Abstract We consider the use of deep learning methods for modeling complex phenomena like those occurring in natural physical processes. With large amount data gathered on these intensive paradigm could begin to challenge more traditional approaches elaborated over years fields maths or physics. However, despite considerable successes a variety application domains, machine field is not yet ready handle level complexity required by such problems. Using an example application, namely sea...

Forecasting complex dynamical phenomena in settings where only partial knowledge of their dynamics is available a prevalent problem across various scientific fields. While purely data-driven approaches are arguably insufficient this context, standard physical modeling based tend to be over-simplistic, inducing non-negligible errors. In work, we introduce the APHYNITY framework, principled approach for augmenting incomplete described by differential equations with deep models. It consists...

We consider the problem of forecasting complex, nonlinear space-time processes when observations provide only partial information on system's state. propose a natural data-driven framework, where dynamics are modelled by an unknown time-varying differential equation, and evolution term is estimated from data, using neural network. Any future state can then be computed placing associated equation in ODE solver. first evaluate our approach shallow water Euler simulations. find that method not...

A crucial challenge in critical settings like medical diagnosis is making deep learning models used decision-making systems interpretable. Efforts Explainable Artificial Intelligence (XAI) are underway to address this challenge. Yet, many XAI methods evaluated on broad classifiers and fail complex, real-world issues, such as diagnosis. In our study, we focus enhancing user trust confidence automated AI systems, particularly for diagnosing skin lesions, by tailoring an method explain model’s...

Large Language Models (LLMs), when used for conditional text generation, often produce hallucinations, i.e., information that is unfaithful or not grounded in the input context. This issue arises typical generation tasks, such as summarization and data-to-text where goal to fluent based on contextual input. When fine-tuned specific domains, LLMs struggle provide faithful answers a given context, adding generating errors. One underlying cause of this rely statistical patterns learned from...

We investigate a new approach for online handwritten shape recognition. Interesting features of this include learning without manual tuning, from very few training samples, incremental characters, and adaptation to the user-specific needs. The proposed system can deal with two-dimensional graphical shapes such as Latin Asian command gestures, symbols, small drawings, geometric shapes. It be used building block series recognition tasks many applications.

Abstract Data-to-Text Generation (DTG) is a subfield of Natural Language aiming at transcribing structured data in natural language descriptions. The field has been recently boosted by the use neural-based generators which exhibit on one side great syntactic skills without need hand-crafted pipelines; other side, quality generated text reflects training data, realistic settings only offer imperfectly aligned structure-text pairs. Consequently, state-of-art neural models include misleading...

Standard neural networks struggle to generalize under distribution shifts in computer vision. Fortunately, combining multiple can consistently improve out-of-distribution generalization. In particular, weight averaging (WA) strategies were shown perform best on the competitive DomainBed benchmark; they directly average weights of despite their nonlinearities. this paper, we propose Diverse Weight Averaging (DiWA), a new WA strategy whose main motivation is increase functional diversity...

Abstract A key issue in critical contexts such as medical diagnosis is the interpretability of deep learning models adopted decision-making systems. Research eXplainable Artificial Intelligence (XAI) trying to solve this issue. However, often XAI approaches are only tested on generalist classifier and do not represent realistic problems those diagnosis. In paper, we aim at improving trust confidence users towards automatic AI decision systems field skin lesion by customizing an existing...

In this paper, we propose a fully automated method to individually classify patients with Alzheimer's disease (AD) and elderly control subjects based on diffusion tensor (DTI) anatomical magnetic resonance imaging (MRI). We new multimodal measure that combines diffusivity measures at the voxel level. Our approach relies whole-brain parcellation into 73 regions extraction of characteristics in these regions. Discriminative features are identified using different feature selection (FS) methods...

Abstract The internal variability pertains to fluctuations originating from processes inherent the climate component and their mutual interactions. On other hand, forced delineates influence of external boundary conditions on physical system. A methodology is formulated distinguish between within surface air temperature. noise‐to‐noise approach employed for training a neural network, drawing an analogy image noise. large data set compiled using temperature spanning 1901 2020, obtained...

Mesoscale oceanic eddies have a visible signature on sea surface temperature (SST) satellite images, portraying diverse patterns of coherent vortices, gradients, and swirling filaments. However, learning the regularities such signatures defines challenging pattern recognition task, due to their complex structure but also cloud coverage which can corrupt large fraction image. We introduce novel deep approach classify eddy signatures, even if they are corrupted by strong coverage. A dataset...

Until now, mesoscale oceanic eddies have been automatically detected through physical methods on satellite altimetry. Nevertheless, they often a visible signature Sea Surface Temperature (SST) images, which not yet sufficiently exploited. We introduce novel method that employs Deep Learning to detect eddy signatures such input. provide the first available dataset for this task, retaining SST images altimetric-based region proposal. train CNN-based classifier succeeds in accurately detecting...