Unseen Locations Research Articles

Detecting floating litter in freshwater bodies with semi-supervised deep learning

Researchers and practitioners have extensively utilized supervised Deep Learning methods to quantify floating litter in rivers and canals. These methods require the availability of large amount of labeled data for training. The labeling work is expensive and laborious, resulting in small open datasets available in the field compared to the comprehensive datasets for computer vision, e.g., ImageNet. Fine-tuning models pre-trained on these larger datasets helps improve litter detection performances and reduces data requirements. Yet, the effectiveness of using features learned from generic datasets is limited in large-scale monitoring, where automated detection must adapt across different locations, environmental conditions, and sensor settings. To address this issue, we propose a two-stage semi-supervised learning method to detect floating litter based on the Swapping Assignments between multiple Views of the same image (SwAV). SwAV is a self-supervised learning approach that learns the underlying feature representation from unlabeled data. In the first stage, we used SwAV to pre-train a ResNet50 backbone architecture on about 100k unlabeled images. In the second stage, we added new layers to the pre-trained ResNet50 to create a Faster R-CNN architecture, and fine-tuned it with a limited number of labeled images (≈1.8k images with 2.6k annotated litter items). We developed and validated our semi-supervised floating litter detection methodology for images collected in canals and waterways of Delft (the Netherlands) and Jakarta (Indonesia). We tested for out-of-domain generalization performances in a zero-shot fashion using additional data from Ho Chi Minh City (Vietnam), Amsterdam and Groningen (the Netherlands). We benchmarked our results against the same Faster R-CNN architecture trained via supervised learning alone by fine-tuning ImageNet pre-trained weights. The findings indicate that the semi-supervised learning method matches or surpasses the supervised learning benchmark when tested on new images from the same training locations. We measured better performances when little data (≈200 images with about 300 annotated litter items) is available for fine-tuning and with respect to reducing false positive predictions. More importantly, the proposed approach demonstrates clear superiority for generalization on the unseen locations, with improvements in average precision of up to 12.7%. We attribute this superior performance to the more effective high-level feature extraction from SwAV pre-training from relevant unlabeled images. Our findings highlight a promising direction to leverage semi-supervised learning for developing foundational models, which have revolutionized artificial intelligence applications in most fields. By scaling our proposed approach with more data and compute, we can make significant strides in monitoring to address the global challenge of litter pollution in water bodies.

Domain knowledge-driven variational recurrent networks for drought monitoring

In the context of climate change, droughts, increasingly frequent and severe, necessitate effective monitoring. Existing methods, such as drought indices and data-driven models, face important limitations. Drought indices are built on prior expert knowledge but lack calibration based on actual drought events, while data-driven models prioritize goodness of fit over real event identification, undermining their credibility and generalization, and also struggling to generalize from regional to large-scale contexts. To address these challenges, here we introduce a hybrid machine learning framework for time series that combines domain knowledge and observational data in a variational recurrent neural network. The network models the joint distribution of total precipitation, air temperature, and real drought events, providing accurate predictions and uncertainty estimates. Extensive experiments focusing on a wide range of European drought events from 2011 to 2018 consistently show that our hybrid model surpasses both drought indices and data-driven models in terms of accuracy in drought detection, underlining its effectiveness, robustness, and stability. Our model achieves the best ROC-AUC (%) results in Afghanistan (79.7 ± 0.5), Italy (84.3 ± 0.6), Russia (89.4 ± 0.2), Europe-0 (84.3 ± 0.1), and Europe-1 (82.8 ± 0.4), effectively capturing the starting and ending times of drought events with lower uncertainty, and also generalizing better for unseen locations.

Local-global methods for generalised solar irradiance forecasting

For efficient operation, solar power operators often require generation forecasts for multiple sites with varying data availability. Many proposed methods for forecasting solar irradiance / solar power production formulate the problem as a time-series, using current observations to generate forecasts. This necessitates a real-time data stream and enough historical observations at every location for these methods to be deployed. In this paper, we propose the use of Global methods to train generalised models. Using data from 20 locations distributed throughout the UK, we show that it is possible to learn models without access to data for all locations, enabling them to generate forecasts for unseen locations. We show a single Global model trained on multiple locations can produce more consistent and accurate results across locations. Furthermore, by leveraging weather observations and measurements from other locations we show it is possible to create models capable of accurately forecasting irradiance at locations without any real-time data. We apply our approaches to both classical and state-of-the-art Machine Learning methods, including a Transformer architecture. We compare models using satellite imagery or point observations (temperature, pressure, etc.) as weather data. These methods could facilitate planning and optimisation for both newly deployed solar farms and domestic installations from the moment they come online.

