agents into believing contrarian theories that go against well-established knowledge. In this paper, I explore one mechanism by which manipulators get epistemic agents to believe contrarian theories. I do so by looking at a prominent empirical model of trustworthiness. This model identifies three major factors that epistemic agents look for when trying to determine who is trustworthy. These are (i) ability, (ii) benevolence, and (iii) moral integrity. I then show how manipulators can manufacture the illusion that they possess these factors. This leads epistemic agents to view manipulators as trustworthy sources of information. Additionally, I argue that fact-checking will be an ineffective – or even harmful – practice when correcting the beliefs of epistemic agents who have been tricked by this illusion of epistemic trustworthiness. I suggest that in such cases we should use an alternative correction, which I call trust undercutting.

TICPS: A trustworthy collaborative intrusion detection framework for industrial cyber–physical systems

The networking of industrial cyber–physical systems (CPS) introduces increased security vulnerabilities, necessitating advanced intrusion detection systems (IDS). Many current studies aiming to enhance IDS capabilities leverage Federated Learning (FL) technology for collaborative intrusion detection. However, devices deployed in an industrial setting in a distributed manner are vulnerable to cyber and poisoning attacks. Compromised clients can create malicious parameters to disrupt intrusion detection models, making them ineffective in identifying attacks. Nevertheless, existing FL-based intrusion detection methods exhibit suboptimal performance in detecting malicious clients and resisting poisoning attacks. To address these issues, we propose TICPS, a collaborative intrusion detection framework based on a trustworthy model update strategy to detect cyber threats from industrial CPS. The framework enables multiple industrial CPS to collaboratively construct an intrusion detection model and evaluate the security of each industrial CPS node using an update evaluation mechanism, ensuring effective intrusion detection even in the presence of poisoning. Extensive experiments on real-world industrial CPS datasets demonstrate that TICPS can effectively detect various types of cyber threats targeting industrial CPS. In particular, the framework achieves an intrusion detection accuracy of 94% even when the proportion of malicious agents reaches 80% under three typical poisoning attacks.

Evaluating the robustness of multimodal task load estimation models

Numerous studies have focused on constructing multimodal machine learning models for estimating a person's cognitive load. However, a prevalent limitation is that these models are typically evaluated on data from the same scenario they were trained on. Little attention has been given to their robustness against data distribution shifts, which may occur during deployment. The aim of this paper is to investigate the performance of these models when confronted with a scenario different from the one on which they were trained. For this evaluation, we utilized a dataset encompassing two distinct scenarios: an n-Back test and a driving simulation. We selected a variety of classic machine learning and deep learning architectures, which were further complemented by various fusion techniques. The models were trained on the data from the n-Back task and tested on both scenarios to evaluate their predictive performance. However, the predictive performance alone may not lead to a trustworthy model. Therefore, we looked at the uncertainty estimates of these models. By leveraging these estimates, we can reduce misclassification by resorting to alternative measures in situations of high uncertainty. The findings indicate that late fusion produces stable classification results across the examined models for both scenarios, enhancing robustness compared to feature-based fusion methods. Although a simple logistic regression tends to provide the best predictive performance for n-Back, this is not always the case if the data distribution is shifted. Finally, the predictive performance of individual modalities differs significantly between the two scenarios. This research provides insights into the capabilities and limitations of multimodal machine learning models in handling distribution shifts and identifies which approaches may potentially be suitable for achieving robust results.

The pre-trained explainable deep learning model with stacked denoising autoencoders for slope stability analysis

In this work, we proposed a deeply-integrated explainable pre-trained deep learning framework with stacked denoising autoencoders in the assessment of slope stability. The deep learning model consists of a deep neural network as a trunk net for prediction and autoencoders as branch nets for denoising. A comprehensive review of machine learning algorithms in slope stability evaluation is first given in the introduction section. A series of 530 data is then collected from real slope records, which are visualized and investigated in feature engineering and further preprocessed for model training. To ensure reliable and trustworthy model interpretability, a unified model from both local and global perspectives is integrated into the deep learning model, which incorporated the ad hoc back-propagation based Deep SHAP, perturbation based Kernel SHAP and PDPs, and distillation based LIME and Anchors. For a fair evaluation, repeated stratified 10-fold cross-validation is adopted in model evaluation. The obtained results manifest that the constructed model outperforms commonly used machine learning methods in terms of accuracy and stability on the real-world slope data. The explainable model provides a reasonable explanation and validates the capability of the proposed model, and reflects the causes and dependencies of model predictions for a given sample.

