Javier Salazar Cavazos

Javier Salazar Cavazos

EE PhD Candidate @ University of Michigan

Posts by Collection

certifications

industry

Olsson - Assistant Electrical Engineer

2020-2021. Designed electrical power distribution systems for data center clients and analyzed systems for short circuit analysis, arc flash mitigation, and breaker setting coordination. Moreover, created and revised electrical design plans (from audits) for mainly data center configurations as seen below.

Brohard Architecture - Architectural Draftsman

2017-2019. Designed 3D building/object models using Autodesk Revit (pre-construction plan process). Created floor plans, ceiling plans, interior/exterior elevation plans, and perspective plans for commercial and residential projects. Image below is a project I assisted Mr. Brohard with.

Premier Surveying - AutoCAD Technician

2015-2017. Drafted residential title surveys using Autodesk AutoCAD Civil3D and analyzed field data, city plats, FEMA flood maps, title commitments, and easements to construct drawings as illustrated below.

portfolio

Autoencoders & CLIP models

Implemented Autoencoder model for image compression and representational learning. Moreover, the CLIP model is explored for query tasks as shown below. Press blue link above for more details.

Diffusion Models

Diffusion-based neural networks have gained traction in recent years for tasks such as generative AI. In this project, I implement a diffusion-based network for image generation using CIFAR10 dataset. Press blue link above for more details.

Laplacian Pyramids: Image Blending

Laplacian pyramids are useful to capture information of images at various scales. In this project, I use this to blend images together as seen below. Press blue link above for more details.

Panorama Stitching with ORB/SIFT Features

Panorama stitching involves joining images taken at different angles. This is nontrivial since it’s challenging to line them up. By using ORB features (key landmarks), one can use this to find the affine mapping related to the two images as shown below. Press blue link above for more details.

VGG Network: Scene Recognition

In this project, I perform classification (scene recognition) with a VGG-type neural network using the miniplaces dataset to recognize the scene of images. Press blue link above for more details.

publications

ALPCAH: Sample-wise Heteroscedastic PCA with Tail Singular Value Regularization

Published in Sampling Theory and Applications (SampTA), 2023

Principal component analysis (PCA) is a key tool in the field of data dimensionality reduction that is useful for various data science problems. However, many applications involve heterogeneous data that varies in quality due to noise characteristics associated with different sources of the data. Methods that deal with this mixed dataset are known as heteroscedastic methods. Current methods like HePPCAT make Gaussian assumptions of the basis coefficients that may not hold in practice. Other methods such as Weighted PCA (WPCA) assume the noise variances are known, which may be difficult to know in practice. This paper develops a PCA method that can estimate the sample-wise noise variances and use this information in the model to improve the estimate of the subspace basis associated with the low-rank structure of the data. This is done without distributional assumptions of the low-rank component and without assuming the noise variances are known. Simulations show the effectiveness of accounting for such heteroscedasticity in the data, the benefits of using such a method with all of the data versus retaining only good data, and comparisons are made against other PCA methods established in the literature like PCA, Robust PCA (RPCA), and HePPCAT. Code available at https://github.com/javiersc1/ALPCAH

Recommended citation: J. A. S. Cavazos, J. A. Fessler, and L. Balzano. ALPCAH: Sample-wise heteroscedastic PCA with tail singular value regularization. In Fourteenth International Conference on Sampling Theory and Applications, 2023. http://javiersc1.github.io/files/paper_alpcah.pdf

talks

Image Segmentation of Pelvic Floor Features & Slings/Meshes

Published:

A poster session given at Tufts University for the Conference on Modern Challenges in Imaging. Researched machine learning methods of identification and segmentation of meshes & slings, along with common pelvic floor features, in MR images for radiologist assistance in 3D visualization-guided surgery planning. Press image below for more details. segmentation

Wheelchair Dynamometer

Published:

A poster session at the UTASixty event regarding my captstone project during undergrad. Fitness is an important aspect in today’s society as a rapidly growing industry. Many people exercise to live a healthier lifestyle but also for sport performance. Several machines have been invented to exercise the body in different ways and help stay fit although there are very few machines specifically meant for wheelchair users. The wheelchair dynamometer is designed for wheelchair users to exercise, like an able-bodied individual, when needed as well as track performance parameters important in their field of sport. With a resistance setting, the dynamometer can accommodate a wide range of users at different fitness levels that can see workout/performance parameters through interaction with the included display. dyno

Sparse Subspace Clustering (SSC)

Published:

A presentation on a subspace clustering method known as SSC that is useful for Union of Subspaces (UoS) problems. Press image below for full PDF. SSC

Heteroscedastic PCA (ALPCAH)

Published:

A presentation on a novel subspace basis approximation method known as ALPCAH that is useful for various tasks like Dynamic MRI reconstruction, lesion detection, and more. Specifically, useful over PCA when the data is noisy (heteroscedastic). Image below illustrates the problems with heteroscedastic data and the regular PCA algorithm. Press image below for full PDF. SSC

ALPCAH for Union of Subspaces (ALPCAHUS)

Published:

A presentation on a novel subspace clustering method known as ALPCAHUS that is useful for Union of Subspaces (UoS) problems when the dataset itself is heteroscedastic (noisy). Press image below for full PDF. ALPCAHUS

Deep Learning 4 Dummies 1

Published:

A presentation on the basics of deep learning from the linear algebra perspective. Press image below for full PDF. DL4D1

teaching

Award: Math Academic Excellence

Award, University of Texas, Math Department, 2019

The Math Academic Excellence Award is a departmental award granted by the math department from UTA to students that demonstrate excellence in math through course grades.

