Publications

2025

1. Ecol. Sol. Evid.

   

   Monitoring protected areas by integrating machine learning, remote sensing and citizen science

   Thijs L Plas, David G Alexander, and Michael JO Pocock

   Ecological Solutions and Evidence, May 2025

   Poster Award Abstract DOI

   

   Awarded the Best Poster Award at the 2025 Dutch Computational Sciences (DuComs) conference.

   1. Protected Areas (PAs) are central to addressing the world’s biodiversity crisis, but their effectiveness for conservation varies. Therefore, high-resolution habitat condition monitoring is needed to evaluate their individual impacts. Critically, monitoring must efficiently scale to cover large areas and be conducted at regular intervals. 2. Remote sensing (RS) data and citizen-science (CS) species data are two sources of global data available for habitat condition monitoring, and integrating these could provide high-resolution, scalable biodiversity data required for the detailed monitoring of PAs. However, integrating these presents four data analysis challenges: RS data are large and complex, large-scale CS data are biased, integrating RS and CS data is non-trivial, and fine-tuning to local priorities is required. 3. Machine Learning (ML) methods can address these challenges: geospatial foundation models for RS data can compress large data volumes, ML de-biasing techniques can improve CS data quality, deep learning and multimodal ML can help to integrate RS and CS data, and transfer learning can fine-tune models to local priorities. Here, we review these techniques and discuss how they can be applied to habitat condition monitoring. 4. Practical implication. Together, these advances in ML can deliver high-resolution biodiversity data that can be tailored to local priorities, enabling the efficient monitoring of PAs at scale, with the potential to support spatial land use decision-making.

2. Nat. Mach. Intell.

   

   Towards deployment-centric multimodal AI beyond vision and language

   Xianyuan Liu, Jiayang Zhang, Shuo Zhou, and 45 more authors

   Nature Machine Intelligence, Oct 2025

   Abstract DOI

   

   Multimodal artificial intelligence (AI) integrates diverse types of data via machine learning to improve understanding, prediction, and decision-making across disciplines such as healthcare, science, and engineering. However, most multimodal AI advances focus on models for vision and language data, while their deployability remains a key challenge. We advocate a deployment-centric workflow that incorporates deployment constraints early to reduce the likelihood of undeployable solutions, complementing data-centric and model-centric approaches. We also emphasise deeper integration across multiple levels of multimodality and multidisciplinary collaboration to significantly broaden the research scope beyond vision and language. To facilitate this approach, we identify common multimodal-AI-specific challenges shared across disciplines and examine three real-world use cases: pandemic response, self-driving car design, and climate change adaptation, drawing expertise from healthcare, social science, engineering, science, sustainability, and finance. By fostering multidisciplinary dialogue and open research practices, our community can accelerate deployment-centric development for broad societal impact.

3. CVPRW

   

   Predicting butterfly species presence from satellite imagery using soft contrastive regularisation

   Thijs Plas, Stephen Law, and Michael JO Pocock

   In Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR) Workshops, Jun 2025

   Abstract DOI HTML

   

   The growing demand for scalable biodiversity monitoring methods has fuelled interest in remote sensing data, due to its widespread availability and extensive coverage. Traditionally, the application of remote sensing to biodiversity research has focused on mapping and monitoring habitats, but with increasing availability of large-scale citizen-science wildlife observation data, recent methods have started to explore predicting multi-species presence directly from satellite images. This paper presents a new data set for predicting butterfly species presence from satellite data in the United Kingdom. We experimentally optimise a Resnet-based model to predict multi-species presence from 4-band satellite images, and find that this model especially outperforms the mean rate baseline for locations with high species biodiversity. To improve performance, we develop a soft, supervised contrastive regularisation loss that is tailored to probabilistic labels (such as species-presence data), and demonstrate that this improves prediction accuracy. In summary, our new data set and contrastive regularisation method contribute to the open challenge of accurately predicting species biodiversity from remote sensing data, which is key for efficient biodiversity monitoring.

4. TPDL

   

   Flexible metadata harvesting for ecology using large language models

   Zehao Lu, Thijs L Plas, Parinaz Rashidi, and 2 more authors

   In International Conference on Theory and Practice of Digital Libraries, Sep 2025

   Oral talk Abstract DOI

   

   Selected for oral talk at TPDL EcoDL 2025.