U-trustworthy Models. Reliability, Competence, and Confidence in Decision-Making

With growing concerns regarding bias and discrimination in predictive models, the AI community has increasingly focused on assessing AI system trustworthiness. Conventionally, trustworthy AI literature relies on the probabilistic framework and calibration as prerequisites for trustworthiness. In this work, we depart from this viewpoint by proposing a novel trust framework inspired by the philosophy literature on trust. We present a precise mathematical definition of trustworthiness, termed U-trustworthiness, specifically tailored for a subset of tasks aimed at maximizing a utility function. We argue that a model’s U-trustworthiness is contingent upon its ability to maximize Bayes utility within this task subset. Our first set of results challenges the probabilistic framework by demonstrating its potential to favor less trustworthy models and introduce the risk of misleading trustworthiness assessments. Within the context of U-trustworthiness, we prove that properly-ranked models are inherently U-trustworthy. Furthermore, we advocate for the adoption of the AUC metric as the preferred measure of trustworthiness. By offering both theoretical guarantees and experimental validation, AUC enables robust evaluation of trustworthiness, thereby enhancing model selection and hyperparameter tuning to yield more trustworthy outcomes.

A Blockchain-based federated learning framework for secure aggregation and fair incentives

Federated Learning (FL) has gained prominence as a machine learning framework incorporating privacy-preserving mechanisms. However, challenges such as poisoning attacks and free rider attacks underscore the need for advanced security measures. Therefore, this paper proposes a novel framework that integrates federated learning with blockchain technology to facilitate secure model aggregation and fair incentives in untrustworthy environments. The framework designs a reputation evaluation method using quality as an indicator, and a consensus method based on reputation feedback. The trustworthiness of nodes is dynamically assessed to achieve an efficient and trustworthy model aggregation process while avoiding reputation monopolisation. Furthermore, the paper defines a tailored contribution calculation process for nodes in different roles in an untrusted environment. A reward and punishment scheme based on the joint constraints of contribution and reputation is proposed to attract highly qualified workers to actively participate in federated learning tasks. Theoretical analysis and simulation experiments demonstrate the framework's ability to maintain efficient and secure aggregation under a certain degree of attack while achieving fair incentives for each role node with significantly reduced consensus consumption.

Trustworthiness modeling and evaluation for a nearly autonomous management and control system

The Nearly Autonomous Management and Control (NAMAC) system supports the advanced reactor operation by recommending control actions to operators based on real-time measurements and digital twins (DTs) learning from the knowledge base. To enable the safe and reliable use of autonomous technologies, NAMAC and its recommendations should be trustworthy to operators and regulators at both the design and operation stages. This study proposes a NAMAC trustworthiness modeling and evaluation framework supported by trustworthiness ontologies and evidence-based approaches. The development-time and run-time ontologies are separately constructed and then converted to Bayesian networks to quantitatively evaluate the NAMAC trustworthiness. This evaluation is demonstrated by collecting and characterizing evidence from NAMAC practices, such as the development and assessment of the NAMAC system, data coverage assessment, and the training and optimizations of neural-network-based DTs. Our proposed approach can aggregate various trustworthiness attributes of complex artificial-intelligence-supported systems for safety-critical applications. It also considers the interaction between different DTs and extends beyond the trustworthiness evaluation of a single DT. The evidence-based method enhances the transparency of the trustworthiness modeling and evaluation processes and helps identify uncertainties and subjectivity involved in the processes.

Racial Prejudice Affects Representations of Facial Trustworthiness

What makes faces seem trustworthy? We investigated how racial prejudice predicts the extent to which perceivers employ racially prototypical cues to infer trustworthiness from faces. We constructed participant-level computational models of trustworthiness and White-to-Black prototypicality from U.S. college students’ judgments of White (Study 1, N = 206) and Black–White morphed (Study 3, N = 386) synthetic faces. Although the average relationships between models differed across stimuli, both studies revealed that as participants’ anti-Black prejudice increased and/or intergroup contact decreased, so too did participants’ tendency to conflate White prototypical features with trustworthiness and Black prototypical features with untrustworthiness. Study 2 (N = 324) and Study 4 (N = 397) corroborated that untrustworthy faces constructed from participants with pro-White preferences appeared more Black prototypical to naive U.S. adults, relative to untrustworthy faces modeled from other participants. This work highlights the important role of racial biases in shaping impressions of facial trustworthiness.

Trust in algorithmic decision-making systems in health: A comparison between ADA health and IBM Watson.

Algorithmic decision-making systems (ADMs) support an ever-growing number of decision-making processes. We conducted an online survey study in Flanders (n = 1,082) to understand how laypeople perceive and trust health ADMs. Inspired by the ability, benevolence, and integrity trustworthiness model (Mayer et al., 1995), this study investigated how trust is constructed in health ADMs. In addition, we investigated how trust construction differs between ADA Health (a self-diagnosis medical chatbot) and IBM Watson Oncology (a system that suggests treatments for cancer in hospitals). Our results show that accuracy and fairness are the biggest predictors of trust in both ADMs, whereas control plays a smaller yet significant role. Interestingly, control plays a bigger role in explaining trust in ADA Health than IBM Watson Oncology. Moreover, how appropriate people evaluate data-driven healthcare and how concerned they are with algorithmic systems prove to be good predictors for accuracy, fairness, and control in these specific health ADMs. The appropriateness of data-driven healthcare had a bigger effect with IBM Watson Oncology than with ADA Health. Overall, our results show the importance of considering the broader contextual, algorithmic, and case-specific characteristics when investigating trust construction in ADMs.