Can end-to-end data-driven models outperform traditional semi-physical models in separating 1-min irradiance?

As a crucial component of the model chain, which facilitates irradiance-to-power conversion during solar resource assessment and forecasting, separation modeling continues to draw attention in both academia and industry. However, when evaluating even the best separation model today, one can quickly recognize its limited accuracy compared to other energy meteorology models such as transposition models. The task of separating global horizontal irradiance into diffuse and beam components does not seem soluble by any derivative effort aimed at tweaking the existing semi-physical models. As a result, an appealing alternative is to consider end-to-end data-driven models, which have demonstrated predictive capability in scenarios where the volume of data is substantial and the interaction among variables is complex. This work discusses the separation of 1-min irradiance from a data-driven perspective. In this preliminary study, a total of 10 representative data-driven separation models are developed and compared to the state-of-the-art semi-physical models, using a comprehensive 1-min irradiance database that spans five years and covers numerous climate types. The average error of the data-driven models is found to be 15.2% to 22.6% lower than that of the semi-physical models for training locations and 7.9% to 17.6% lower for completely unseen locations. Data-driven models also have significantly lower standard deviations (up to 87.2% even for completely unseen locations), highlighting their robustness. In addition, this work provides a guideline for choosing between data-driven and semi-physical models based on data availability, application needs, computational resources, interpretability, and model adaptability. Furthermore, the study underscores the challenges in accurately predicting the diffuse fraction using available input features and indicates that the incorporation of additional weather-related variables and domain knowledge could enhance the performance of data-driven separation models.

Impact localization in composite structures with guided wave and 1D convolutional neural network

Low velocity impacts may cause damages to the composite material. Piezoelectric transducers can be mounted onto the plate surface to monitor the impact by recording the impact induced guided waves. In this study, an impact localization algorithm based on 1D convolutional neural network is proposed. To test the effectiveness of the proposed algorithm, 4 PZTs were glued to a composite plate to monitor a region of 200 mm × 200 mm, and an impact experiment was conducted by dropping a glass ball at 117 different locations on the plate. The time-domain guided wave signals obtained by 4 PZTs were used as input to the 1D convolutional neural network. The output of the network was the coordinates of impact. The trained network was afterwards applied to locate impacts at unseen locations, and the mean localization error is 15.2 mm.

Measuring configural spatial knowledge: Individual differences in correlations between pointing and shortcutting

People use environmental knowledge to maintain a sense of direction in daily life. This knowledge is typically measured by having people point to unseen locations (judgments of relative direction) or navigate efficiently in the environment (shortcutting). Some people can estimate directions precisely, while others point randomly. Similarly, some people take shortcuts not experienced during learning, while others mainly follow learned paths. Notably, few studies have directly tested the correlation between pointing and shortcutting performance. We compared pointing and shortcutting in two experiments, one using desktop virtual reality (VR) (N = 57) and one using immersive VR (N = 48). Participants learned a new environment by following a fixed route and were then asked to point to unseen locations and navigate to targets by the shortest path. Participants’ performance was clustered into two groups using K-means clustering. One (lower ability) group pointed randomly and showed low internal consistency across trials in pointing, but were able to find efficient routes, and their pointing and efficiency scores were not correlated. The others (higher ability) pointed precisely, navigated by efficient routes, and their pointing and efficiency scores were correlated. These results suggest that with the same egocentric learning experience, the correlation between pointing and shortcutting depends on participants’ learning ability, and internal consistency and discriminating power of the measures. Inconsistency and limited discriminating power can lead to low correlations and mask factors driving human variation. Psychometric properties, largely under-reported in spatial cognition, can advance our understanding of individual differences and cognitive processes for complex spatial tasks.

