Research internship on developing defensive strategies against adversarial attacks in multimodal agents.

Responsibilities include:

• Characterization of different types of adversarial perturbations including localized patches with spatial constraints, globally distributed subtle perturbations, semantically coherent attacks, and transferable perturbations across architectures.
• Implementation of the Projected Displacement (PD) model to construct abstract security spaces around inputs, leveraging anisotropy and locality properties of adversarial perturbations.
• Development of detection techniques based on adversarial perturbation properties, pre-processing filters for input purification, and defensive ensemble approaches.
• Systematic experimental evaluation and comparative analysis of proposed defense strategies against malicious image patch attacks.

Research Focus: Mitigating vulnerabilities of vision-language multimodal agents against adversarial attacks with malicious image patches (MIPs) that exploit weaknesses in visual processing. Expected to develop an adaptable defense framework and security metrics for evaluating multimodal systems.

Design computational algorithms for estimating the physicochemical properties of cocoa using selectednear-infrared spectral bands and Deep Learning algorithms.

Responsibilities include:

• Analyze and mathematically model the estimation of physicochemical properties and classification ofmaterials using spectral images and deep learning, within the framework of cocoa beans.
• Obtaining and organising the multispectral cocoa image database.
• Analyze and mathematically model a neural network to obtain the most significant bands in the nearinfrared for the estimation of the physicochemical properties of cocoa.

Achievement: Construction of a low-cost multispectral images camera to analyze the quality of Cocoa.

Professional R&D internship to develop an algorithm for wood-sand differentiation in drone imagery.

Responsibilities include:

• Data cleaning using the Fast Fourier Transform in the Matlab platform.
• Construction of the algorithm for obtaining the physical characteristics of wood and sand.
  

Achievement: Design and implementation of an algorithm with the objective of differentiating wood from sand on the beach of Puerto Velero, Colombia.

Volunteering to develop a protection system against COVID-19.

Responsibilities include:

• Mechanical design of the automatic cabin.
• Design and simulation of the electrical system using PROTEUS.
• Construction of the electronic system.

Achievement: Design and implementation of an automatic booth for the protection of people from COVID-19.

Research project to develop an algorithm for encoding car number plates in Special Topics in Optics.

Responsibilities include:

• Data cleaning using the Fast Fourier Transform in the Matlab platform.
• Development of an interactive application in Matlab to obtaincar number plate digits.

Achievement: Development of an algorithm to improve vehicle number plate resolutions.

R&D internship grant to implementan otitis classification algorithm.

Responsibilities include:

• Obtaining and organising the database using a digital otoscope and Matlab.
• Development of a mathematical model to obtain intrinsic characteristic of Otitis images.
• Creation of neural networks in Matlab to differentiate the different types of Otitis.Achievement: Implementation of an algorithm to differentiate between viral and bacterial otitis.
  

Achievement: Implementation of an algorithm to differentiate between viral and bacterial otitis.

Research project to develop an orchid classification algorithm using Digital Image Processing.

Responsibilities include:

• Obtaining and organising the Orchid database using a Sonic Camera, Microsoft Excel and Matlab.
• Creation of neural networks in Matlab to classify orchids.