The modelling of Earth observation data is a challenging problem, typically approached by either purely mechanistic or purely data-driven methods. Mechanistic models encode the domain knowledge and physical rules governing the system. Such models, however, need the correct specification of all interactions between variables in the problem and the appropriate parameterization is a challenge in itself. On the other hand, machine learning approaches are flexible data-driven tools, able to approximate arbitrarily complex functions, but lack interpretability and struggle when data is scarce or in extrapolation regimes. In this paper, we argue that hybrid learning schemes that combine both approaches can address all these issues efficiently. We introduce Gaussian process (GP) convolution models for hybrid modelling in Earth observation (EO) problems. We specifically propose the use of a class of GP convolution models called latent force models (LFMs) for EO time series modelling, analysis and understanding. LFMs are hybrid models that incorporate physical knowledge encoded in differential equations into a multioutput GP model. LFMs can transfer information across time-series, cope with missing observations, infer explicit latent functions forcing the system, and learn parameterizations which are very helpful for system analysis and interpretability. We consider time series of soil moisture from active (ASCAT) and passive (SMOS, AMSR2) microwave satellites. We show how assuming a first order differential equation as governing equation, the model automatically estimates the e-folding time or decay rate related to soil moisture persistence and discovers latent forces related to precipitation. The proposed hybrid methodology reconciles the two main approaches in remote sensing parameter estimation by blending statistical learning and mechanistic modeling.

Federated learning (FL) is an emerging distributed machine learning paradigm that protects privacy and tackles the problem of isolated data islands. At present, there are two main communication strategies of FL: synchronous FL and asynchronous FL. The advantages of synchronous FL are that the model has high precision and fast convergence speed. However, this synchronous communication strategy has the risk that the central server waits too long for the devices, namely, the straggler effect which has a negative impact on some time-critical applications. Asynchronous FL has a natural advantage in mitigating the straggler effect, but there are threats of model quality degradation and server crash. Therefore, we combine the advantages of these two strategies to propose a clustered semi-asynchronous federated learning (CSAFL) framework. We evaluate CSAFL based on four imbalanced federated datasets in a non-IID setting and compare CSAFL to the baseline methods. The experimental results show that CSAFL significantly improves test accuracy by more than +5% on the four datasets compared to TA-FedAvg. In particular, CSAFL improves absolute test accuracy by +34.4% on non-IID FEMNIST compared to TA-FedAvg.

Transformer-based pretrained language models (PLMs) have started a new era in modern natural language processing (NLP). These models combine the power of transformers, transfer learning, and self-supervised learning (SSL). Following the success of these models in the general domain, the biomedical research community has developed various in-domain PLMs starting from BioBERT to the latest BioMegatron and CoderBERT models. We strongly believe there is a need for a survey paper that can provide a comprehensive survey of various transformer-based biomedical pretrained language models (BPLMs). In this survey, we start with a brief overview of foundational concepts like self-supervised learning, embedding layer and transformer encoder layers. We discuss core concepts of transformer-based PLMs like pretraining methods, pretraining tasks, fine-tuning methods, and various embedding types specific to biomedical domain. We introduce a taxonomy for transformer-based BPLMs and then discuss all the models. We discuss various challenges and present possible solutions. We conclude by highlighting some of the open issues which will drive the research community to further improve transformer-based BPLMs.

In this study, we address the problem of chaotic synchronization over a noisy channel by introducing a novel Deep Chaos Synchronization (DCS) system using a Convolutional Neural Network (CNN). Conventional Deep Learning (DL) based communication strategies are extremely powerful but training on large data sets is usually a difficult and time-consuming procedure. To tackle this challenge, DCS does not require prior information or large data sets. In addition, we provide a novel Recurrent Neural Network (RNN)-based chaotic synchronization system for comparative analysis. The results show that the proposed DCS architecture is competitive with RNN-based synchronization in terms of robustness against noise, convergence, and training. Hence, with these features, the DCS scheme will open the door for a new class of modulator schemes and meet the robustness against noise, convergence, and training requirements of the Ultra Reliable Low Latency Communications (URLLC) and Industrial Internet of Things (IIoT).

Annotating the right set of data amongst all available data points is a key challenge in many machine learning applications. Batch active learning is a popular approach to address this, in which batches of unlabeled data points are selected for annotation, while an underlying learning algorithm gets subsequently updated. Increasingly larger batches are particularly appealing in settings where data can be annotated in parallel, and model training is computationally expensive. A key challenge here is scale - typical active learning methods rely on diversity techniques, which select a diverse set of data points to annotate, from an unlabeled pool. In this work, we introduce Active Data Shapley (ADS) -- a filtering layer for batch active learning that significantly increases the efficiency of active learning by pre-selecting, using a linear time computation, the highest-value points from an unlabeled dataset. Using the notion of the Shapley value of data, our method estimates the value of unlabeled data points with regards to the prediction task at hand. We show that ADS is particularly effective when the pool of unlabeled data exhibits real-world caveats: noise, heterogeneity, and domain shift. We run experiments demonstrating that when ADS is used to pre-select the highest-ranking portion of an unlabeled dataset, the efficiency of state-of-the-art batch active learning methods increases by an average factor of 6x, while preserving performance effectiveness.

