Active object recognition, fundamental to tasks like reading and driving, relies on the ability make time-sensitive decisions. People exhibit a flexible tradeoff between speed accuracy, crucial human skill. However, current computational models struggle incorporate time. To address this gap, we present first dataset (with 148 observers) exploring speed–accuracy (SAT) in ImageNet recognition. Participants performed 16-way categorization task where their responses counted only if they occurred...

Abstract Background RNA secondary structure prediction, or folding, is a classic problem in bioinformatics: given sequence of nucleotides, the aim to predict base pairs formed its three dimensional conformation. The inverse designing folding into particular target has only more recently received notable interest. With growing appreciation and understanding functional structural properties motifs, interest utilising biomolecules nano-scale designs, bound increase. However, whereas from an...

The TRAnsient Pockets in Proteins (TRAPP) webserver provides an automated workflow that allows users to explore the dynamics of a protein binding site and detect pockets or sub-pockets may transiently open due internal motion. These transient cryptic be interest design optimization small molecular inhibitors for target interest. TRAPP consists following three modules: (i) structure- generation ensemble structures using one more four possible simulation methods; (ii) analysis-superposition...

Autism spectrum disorder (ASD) is a lifelong neurodevelopmental with very high prevalence around the world. Research progress in field of ASD facial analysis pediatric patients has been hindered due to lack well-established baselines. In this paper, we propose use Vision Transformer (ViT) for computational ASD. The presented model, known as ViTASD, distills knowledge from large expression datasets and offers model structure transferability. Specifically, ViTASD employs vanilla ViT extract...

Abstract Collections of objects such as images are often presented visually in a grid because it is compact representation that lends itself well for search and exploration. Most layouts sorted using very basic criteria, date or filename. In this work we present method to arrange collections respecting an arbitrary distance measure. Pairwise distances preserved much possible, while still producing the specific target arrangement which may be 2D grid, surface sphere, hierarchy, any other...

Existing approaches to unsupervised object discovery (UOD) do not scale up large datasets without approximations that compromise their performance. We propose a novel formulation of UOD as ranking problem, amenable the arsenal distributed methods available for eigenvalue problems and link analysis. Through use self-supervised features, we also demonstrate first effective fully pipeline UOD. Extensive experiments on COCO OpenImages show that, in single-object setting where single prominent is...

Global reconstruction of two-dimensional wall paintings (frescoes) from fragments is an important problem for many archaeological sites. The goal to find the global position and rotation each fragment so that all jointly “reconstruct” original surface (i.e., solve puzzle). Manual placement difficult time-consuming, especially when are irregularly shaped uncolored. Systems have been proposed first acquire 3D scans then use computer algorithms problem. These systems work well small test cases...

Nowadays, autonomous vehicle technology is becoming more and mature. Critical to progress safety, high-definition (HD) maps, a type of centimeter-level map collected using laser sensor, provide accurate descriptions the surrounding environment. The key challenge HD production efficient, high-quality collection annotation large-volume datasets. Due demand for high quality, requires significant manual human effort create annotations, very time-consuming costly process industry. In order reduce...

Abstract A key challenge for the development and deployment of artificial intelligence (AI) solutions in radiology is solving associated data limitations. Obtaining sufficient representative patient datasets with appropriate annotations may be burdensome due to high acquisition cost, safety limitations, privacy restrictions, or low disease prevalence rates. In silico offers a number potential advantages data, such as diminished harm, reduced simplified acquisition, scalability, improved...

To generate evidence regarding the safety and efficacy of artificial intelligence (AI) enabled medical devices, AI models need to be evaluated on a diverse population patient cases, some which may not readily available. We propose an evaluation approach for testing imaging that relies in silico pipelines stochastic digital human anatomy (in object space) with without pathology are imaged using replica acquisition system realistic synthetic image datasets. Here, we release M-SYNTH, dataset...

Abstract As autonomous vehicle technology advances, high‐definition (HD) maps have become essential for ensuring safety and navigation accuracy. However, creating HD with accurate annotations demands substantial human effort, leading to a time‐consuming costly process. Although artificial intelligence (AI) computer vision (CV) algorithms been developed prelabeling maps, significant gap remains in accuracy robustness between AI‐based methods traditional manual pipelines. Additionally,...