Bayesian Kernelized Matrix Factorization for Spatiotemporal Traffic Data Imputation and Kriging

Missingness and corruption are common problems for real-world traffic data. How to accurately perform imputation and prediction based on incomplete or even sparse traffic data becomes a critical research question in intelligent transportation systems. Low-rank matrix factorization (MF) is a common solution for the general missing value imputation problem. To better characterize and encode the strong spatial and temporal consistency in traffic data, existing work has introduced flexible spatial/temporal Gaussian process (GP) priors to model the latent factors in MF framework, which also allows us to perform kriging for unseen locations and virtual sensors. However, learning the hyperparameters in GP kernels remains a challenging task. In this paper, we present a Bayesian kernelized matrix factorization (BKMF) model with an efficient Markov chain Monte Carlo (MCMC) sampling algorithm for model inference. By learning the kernel hyperparameters from their marginal posteriors through a slice sampling treatment and updating the latent factors alternatively with Gibbs sampling, we achieve a fully Bayesian model for the spatiotemporally kernelized (i.e., GP prior regularized) MF framework. We apply BKMF on both imputation and kriging tasks, and our results demonstrate the superiority of BKMF compared with state-of-the-art spatiotemporal models. In addition, we also explore the effects of different GP kernels in characterizing networked spatiotemporal traffic state data.

Image-Based Classification of Double-Barred Beach States Using a Convolutional Neural Network and Transfer Learning

Nearshore sandbars characterize many sandy coasts, and unravelling their dynamics is crucial to understanding nearshore sediment pathways. Sandbar morphologies exhibit complex patterns that can be classified into distinct states. The tremendous progress in data-driven learning in image recognition has recently led to the first automated classification of single-barred beach states from Argus imagery using a Convolutional Neural Network (CNN). Herein, we extend this method for the classification of beach states in a double-barred system. We used transfer learning to fine-tune the pre-trained network of ResNet50. Our data consisted of labelled single-bar time-averaged images from the beaches of Narrabeen (Australia) and Duck (US), complemented by 9+ years of daily averaged low-tide images of the double-barred beach of the Gold Coast (Australia). We assessed seven different CNNs, of which each model was tested on the test data from the location where its training data came from, the self-tests, and on the test data of alternate, unseen locations, the transfer-tests. When the model trained on the single-barred data of both Duck and Narrabeen was tested on unseen data of the double-barred Gold Coast, we achieved relatively low performances as measured by F1 scores. In contrast, models trained with only the double-barred beach data showed comparable skill in the self-tests with that of the single-barred models. We incrementally added data with labels from the inner or outer bar of the Gold Coast to the training data from both single-barred beaches, and trained models with both single- and double-barred data. The tests with these models showed that which bar the labels used for training the model mattered. The training with the outer bar labels led to overall higher performances, except at the inner bar. Furthermore, only 10% of additional data with the outer bar labels was needed for reasonable transferability, compared to the 20% of additional data needed with the inner bar labels. Additionally, when trained with data from multiple locations, more data from a new location did not always positively affect the model’s performance on other locations. However, the larger diversity of images coming from more locations allowed the transferability of the model to the locations from where new training data were added.

Spatial Prediction of COVID-19 Pandemic Dynamics in the United States

The impact of COVID-19 across the United States (US) has been heterogeneous, with rapid spread and greater mortality in some areas compared with others. We used geographically-linked data to test the hypothesis that the risk for COVID-19 was defined by location and sought to define which demographic features were most closely associated with elevated COVID-19 spread and mortality. We leveraged geographically-restricted social, economic, political, and demographic information from US counties to develop a computational framework using structured Gaussian process to predict county-level case and death counts during the pandemic’s initial and nationwide phases. After identifying the most predictive information sources by location, we applied an unsupervised clustering algorithm and topic modeling to identify groups of features most closely associated with COVID-19 spread. Our model successfully predicted COVID-19 case counts of unseen locations after examining case counts and demographic information of neighboring locations, with overall Pearson’s correlation coefficient and the proportion of variance explained as 0.96 and 0.84 during the initial phase and 0.95 and 0.87 during the nationwide phase, respectively. Aside from population metrics, presidential vote margin was the most consistently selected spatial feature in our COVID-19 prediction models. Urbanicity and 2020 presidential vote margins were more predictive than other demographic features. Models trained using death counts showed similar performance metrics. Topic modeling showed that counties with similar socioeconomic and demographic features tended to group together, and some of these feature sets were associated with COVID-19 dynamics. Clustering of counties based on these feature groups found by topic modeling revealed groups of counties that experienced markedly different COVID-19 spread. We conclude that topic modeling can be used to group similar features and identify counties with similar features in epidemiologic research.