Satellite images are snapshots of the Earth surface. We propose to forecast them. We frame Earth surface forecasting as the task of predicting satellite imagery conditioned on future weather. EarthNet2021 is a large dataset suitable for training deep neural networks on the task. It contains Sentinel 2 satellite imagery at 20m resolution, matching topography and mesoscale (1.28km) meteorological variables packaged into 32000 samples. Additionally we frame EarthNet2021 as a challenge allowing for model intercomparison. Resulting forecasts will greatly improve (>x50) over the spatial resolution found in numerical models. This allows localized impacts from extreme weather to be predicted, thus supporting downstream applications such as crop yield prediction, forest health assessments or biodiversity monitoring. Find data, code, and how to participate at www.earthnet.tech

As neural network model sizes have dramatically increased, so has the interest in various techniques to reduce their parameter counts and accelerate their execution. An active area of research in this field is sparsity - encouraging zero values in parameters that can then be discarded from storage or computations. While most research focuses on high levels of sparsity, there are challenges in universally maintaining model accuracy as well as achieving significant speedups over modern matrix-math hardware. To make sparsity adoption practical, the NVIDIA Ampere GPU architecture introduces sparsity support in its matrix-math units, Tensor Cores. We present the design and behavior of Sparse Tensor Cores, which exploit a 2:4 (50%) sparsity pattern that leads to twice the math throughput of dense matrix units. We also describe a simple workflow for training networks that both satisfy 2:4 sparsity pattern requirements and maintain accuracy, verifying it on a wide range of common tasks and model architectures. This workflow makes it easy to prepare accurate models for efficient deployment on Sparse Tensor Cores.

Autonomous driving consists of a multitude of interacting modules, where each module must contend with errors from the others. Typically, the motion prediction module depends on a robust tracking system to capture each agent's past movement. In this work, we systematically explore the importance of the tracking module for the motion prediction task and ultimately conclude that the tracking module is detrimental to overall motion prediction performance when the module is imperfect (with as low as 1% error). We explicitly compare models that use tracking information to models that do not across multiple scenarios and conditions. We find that the tracking information only improves performance in noise-free conditions. A noise-free tracker is unlikely to remain noise-free in real-world scenarios, and the inevitable noise will subsequently negatively affect performance. We thus argue future work should be mindful of noise when developing and testing motion/tracking modules, or that they should do away with the tracking component entirely.

Automated seizure detection and classification from electroencephalography (EEG) can greatly improve the diagnosis and treatment of seizures. While prior studies mainly used convolutional neural networks (CNNs) that assume image-like structure in EEG signals or spectrograms, this modeling choice does not reflect the natural geometry of or connectivity between EEG electrodes. In this study, we propose modeling EEGs as graphs and present a graph neural network for automated seizure detection and classification. In addition, we leverage unlabeled EEG data using a self-supervised pre-training strategy. Our graph model with self-supervised pre-training significantly outperforms previous state-of-the-art CNN and Long Short-Term Memory (LSTM) models by 6.3 points (7.8%) in Area Under the Receiver Operating Characteristic curve (AUROC) for seizure detection and 6.3 points (9.2%) in weighted F1-score for seizure type classification. Ablation studies show that our graph-based modeling approach significantly outperforms existing CNN or LSTM models, and that self-supervision helps further improve the model performance. Moreover, we find that self-supervised pre-training substantially improves model performance on combined tonic seizures, a low-prevalence seizure type. Furthermore, our model interpretability analysis suggests that our model is better at identifying seizure regions compared to an existing CNN. In summary, our graph-based modeling approach integrates domain knowledge about EEG, sets a new state-of-the-art for seizure detection and classification on a large public dataset (5,499 EEG files), and provides better ability to identify seizure regions.

Vision models trained on multimodal datasets have recently proved very efficient, both in terms of the wide availability of large image-caption datasets, and in terms of the resulting model's ability to generalize to multiple downstream tasks (e.g. zero-shot learning). One might assume that these abilities are derived, at least in part, from a "semantic grounding" of the visual feature space, learning meaningful structure by mirroring the space of linguistic representations. Contrary to this intuition, we show that a visual model (BiT-M) trained on a very large supervised image dataset (ImageNet-21k) can be as efficient for generalization (few-shot learning, unsupervised clustering) as its multimodal counterpart (CLIP). When compared to other standard visual or language models, the latent representations of BiT-M were found to be just as "linguistic" as those of CLIP. Overall, these findings suggest that the main factor driving improvements of generalization in current models is the size of the training dataset, not (solely) the multimodal grounding property.