A Neural Network Model of Visual Attention Integrating Biased Competition and Reinforcement Learning DOI Open Access
Jonathan H. Morgan, Badr F. Albanna, James P. Herman

и другие.

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2024, Номер unknown

Опубликована: Ноя. 9, 2024

Abstract We present a Recurrent Vision Transformer (Recurrent ViT) that integrates capacity-limited spatial memory module with self-attention to emulate primate-like visual attention. Trained via reinforcement learning on spatially cued orientation-change detection task, our model exhibits hallmark behavioral signatures of primate attention—including improved accuracy and faster reaction times for stimuli scale cue validity. Analysis its maps reveals rich temporal dynamics: biases induced by cues are maintained during blank intervals reactivated prior anticipated stimulus changes, mirroring the top–down modulation observed in studies. Moreover, targeted manipulations internal attention weights yield performance changes analogous those produced microstimulation attentional control regions such as frontal eye fields superior colliculus. These findings demonstrate embedding recurrent, memory-driven mechanisms within transformer architectures may provide computational framework linking artificial biological

Язык: Английский

A Study on Optimal Data Bandwidth of Recurrent Neural Network–Based Dynamics Model for Robot Manipulators DOI Creative Commons

Seungcheon Shin,

Minseok Kang, Jaemin Baek

и другие.

Advanced Intelligent Systems, Год журнала: 2025, Номер unknown

Опубликована: Март 11, 2025

In this article, a recurrent neural network (RNN)‐based learning method is propdosed for achieving the overall dynamic model of robot manipulators. Several sections, e.g., data acquisition, model, hidden layers, nodes, activation function, and bandwidth, are designed to make RNN‐based establish The proposed has key point that optimal bandwidth can be obtained by loss function its derivative in Since set effective process, it helps provide high hit rate while significantly reducing time‐consuming tasks caused trial errors any From these benefits, offers compact form simplicity so produce convenience practicing engineers industrial fields. effectiveness one verified through experiments with three scenarios, which compared original real manipulator.

Язык: Английский

Процитировано

0
Explainable Artificial Intelligence for Computer Vision and Quantum Machine Learning DOI Open Access
Asharul Islam Khan,

Ali Al Badi,

Mohammed Alqahtani

и другие.

Procedia Computer Science, Год журнала: 2025, Номер 258, С. 3723 - 3730

Опубликована: Янв. 1, 2025

Язык: Английский

Процитировано

0
Improving skin lesion classification through saliency-guided loss functions DOI

Rym Dakhli,

Walid Barhoumi

Computers in Biology and Medicine, Год журнала: 2025, Номер 192, С. 110299 - 110299

Опубликована: Май 14, 2025

Язык: Английский

Процитировано

0
Explanation strategies in humans versus current explainable artificial intelligence: Insights from image classification DOI Creative Commons
Ruoxi Qi, Yueyuan Zheng, Y. S. Yang

и другие.

British Journal of Psychology, Год журнала: 2024, Номер unknown

Опубликована: Июнь 10, 2024

Abstract Explainable AI (XAI) methods provide explanations of models, but our understanding how they compare with human remains limited. Here, we examined participants' attention strategies when classifying images and explaining classified the through eye‐tracking compared their saliency‐based from current XAI methods. We found that humans adopted more explorative for explanation task than classification itself. Two representative were identified clustering: One involved focused visual scanning on foreground objects conceptual explanations, which contained specific information inferring class labels, whereas other rated higher in effectiveness early category learning. Interestingly, saliency map had highest similarity to strategy humans, highlighting discriminative features invoking observable causality perturbation those internal associated score. Thus, use both during explanation, serve different purposes, highlight informing match better potentially accessible users.

Язык: Английский

Процитировано

1
Exploring Pathogen Presence Prediction in Pastured Poultry Farms through Transformer-Based Models and Attention Mechanism Explainability DOI Creative Commons

Athish Ram Das,

Nisha Pillai, Bindu Nanduri

и другие.

Microorganisms, Год журнала: 2024, Номер 12(7), С. 1274 - 1274

Опубликована: Июнь 23, 2024

In this study, we explore how transformer models, which are known for their attention mechanisms, can improve pathogen prediction in pastured poultry farming. By combining farm management practices with microbiome data, our model outperforms traditional methods terms of the F1 score—an evaluation metric performance—thus fulfilling an essential need predictive microbiology. Additionally, emphasis is on making model’s predictions explainable. We introduce a novel approach identifying feature importance using matrix and PageRank algorithm, offering insights that enhance comprehension established techniques such as DeepLIFT. Our results showcase efficacy models food safety mark noteworthy contribution to progress explainable AI within biomedical sciences. This study sheds light impact effective highlights technological advancements ensuring safety.

Язык: Английский

Процитировано

1
Artificial intelligence correctly classifies developmental stages of monarch caterpillars enabling better conservation through the use of community science photographs DOI Creative Commons
Naresh Neupane,

Rhea Goswami,

Kyle Robert Harrison

и другие.

Scientific Reports, Год журнала: 2024, Номер 14(1)

Опубликована: Ноя. 7, 2024

Rapid technological advances and growing participation from amateur naturalists have made countless images of insects in their natural habitats available on global web portals. Despite automated species identification, traits like developmental stage or health remain underexplored manually annotated, with limited focus automating these features. As a proof-of-concept, we developed computer vision model utilizing the YOLOv5 algorithm to accurately detect monarch butterfly caterpillars photographs classify them into five stages (instars). The training data were obtained iNaturalist portal, first classified annotated by experts allow supervised models. Our best trained demonstrates excellent performance object detection, achieving mean average precision score 95% across all instars. In terms classification, YOLOv5l version yielded performance, reaching 87% instar classification accuracy for classes test set. approach show promise developing detection models insects, resource that can be used large-scale mechanistic studies. These photos hold valuable untapped information, we've released our collection as an open dataset support replication expansion methods.

Язык: Английский

Процитировано

1
A Neural Network Model of Visual Attention Integrating Biased Competition and Reinforcement Learning DOI Open Access
Jonathan H. Morgan, Badr F. Albanna, James P. Herman

и другие.

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2024, Номер unknown

Опубликована: Ноя. 9, 2024

Abstract We present a Recurrent Vision Transformer (Recurrent ViT) that integrates capacity-limited spatial memory module with self-attention to emulate primate-like visual attention. Trained via reinforcement learning on spatially cued orientation-change detection task, our model exhibits hallmark behavioral signatures of primate attention—including improved accuracy and faster reaction times for stimuli scale cue validity. Analysis its maps reveals rich temporal dynamics: biases induced by cues are maintained during blank intervals reactivated prior anticipated stimulus changes, mirroring the top–down modulation observed in studies. Moreover, targeted manipulations internal attention weights yield performance changes analogous those produced microstimulation attentional control regions such as frontal eye fields superior colliculus. These findings demonstrate embedding recurrent, memory-driven mechanisms within transformer architectures may provide computational framework linking artificial biological

Язык: Английский

Процитировано

0