Simulating Oil and Water Production in Reservoirs with Generative Deep Learning

Summary This study investigated the ability to produce accurate multiphase flow profiles simulating the response of producing reservoirs, using generative deep learning (GDL) methods. Historical production data from numerical simulators were used to train a variational autoencoder (VAE) algorithm that was then used to predict the output of new wells in unseen locations. This work describes a procedure in which data analysis techniques can be applied to existing historical production profiles to gain insight into field-level reservoir flow behavior. The procedure includes clustering, dimensionality reduction, correlation, in addition to novel interpretation methodologies that synthesize the results from reservoir simulation output, characterizing flow conditions. The insight was then used to build and select samples to train a VAE algorithm that reproduces the multiphase reservoir behavior for unseen operational conditions with high accuracy. Furthermore, using deep feature space interpolation, the trained algorithm can be used to further generate new predictions of the reservoir response under operational conditions for which we do not have previous examples in the training data set. It is found that VAE can be used as a robust multiphase flow simulator. Applying the methodology to the problem of determining multiphase production rate from new producing wells in undrilled locations showed positive results. The methodology was tested successfully in predicting multiphase production under different scenarios including multiwell channelized and heterogeneous reservoirs. Comparison with other shallow supervised algorithms demonstrated improvements realized by the proposed methodology. The study developed a novel methodology to interpret both data and GDL algorithms, geared toward improving reservoir management. The method was able to predict the performance of new wells in previously undrilled locations, potentially without using a reservoir simulator.

Convolutional conditional neural processes for local climate downscaling

Abstract. A new model is presented for multisite statistical downscaling of temperature and precipitation using convolutional conditional neural processes (convCNPs). ConvCNPs are a recently developed class of models that allow deep-learning techniques to be applied to off-the-grid spatio-temporal data. In contrast to existing methods that map from low-resolution model output to high-resolution predictions at a discrete set of locations, this model outputs a stochastic process that can be queried at an arbitrary latitude–longitude coordinate. The convCNP model is shown to outperform an ensemble of existing downscaling techniques over Europe for both temperature and precipitation taken from the VALUE intercomparison project. The model also outperforms an approach that uses Gaussian processes to interpolate single-site downscaling models at unseen locations. Importantly, substantial improvement is seen in the representation of extreme precipitation events. These results indicate that the convCNP is a robust downscaling model suitable for generating localised projections for use in climate impact studies.

The human visual system and CNNs can both support robust online translation tolerance following extreme displacements.

Visual translation tolerance refers to our capacity to recognize objects over a wide range of different retinal locations. Although translation is perhaps the simplest spatial transform that the visual system needs to cope with, the extent to which the human visual system can identify objects at previously unseen locations is unclear, with some studies reporting near complete invariance over 10 degrees and other reporting zero invariance at 4 degrees of visual angle. Similarly, there is confusion regarding the extent of translation tolerance in computational models of vision, as well as the degree of match between human and model performance. Here, we report a series of eye-tracking studies (total N = 70) demonstrating that novel objects trained at one retinal location can be recognized at high accuracy rates following translations up to 18 degrees. We also show that standard deep convolutional neural networks (DCNNs) support our findings when pretrained to classify another set of stimuli across a range of locations, or when a global average pooling (GAP) layer is added to produce larger receptive fields. Our findings provide a strong constraint for theories of human vision and help explain inconsistent findings previously reported with convolutional neural networks (CNNs).

Neural Manipulation Planning on Constraint Manifolds

The presence of task constraints imposes a significant challenge to motion planning. Despite all recent advancements, existing algorithms are still computationally expensive for most planning problems. In this paper, we present Constrained Motion Planning Networks (CoMPNet), the first neural planner for multimodal kinematic constraints. Our approach comprises the following components: i) constraint and environment perception encoders; ii) neural robot configuration generator that outputs configurations on/near the constraint manifold(s), and iii) a bidirectional planning algorithm that takes the generated configurations to create a feasible robot motion trajectory. We show that CoMPNet solves practical motion planning tasks involving both unconstrained and constrained problems. Furthermore, it generalizes to new unseen locations of the objects, i.e., not seen during training, in the given environments with high success rates. When compared to the state-of-the-art constrained motion planning algorithms, CoMPNet outperforms by order of magnitude improvement in computational speed with a significantly lower variance.