   Large, open datasets can accelerate ecological research, particularly by enabling researchers to develop new insights by reusing datasets from multiple sources. However, to find the most suitable datasets to combine and integrate, researchers must navigate diverse ecological and environmental data provider platforms with varying metadata availability and standards. To overcome this obstacle, we have developed a large language model (LLM)-based metadata harvester that flexibly extracts metadata from any dataset’s landing page, and converts these to a user-defined, unified format using existing metadata standards. We validate that our tool is able to extract both structured and unstructured metadata with equal accuracy, aided by our LLM post-processing protocol. Furthermore, we utilise LLMs to identify links between datasets, both by calculating embedding similarity and by unifying the formats of extracted metadata to enable rule-based processing. Our tool, which flexibly links the metadata of different datasets, can therefore be used for ontology creation or graph-based queries, for example, to find relevant ecological and environmental datasets in a virtual research environment.

5. EurIPS REO

   

   NeurEO: Dissecting Earth embeddings using computational neuroscience

   TL Plas, Jacob Bakermans, and IN Athanasiadis

   In EurIPS 2025 REO workshop: The 1st workshop on Advances on Representation Learning for Earth Observation, Dec 2025

   Oral talk Abstract PDF

   

   Selected for oral talk at EurIPS REO 2025.

   Earth embeddings are compressed representations of rich Earth observation data that enable downstream models to tackle a wide range of complex geospatial tasks. Yet, this increase in performance comes at the cost of decreased interpretability, as the semantic meaning of Earth embeddings is unknown. To address this challenge, we borrow methods from the field of computational neuroscience, which aim to interpret activity in (biological) neurons. Here, we adapt and apply these methods to assign meaning to AlphaEarth embeddings. We analyse embedding features both as stand-alone data, equivalent to spontaneous neural activity, and in relation to land cover and elevation correlates, equivalent to stimulus-response neural activity. We demonstrate correlational and multi-scale structure in the embedding features. Embedding features can display highly specific and directional tuning to land cover and elevation, such as a ‘North vs South woodland edge’ response. When we remove the main trends of combined land cover responses (like water versus land tuning), we discover fine-grained representational details with increased effective dimensionality. Together, our analyses start to unravel what these Earth embeddings represent, which is critical for developing trustworthy downstream applications.

2024

1. Plos CB

   

   What does the mean mean? A simple test for neuroscience

   Alejandro Tlaie, Katharine Shapcott, Thijs L van der Plas, and 7 more authors

   PLOS Computational Biology, Dec 2024

   Abstract DOI

   

   Trial-averaged metrics, e.g. tuning curves or population response vectors, are a ubiquitous way of characterizing neuronal activity. But how relevant are such trial-averaged responses to neuronal computation itself? Here we present a simple test to estimate whether average responses reflect aspects of neuronal activity that contribute to neuronal processing. The test probes two assumptions implicitly made whenever average metrics are treated as meaningful representations of neuronal activity: 1. Reliability: Neuronal responses repeat consistently enough across trials that they convey a recognizable reflection of the average response to downstream regions. 2. Behavioural relevance: If a single-trial response is more similar to the average template, it is more likely to evoke correct behavioural responses. We apply this test to two data sets: (1) Two-photon recordings in primary somatosensory cortices (S1 and S2) of mice trained to detect optogenetic stimulation in S1; and (2) Electrophysiological recordings from 71 brain areas in mice performing a contrast discrimination task. Under the highly controlled settings of Data set 1, both assumptions were largely fulfilled. In contrast, the less restrictive paradigm of Data set 2 met neither assumption. Simulations predict that the larger diversity of neuronal response preferences, rather than higher cross-trial reliability, drives the better performance of Data set 1. We conclude that when behaviour is less tightly restricted, average responses do not seem particularly relevant to neuronal computation, potentially because information is encoded more dynamically. Most importantly, we encourage researchers to apply this simple test of computational relevance whenever using trial-averaged neuronal metrics, in order to gauge how representative cross-trial averages are in a given context.

