Juan F Perez Juste Abascal

Machine Learning

• Notes on machine learning: GitHub repository that provides a summary of the theory with colab notebooks using R and Python.
• Exploratory image analysis - Part 1: Medium blog article on descriptive statistics and exploratory image analysis, with a focus on advanced density plots.
• Exploratory image analysis - Part 2: Medium blog article on embeddings and TensorBoard Embedding Projector.

Architecture NN design

• Spyrit: I contribute to Spyrit, an open-source Python library for deep image reconstruction for single-pixel imaging and other inverse problems.
• Physics-based RNN for unrolling iterative algorithms: Applied to compressed sensing (low-dose) CT (bottom image).

Computer vision

• COVID detection: GitHub repository for training deep learning classifiers with experiment tracking using mlflow and tensorboard (in progress).
• Image classification: Notebook on image classification using transfer learning with Keras.
• Adversarial attacks: Notebook on adversarial attacks and adversarial training with Keras.

Deep learning for high dimensionality

Deep hyperspectral imaging: Publication that proposes higher-order covolutional networks for leveraging high-dimensional data for hyperspectral single-pixel imaging.

NN architecture design

ResPr-UNet-3D-Denoising: GitHub repository for a 3D denoising method that leverages a prior image and a residual learning architecture. Trained using efficient tensorflow pipelines.

Transfer learning human data

Sim2Real transfer approach: Publication proposes and validates a transfer learning approach to leverage realistic computer simulations for solving the material decomposition problem in spectral CT on human data.

AI surveillance

AI surveillance: GitHub repository for a AI surveillance system based on object detection, using several deep learning architectures, and that takes different actions (storing detected images, sending e-mail) upon detection of specific objects (person, ...).

Chat to your data

• Chat to your data: Medium blog post on how to use langchain and openai to chat to any source of data (pdf, doc, spreadsheet, url, audio).
• Talking chatbot: Medium blog post on how to use converse to any source of data (pdf, doc, spreadsheet, url, audio) — a tutorial on leveraging OpenAI's GPT and Whisper models.

NER training

• Named Entity Recognition (NER) for extracting information from pdfs and other sources of data.
• BERT-based NER model trained leveraging Hugging's face transformers library and spaCy.
• NER using prompt-engineering GPT.

Simultaneous motion & image recovery

• Prior motion reconstruction: Data, code and results for prior- and motion-based reconstruction (PRIMOR) method for respiratory gated CT. Motion artefact is removed by simultaneous motion and image reconstruction. Publication.
• Motion estimation CS MRI: This repository contains MATLAB code for B-spline-based compressed sensing (SPLICS). SPLICS generalizes spatiotemporal total variation (ST-TV) by modelling the motion between consecutive frames into the reconstruction framework. Publication.

Compressed sensing

• Compressed-sensing diffusion lung MRI: data, code and results for a novel compressed sensing method presented in the paper Incorporation of prior knowledge of the signal behavior into the reconstruction to accelerate the acquisition of MR diffusion data. Publication.
• Split-Bregman ST-TV MRI: Split Bregman spatiotemporal total variation for cardiac-cine MRI. Publication.
• Split-Bregman PICCS fMRI: Prior image compressed sensing (PICCS) leveraging the Split Bregman for fMRI. Publication.
• Efficient ART Split-Breman: MATLAB code for an iterative reconstruction method that alternates a computationally efficient linear solver (ART) with a fast denoising step based on the Split Bregman formulation. Publication.

Inverse problems - Optimization - Regularization

• Efficient ART Split-Breman: MATLAB code for an iterative reconstruction method that alternates a computationally efficient linear solver (ART) with a fast denoising step based on the Split Bregman formulation. Publication.
• Split-Bregman TV Jacobian: Constrained total variation image reconstruction problem solved using the Split Bregman formulation where the Jacobian is provided as a matrix.
• Publications by topic: Publications on inverse problems, optimization and regularization.

Modelling - FEM

• Anisotropic modelling in EIT: Publication on the influence of anisotropic modelling in electrical impedance tomography of brain function.
• Influence of absorption and scattering on f-DOT: Publication on the influence of absorption and scattering on the quantification of fluorescence diffuse optical tomography.
• Validation of FEM: Publication on the validation of a finite element solution for anisotropic EIT.