Retrieval of Crude Protein in Perennial Ryegrass Using Spectral Data at the Canopy Level

Crude protein estimation is an important parameter for perennial ryegrass (Lolium perenne) management. This study aims to establish an effective and affordable approach for a non-destructive, near-real-time crude protein retrieval based solely on top-of-canopy reflectance. The study contrasts different spectral ranges while selecting a minimal number of bands and analyzing achievable accuracies for crude protein expressed as a dry matter fraction or on a weight-per-area basis. In addition, the model’s prediction performance in known and new locations is compared. This data collection comprised 266 full-range (350–2500 nm) proximal spectral measurements and corresponding ground truth observations in Australia and the Netherlands from May to November 2018. An exhaustive-search (based on a genetic algorithm) successfully selected band subsets within different regions and across the full spectral range, minimizing both the number of bands and an error metric. For field conditions, our results indicate that the best approach for crude protein estimation relies on the use of the visible to near-infrared range (400–1100 nm). Within this range, eleven sparse broad bands (of 10 nm bandwidth) provide performance better than or equivalent to those of previous studies that used a higher number of bands and narrower bandwidths. Additionally, when using top-of-canopy reflectance, our results demonstrate that the highest accuracy is achievable when estimating crude protein on its weight-per-area basis (RMSEP 80 kg.ha−1). These models can be employed in new unseen locations, resulting in a minor decrease in accuracy (RMSEP 85.5 kg.ha−1). Crude protein as a dry matter fraction presents a bottom-line accuracy (RMSEP) ranging from 2.5–3.0 percent dry matter in optimal models (requiring ten bands). However, these models display a low explanatory ability for the observed variability (R2 > 0.5), rendering them only suitable for qualitative grading.

Intact wayfinding abilities in patients with Parkinson's disease

IntroductionPrevious studies have found that patients with Parkinson's Disease (PD) showed impairments in certain aspects of spatial orientation. The current study aimed to systematically investigate whether these impairments extend to wayfinding abilities in patients with PD. Wayfinding refers to the ability to navigate to an unseen location in the environment and is essential to one's everyday functioning. MethodsA total of 24 patients with PD, 20 ability matched controls, 21 college students participated in a series of experimental behavioral tasks and a self-report of environmental abilities. In the route learning task, participants learned and then recalled routes. In the survey learning task, participants were asked to form configurational or survey knowledge. In the map tracing tack, participants were asked to trace the turning directions of a route on a map. ResultsPatients with PD showed no impairments in the behavioral measures of wayfinding relative to ability matched controls. Both groups performed worse than college students, who had higher cognitive levels. Patients with PD, however, reported a higher competency in environmental abilities than college students. ConclusionAlthough wayfinding abilities may decrease as cognitive abilities decline, they do not appear as a unique impairment for patients with PD relative to their cognitive level.

Model-Aided Deep Learning Method for Path Loss Prediction in Mobile Communication Systems at 2.6 GHz

Accurate channel models are essential to evaluate mobile communication system performance and optimize coverage for existing deployments. The introduction of various transmission frequencies for 5G imposes new challenges for accurate radio performance prediction. This paper compares traditional channel models to a channel model obtained using Deep Learning (DL)-techniques utilizing satellite images aided by a simple path loss model. Experimental measurements are gathered and compose the training and test set. This paper considers path loss modelling techniques offered by state-of-the-art stochastic models and a ray-tracing model for comparison and evaluation. The results show that 1) the satellite images offer an increase in predictive performance by ≈ 0.8 dB, 2) The model-aided technique offers an improvement of ≈ 1 dB, and 3) that the proposed DL model is capable of improving path loss prediction at unseen locations for 811 MHz with ≈ 1 dB and ≈ 4.7 dB for 2630 MHz.

Exploration vs. Data Refinement via Multiple Mobile Sensors.

We examine the deployment of multiple mobile sensors to explore an unknown region to map regions containing concentration of a physical quantity such as heat, electron density, and so on. The exploration trades off between two desiderata: to continue taking data in a region known to contain the quantity of interest with the intent of refining the measurements vs. taking data in unobserved areas to attempt to discover new regions where the quantity may exist. Making reasonable and practical decisions to simultaneously fulfill both goals of exploration and data refinement seem to be hard and contradictory. For this purpose, we propose a general framework that makes value-laden decisions for the trajectory of mobile sensors. The framework employs a Gaussian process regression model to predict the distribution of the physical quantity of interest at unseen locations. Then, the decision-making on the trajectories of sensors is performed using an epistemic utility controller. An example is provided to illustrate the merit and applicability of the proposed framework.