2. ICLR ML4RS

   

   Multi-stage semantic segmentation to map small and sparsely distributed habitats

   Thijs L van der Plas, Simon T Geikie, David G Alexander, and 1 more author

   ICLR Machine Learning 4 Remote Sensing Workshop Proceedings, Dec 2024

   Abstract

   

   Land cover (LC) maps are used extensively for nature conservation and landscape planning, but low spatial resolution and coarse LC schemes typically limit their applicability to large, broadly-defined habitats. In order to target smaller and more specific habitats, LC maps must be developed at high resolution and fine class detail, using methods that can handle strong class imbalance. In this work, we present a new aerial photography data set with 12.5 cm ground resolution, annotated using a detailed, hierarchical land cover schema. We show that splitting up the semantic segmentation process into multiple stages critically improves the predictive performance, in particular by including the rare LC classes. We then apply this method to create a new LC map of the Peak District National Park (1439 km2), England, at 12.5 cm resolution.

2023

1. eLife

   

   Neural assemblies uncovered by generative modeling explain whole-brain activity statistics and reflect structural connectivity

   Thijs L van der Plas^*, Jérôme Tubiana^*, Guillaume Le Goc, and 6 more authors

   eLife, Dec 2023

   Reproducibility Award Abstract DOI

   

   Awarded the Reproducibility Award at the 2023 BioInference conference.

   Patterns of endogenous activity in the brain reflect a stochastic exploration of the neuronal state space that is constrained by the underlying assembly organization of neurons. Yet, it remains to be shown that this interplay between neurons and their assembly dynamics indeed suffices to generate whole-brain data statistics. Here, we recorded the activity from ∼40,000 neurons simultaneously in zebrafish larvae, and show that a data-driven generative model of neuron-assembly interactions can accurately reproduce the mean activity and pairwise correlation statistics of their spontaneous activity. This model, the compositional Restricted Boltzmann Machine (cRBM), unveils ∼200 neural assemblies, which compose neurophysiological circuits and whose various combinations form successive brain states. We then performed in silico perturbation experiments to determine the interregional functional connectivity, which is conserved across individual animals and correlates well with structural connectivity. Our results showcase how cRBMs can capture the coarse-grained organization of the zebrafish brain. Notably, this generative model can readily be deployed to parse neural data obtained by other large-scale recording techniques.

2. Nat. Neuro.

   

   Propagation of activity through the cortical hierarchy and perception are determined by neural variability

   James M Rowland^*, Thijs L van der Plas^*, Matthias Loidolt^*, and 6 more authors

   Nature Neuroscience, Dec 2023

   Abstract DOI

   

   Brains are composed of anatomically and functionally distinct regions performing specialized tasks, but regions do not operate in isolation. Orchestration of complex behaviors requires communication between brain regions, but how neural dynamics are organized to facilitate reliable transmission is not well understood. Here we studied this process directly by generating neural activity that propagates between brain regions and drives behavior, assessing how neural populations in sensory cortex cooperate to transmit information. We achieved this by imaging two densely interconnected regions—the primary and secondary somatosensory cortex (S1 and S2)—in mice while performing two-photon photostimulation of S1 neurons and assigning behavioral salience to the photostimulation. We found that the probability of perception is determined not only by the strength of the photostimulation but also by the variability of S1 neural activity. Therefore, maximizing the signal-to-noise ratio of the stimulus representation in cortex relative to the noise or variability is critical to facilitate activity propagation and perception.

3. Rem. Sens.

   

   Multi-stage semantic segmentation quantifies fragmentation of small habitats at a landscape scale

   Thijs L van der Plas, Simon T Geikie, David G Alexander, and 1 more author

   Remote Sensing, Dec 2023

   Best Poster Abstract DOI

   

   Awarded a Best Poster Award at the IADF 2024 Computer Vision for Earth Observation summer school.

   Land cover (LC) maps are used extensively for nature conservation and landscape planning, but low spatial resolution and coarse LC schemas typically limit their applicability to large, broadly defined habitats. In order to target smaller and more-specific habitats, LC maps must be developed at high resolution and fine class detail using automated methods that can efficiently scale to large areas of interest. In this work, we present a Machine Learning approach that addresses this challenge. First, we developed a multi-stage semantic segmentation approach that uses Convolutional Neural Networks (CNNs) to classify LC across the Peak District National Park (PDNP, 1439 km2) in the UK using a detailed, hierarchical LC schema. High-level classes were predicted with 95% accuracy and were subsequently used as masks to predict low-level classes with 72% to 92% accuracy. Next, we used these predictions to analyse the degree and distribution of fragmentation of one specific habitat—wet grassland and rush pasture—at the landscape scale in the PDNP. We found that fragmentation varied across areas designated as primary habitat, highlighting the importance of high-resolution LC maps provided by CNN-powered analysis for nature conservation.