Many applications in 3D shape design and augmentation require the ability to make specific edits an object's semantic parameters (e.g., pose of a person's arm or length airplane's wing) while preserving as much existing details possible. We propose learn deep network that infers input then allows user manipulate those parameters. The is trained jointly on shapes from auxiliary synthetic template unlabeled realistic models, ensuring robustness variability relieving need label exemplars. At...

The evaluation of infectious disease processes on radio-logic images is an important and challenging task in medical image analysis. Pulmonary infections can often be best imaged evaluated through computed tomography (CT) scans, which are not available low-resource environments difficult to obtain for critically ill patients. On the other hand, X-ray, a different type imaging procedure, inexpensive, at bedside more widely available, but offers simpler, two dimensional image. We show that by...

Recent advances in deep learning and computer vision have reduced many barriers to automated medical image analysis, allowing algorithms process label-free images improve performance. However, existing techniques extreme computational requirements drop a lot of performance with reduction batch size or training epochs. This paper presents Cross Architectural - Self Supervision (CASS), novel self-supervised approach that leverages Transformer CNN simultaneously. Compared the state art...

Neural networks today often recognize objects as well people do, and thus might serve models of the human recognition process. However, most such provide their answer after a fixed computational effort, whereas reaction time varies, e.g. from 0.2 to 10 s, depending on properties stimulus task. To model effect difficulty time, we considered classification network that uses early-exit classifiers make anytime predictions. Comparing MSDNet accuracy in classifying CIFAR-10 images added Gaussian...

Global reconstruction of two-dimensional wall paintings (frescoes) from fragments is an important problem for many archaeological sites. The goal to find the global position and rotation each fragment so that all jointly reconstruct original surface (i.e., solve puzzle). Manual placement difficult time-consuming, especially when are irregularly shaped uncolored. Systems have been proposed first acquire 3D scans then use computer algorithms problem. These systems work well small test cases...

The core of everyday tasks like reading and driving is active object recognition. Attempts to model such are currently stymied by the inability incorporate time. People show a flexible tradeoff between speed accuracy this crucial human skill. Deep neural networks have emerged as promising candidates for predicting peak recognition performance activity. However, modeling temporal dimension i.e., speed-accuracy (SAT), essential them serve useful computational models how humans recognize...

We propose new semi-supervised nonnegative matrix factorization (SSNMF) models for document classification and provide motivation these as maximum likelihood estimators. The proposed SSNMF simultaneously both a topic model classification, thereby offering highly interpretable results. derive training methods using multiplicative updates each model, demonstrate the application of to single-label multi-label although are flexible other supervised learning tasks such regression. illustrate...

The growing utilization of synthetic medical data (SMD) in training and testing AI-driven tools healthcare necessitates a systematic framework for assessing SMD quality. current lack standardized methodology to evaluate SMD, particularly terms its applicability various scenarios, is significant hindrance broader acceptance applications. Here, we outline an evaluation designed meet the unique requirements applications, introduce concept scorecards, which can serve as comprehensive reports...

A key challenge for the development and deployment of artificial intelligence (AI) solutions in radiology is solving associated data limitations. Obtaining sufficient representative patient datasets with appropriate annotations may be burdensome due to high acquisition cost, safety limitations, privacy restrictions or low disease prevalence rates. In silico offers a number potential advantages data, such as diminished harm, reduced simplified acquisition, scalability, improved quality...

Development of artificial intelligence (AI) techniques in medical imaging requires access to large-scale and diverse datasets for training evaluation. In dermatology, obtaining such remains challenging due significant variations patient populations, illumination conditions, acquisition system characteristics. this work, we propose S-SYNTH, the first knowledge-based, adaptable open-source skin simulation framework rapidly generate synthetic skin, 3D models digitally rendered images, using an...

We propose several new models for semi-supervised nonnegative matrix factorization (SSNMF) and provide motivation SSNMF as maximum likelihood estimators given specific distributions of uncertainty. present multiplicative updates training methods each model, demonstrate the application these to classification, although they are flexible other supervised learning tasks. illustrate promise on both synthetic real data, achieve high classification accuracy 20 Newsgroups dataset.