Responding to sounds from unseen locations: crossmodal attentional orienting in response to sounds presented from the rear.

To date, most of the research on spatial attention has focused on probing people's responses to stimuli presented in frontal space. That is, few researchers have attempted to assess what happens in the space that is currently unseen (essentially rear space). In a sense, then, 'out of sight' is, very much, 'out of mind'. In this review, we highlight what is presently known about the perception and processing of sensory stimuli (focusing on sounds) whose source is not currently visible. We briefly summarize known differences in the localizability of sounds presented from different locations in 3D space, and discuss the consequences for the crossmodal attentional and multisensory perceptual interactions taking place in various regions of space. The latest research now clearly shows that the kinds of crossmodal interactions that take place in rear space are very often different in kind from those that have been documented in frontal space. Developing a better understanding of how people respond to unseen sound sources in naturalistic environments by integrating findings emerging from multiple fields of research will likely lead to the design of better warning signals in the future. This review highlights the need for neuroscientists interested in spatial attention to spend more time researching what happens (in terms of the covert and overt crossmodal orienting of attention) in rear space.

Fast Spatial Interpolation using Sparse Gaussian Processes

The estimation of the natural ambient radioactivity in this entry to the Spatial Interpolation Comparison 2004 (SIC2004) uses Gaussian processes (GPs) to predict the underlying dispersal process. GPs enable us to predict easily levels of radioactivity at previously unseen locations and in addition they allow us to assess the uncertainty in the predicted value. To speed up computation time, which is cubic in the number of examples, a sequential, sparse implementation of the Gaussian process inference (SSGP) was used together with a Gaussian observational noise assumption. The examination of the available data led to a covariance function which is a mixture of exponential and squared-exponential functions. The mixture was chosen so that it incorporates both the local ambiguity in the data and also at the same time it captures the larger-scale variation of the observations.A further characteristic of the competition was the availability of 10 days of prior observations. The individual sets comprised of data that was recorded at the same time but at different locations across Germany. Consequently in the modelling stage we assumed that the underlying process governing the observations was invariant across the 10 days of prior observations, but that each day, i.e. each dataset, should be treated independently. These prior observations were used to infer the parameters of the covariance function, which are called hyper-parameters. Conforming to the Bayesian inference method, the exact values of the hyper-parameters were not fixed; rather we assigned a probability distribution to these parameters, giving a two level inference scheme. The inferred values of the hyper-parameters were used to set up a prior GP, and the same two level inference was employed when evaluating the method on the released SIC2004 dataset.We show that the sparse approximation with the SSGP method can be used instead of the conventional GPs without significant loss in accuracy leading to a greatly reduced calculation time. We also compare the SSGP performance with standard machine learning techniques: the SSGP results compare favourably to the other benchmark techniques.

Dorso-medial and ventro-lateral functional specialization of the human retrosplenial complex in spatial updating and orienting.

The retrosplenial complex is a region within the posterior cingulate cortex implicated in spatial navigation. Here, we investigated the functional specialization of this large and anatomically heterogeneous region using fMRI and resting-state functional connectivity combined with a spatial task with distinct phases of spatial 'updating' (i.e., integrating and maintaining object locations in memory during spatial displacement) and 'orienting' (i.e., recalling unseen locations from current position in space). Both spatial 'updating' and 'orienting' produced bilateral activity in the retrosplenial complex, among other areas. However, spatial 'updating' produced slightly greater activity in ventro-lateral portions, of the retrosplenial complex, whereas spatial 'orienting' produced greater activity in a more dorsal and medial portion of it (both regions localized along the parieto-occipital fissure). At rest, both ventro-lateral and dorso-medial subregions of the retrosplenial complex were functionally connected to the hippocampus and parahippocampus, regions both involved in spatial orientation and navigation. However, the ventro-lateral subregion of the retrosplenial complex displayed more positive functional connectivity with ventral occipital and temporal object recognition regions, whereas the dorso-medial subregion activity was more correlated to dorsal activity and frontal activity, as well as negatively correlated with more ventral parietal structures. These findings provide evidence for a dorso-medial to ventro-lateral functional specialization within the human retrosplenial complex that may shed more light on the complex neural mechanisms underlying spatial orientation and navigation in humans.