Award: Rackham Merit Fellow

Fellowship Award, University of Michigan, Rackham Graduate School, 2021

The Rackham Merit Fellowship (RMF) Program helps sustain the academic excellence and inclusiveness of the Michigan graduate community, one that embraces students with diverse experiences and goals, and who come from many educational, cultural, geographic, and familial backgrounds. By offering financial assistance to those students who might not otherwise have access, we aim to reduce disparities in graduate education. We also aim to promote the values of diversity and inclusion by encouraging the admission and funding of students who represent a broad array of life experiences and perspectives, because this enhances the quality of the intellectual environment for all students.

Teaching: EECS351 - Digital Signal Processing

Undergraduate course, Graduate Student Instructor, University of Michigan, EECS Department, 2023

EECS351 is a Digital Signal Processing (DSP) course where I lead a weekly discussion reinforcing course material through further mathematical analysis and examples. Moreover, I moderated the Canvas/Piazza forum section to answer student questions related to homework and general course questions. This is besides the usual GSI expectations of dealing with homework assignments and tests.

Teaching: EECS551 - Matrix Methods (Linear Algebra)

Graduate course, Graduate Student Instructor, University of Michigan, EECS Department, 2023

EECS551 is a linear algebra based course called matrix methods where I lead a weekly discussion that covers coding algorithms in Julia for classification tasks using linear algebra techniques. Moreover, we go through example questions to reinforce course material and exam preparation. Moreover, I moderated the Canvas/Piazza forum section to answer student questions related to homework and general course questions. This is besides the usual GSI expectations of dealing with homework assignments, tests, and office hours.

undergrad

Image Segmentation of Pelvic Floor Features & Slings/Meshes

Pelvic floor disorders, which include pelvic organ prolapse, affect nearly 1 in 4 women in the US. Each year more than 300,000 surgical procedures are performed in the US to treat these disorders. Most common surgical repair procedures of pelvic floor include placement of synthetics implants such as urethral slings and vaginal meshes. Unfortunately, postoperative complications of these procedures (e.g. dyspareunia, chronic pain, extrusion, or recurrent infection) are not rare, and their treatment may require follow-up surgeries to remove the implants. In these cases, imaging of the pelvic floor is a vital necessity for surgeons planning the procedure, seeking to identify the implants, their relative location and distance from various organs. The process is further complicated by the poor contrast between the implants and scar tissue, as well as the need for the surgeons to mentally generate a 3D volumetric image from 2D projections generated by standard MRI. This work aims to address some of these issues and assist surgeons by improving the currently available imaging techniques. In particular, we study the possibility of using multi-atlas segmentation (MAS) to automatically segment 2D MR images of pelvic floor, as well as utilizing such segmentations to create 3D semi-transparent volumetric images that can be rotated and zoomed with a mouse on the computer screen. Press blue link above for poster PDF.

REU: Estimating Iceberg Drag Coefficients w/ Bayesian Inference Techniques

Researched the mathematics of Bayesian inference methods for iceberg prediction tracking by modeling ordinary differential equations, simulated the model using Markov chain Monte Carlo sampling techniques, and predicted drag coefficients for iceberg prediction. Press blue link above for presentation PDF.

REU: Mathematics Behind CT Image Reconstruction

Researched and implemented the mathematics of computerized tomography such as Radon transform and filtered backprojection for image reconstruction of artificial phantoms. Press blue link above for PDF report.

WiFi Antenna Design w/ ANSYS HFSS

Designed, optimized, and tested 2.4GHz/5GHz wifi antenna using ANSYS HFSS and electronic equipment for an electromagnetics course. Press blue link above for PDF report.

Capstone Project: Wheelchair Dynamometer

Fitness is an important aspect in today’s society as a rapidly growing industry. Many people exercise to live a healthier lifestyle but also for sport performance. Several machines have been invented to exercise the body in different ways and help stay fit although there are very few machines specifically meant for wheelchair users. The wheelchair dynamometer is designed for wheelchair users to exercise, like an able-bodied individual, when needed as well as track performance parameters important in their field of sport. With a resistance setting, the dynamometer can accommodate a wide range of users at different fitness levels that can see workout/performance parameters through interaction with the included display. Press blue link above for PDF report.

Machine Learning Algorithms

Machine learning is a hot topic where one wants to learn from data to achieve a certain task. I have implemented various algorithms from LDA, interpolation, k-NN, clustering, MLP, and more. Press blue link above for more details and all algorithms I worked on.

Stochastic Models & Simulations

Stochastic modeling is useful for many real-world scenerios. From random number generation, Markov chains, random walks, and Black-Scholes option pricing, and more. Press blue link above for more details and all algorithms I worked on.

Numerical Analysis Algorithms

Numerical Analysis is the study of mathematical numerical algorithms such as ODE solvers, Taylor approximations, matrix solvers, root finders, numerical integration, and more. Press blue link above for more details and all algorithms I worked on.