Algorithm Design and Multidimensional Data Processing (HDSP)

Artificial intelligence combines numerical optimization, machine learning, data processing, and physical-mathematical principles to extract patterns, perform inference, and support decision-making in complex real-world problems.

Unlike approaches based solely on data, this research line seeks to develop artificial intelligence frameworks that incorporate physical, mathematical, and structural relationships of the phenomena under study, enabling the construction of models that are more interpretable, robust, and adaptable to scenarios involving limited, noisy, or heterogeneous data.

This line focuses on the fundamental development of methods capable of analyzing data from multiple modalities and extracting high-level information, with applications in areas such as computational optics, geophysical imaging, remote sensing, medical imaging, and others.

Spectral imaging is an interdisciplinary field that combines optical design, data acquisition, signal processing, and computational algorithms to capture and interpret visual information beyond conventional intensity and color.

Unlike traditional imaging systems, which primarily record spatial information within limited bands of the visible spectrum, spectral imaging systems aim to measure a scene’s response across multiple wavelengths, making it possible to reveal physical, chemical, or material properties that are not directly observable in conventional images.

This research line addresses the design, implementation, and characterization of optical and computational systems for spectral data acquisition, supporting high-level tasks such as the classification, detection, identification, and characterization of materials and objects.

Computational imaging is an interdisciplinary field that combines optical design, signal processing, and computational algorithms to capture, reconstruct, and interpret visual information beyond the limits of conventional photography.

Unlike traditional imaging systems, where the camera seeks to produce an image that directly and faithfully represents a scene, computational imaging systems co-design the acquisition hardware and reconstruction algorithms to extract information that would otherwise be inaccessible, such as depth, spectral content, light fields, temporal dynamics, or the physical properties of materials.

Seismic processing and design integrates data acquisition, physics-based modeling, signal processing, and artificial intelligence to reconstruct and interpret subsurface information from indirect measurements.

Unlike conventional seismic approaches, which often rely on dense and regular acquisition geometries, this research line explores computational strategies and physics-guided models capable of operating in scenarios with sparse, irregular, or incomplete sampling.

Through the design and implementation of inversion algorithms, neural networks, and advanced reconstruction methods, this line seeks to generate subsurface images that enable the characterization of geological structures, improve seismic interpretation, and extract valuable information from limited or complex datasets.

Multidimensional data processing is a field that combines advanced signal processing, artificial intelligence, mathematical representations, and computational algorithms to analyze, reconstruct, and interpret information contained in high-dimensional data.

Unlike traditional approaches, which often treat signals, images, or measurements as isolated or low-complexity structures, this research line addresses complex data with multiple spatial, temporal, spectral, or structural dimensions, including images, video, signals, spectral data, and high-dimensional arrays.

Its objective is to design and implement methods that enable the representation, compression, reconstruction, and extraction of meaningful information from these datasets, supporting tasks such as classification, detection, estimation, prediction, and decision-making.

An artificial intelligence agent is an autonomous computational system capable of making decisions through interaction with its environment. During training, the agent participates in a continuous feedback loop: it observes the current state of the environment, uses a deep neural network to select an action aimed at achieving a long-term goal, receives a new state along with a reward or penalty, and adjusts its behavior to maximize cumulative rewards in future interactions.

Once training is completed, the agent can predict how the environment is likely to evolve based on its observations and make decisions that provide the greatest benefit for the task at hand.

AI agents have applications across a wide range of intelligent systems, including robots capable of navigating and manipulating objects, video games in which agents learn strategies to overcome challenges or compete against other players, autonomous vehicles that make real-time decisions, recommendation systems that learn user preferences, virtual assistants that can plan tasks, and optimization models used in fields such as logistics, finance, healthcare, and industrial automation.