4. D.Phil. thesis

   

   Neural mechanisms of state-dependent sensory processing

   Thijs L van der Plas

   Dec 2023

   D.Phil. Abstract DOI

   

   Passed D.Phil. viva with no corrections.

   A key purpose of the central nervous system is to transform sensation into action. During this process, identical stimuli can cause variable behavioural responses because of changes in the condition – or, state – of the brain. This response variability is a common phenomenon, yet the underlying neural mechanisms of how sensory information differentially interacts with varying network states remain obscure. In this thesis, we consider three such mechanisms in depth: inhibition stabilisation, neural variability and predictive learning. First, we analyse all-optical experiments where we were able to simultaneously record and insert neural activity in somatosensory cortex in vivo. Our results show that population activity is stabilised by inhibition, thereby allowing faint sensory stimulation to be detected and propagated downstream. In mice trained to detect these stimuli, we further observe that the neural variability across neurons modulated their perceptual performance, whereas this effect did not occur during passive stimulation. Therefore, maximising the cortical signal-to-noise ratio of behaviourally relevant stimuli is critical to facilitate activity propagation and perception. Separately, in a study of artificial neural networks, we demonstrate that it is advantageous for networks to continuously predict upcoming sensory information, as this facilitates learning of more complex working memory tasks. This could explain recent experimental findings showing that cortical neurons engage in predictive learning even when reward is absent. In sum, this thesis explores how sensory information interacts with ongoing neural activity in both biological and artificial neural networks, and explains mechanisms that underpin the variability of our behavioural response.

2022

1. PMLR

   

   Predictive learning enables neural networks to learn complex working memory tasks

   Thijs L van der Plas, Sanjay G Manohar, and Tim P Vogels

   Proceedings of Machine Learning Research, Dec 2022

   Oral talk Abstract URL HTML

   

   Selected for oral talk at CoLLAs 2022.

   Brains are thought to engage in predictive learning-learning to predict upcoming stimuli-to construct an internal model of their environment. This is especially notable for spatial navigation, as first described by Tolman’s latent learning tasks. However, predictive learning has also been observed in sensory cortex, in settings unrelated to spatial navigation. Apart from normative frameworks such as active inference or efficient coding, what could be the utility of learning to predict the patterns of occurrence of correlated stimuli? Here we show that prediction, and thereby the construction of an internal model of sequential stimuli, can bootstrap the learning process of a working memory task in a recurrent neural network. We implemented predictive learning alongside working memory match-tasks, and networks emerged to solve the prediction task first by encoding information across time to predict upcoming stimuli, and then eavesdropped on this solution to solve the matching task. Eavesdropping was most beneficial when neural resources were limited. Hence, predictive learning acts as a general neural mechanism to learn to store sensory information that can later be essential for working memory tasks.

2018

1. Curr. Biol.

   

   Whole-brain calcium imaging during physiological vestibular stimulation in larval zebrafish

   Geoffrey Migault, Thijs L van der Plas, Hugo Trentesaux, and 6 more authors

   Current Biology, Dec 2018

   Abstract DOI

   

   The vestibular apparatus provides animals with postural and movement-related information that is essential to adequately execute numerous sensorimotor tasks. In order to activate this sensory system in a physiological manner, one needs to macroscopically rotate or translate the animal’s head, which in turn renders simultaneous neural recordings highly challenging. Here we report on a novel miniaturized, light-sheet microscope that can be dynamically co-rotated with a head-restrained zebrafish larva, enabling controlled vestibular stimulation. The mechanical rigidity of the microscope allows one to perform whole-brain functional imaging with state-of-the-art resolution and signal-to-noise ratio while imposing up to 25° in angular position and 6,000°/s2 in rotational acceleration. We illustrate the potential of this novel setup by producing the first whole-brain response maps to sinusoidal and stepwise vestibular stimulation. The responsive population spans multiple brain areas and displays bilateral symmetry, and its organization is highly stereotypic across individuals. Using Fourier and regression analysis, we identified three major functional clusters that exhibit well-defined phasic and tonic response patterns to vestibular stimulation. Our rotatable light-sheet microscope provides a unique tool for systematically studying vestibular processing in the vertebrate brain and extends the potential of virtual-reality systems to explore complex multisensory and motor integration during simulated 3D navigation.