Navid Shokouhi
ML Operations and Research
I’m a machine learning researcher and practitioner trained in probabilistic approaches, information theory, and statistical learning. My work leverages these skills to develop machine learning algorithms. I’m also deeply interested in ML Ops – the practice of productionizing robust, scalable machine learning software systems.
Here’s a look at my current projects, interests, and past work. While some recent code repositories may be private (pending publication or IP reasons), I welcome discussions and potential collaborations.
Projects
2024
Energy Consumption, Emissions, and Sustainability of the Transformer Architecture for Large Language Models (code).
Enterprise-wide Responsible AI/ML.
Software Design Patterns for modern MLOps – Foundation Models (code)(talk)
2023
Software Design Patterns for Robust Machine Learning Productionization (code)(talk).
Identifying Risk of Falls for People with Parkinson’s Disease (submitted).
2022
Transfer learning for Human Activity Recognition.
Step Count Accuracy estimation for Parkinson’s Disease patients (paper)
2021
Model monitoring through Microservices: An AWS Lambda Approach (code).
2020
Causal Bayesian Networks: Specifically focused on Causal Discovery Algorithms.
Boosting: PAC learning, AdaBoost, and their relation to maximum likelihood.
2019
Adaptive robust Neural Network trianing via alpha-divergence (paper)(codeocean)
Canonical Correlation Analysis Dimensionality Estimation (paper)(codeocean)
Sparse Canonical Correlation Analysis (private repo)
2018
Sparse Principal Component Analysis (code) (paper) (codeocean)
Sparse representation using Dictionary Learning (slides)
Concurrent Spatial and Temporal alignment of multimodal data (private repo)
Model order selection for CCA (code)
Estimating Dimensionality of PCA (paper)(code)
Estimating Dimensionality of ICA (paper)(code)
Robust HRF estimation (fNIRS) (private repo)
Multi-target Speaker Identification (private repo)
2017
On the various forms of training Radial Basis Function Neural Networks (code)
Model order selection (code) (paper)
2D-Whitening for face recognition (paper)
Speaker Diarization (python toolbox)
Previous Work (2011-2016)
PhD (UT Dallas, 2017) dissertation: Speaker Recognition and Diarization in Multi-Speaker Signals. A link to my dissertation can be found here.
Some code excerpts from my PhD work:
Overlapped Speech Detection (code) and (paper)
UTDallas-CRSS Speaker Diarization tool-box – (private)
Speech Activity Detection for UT-Dallas projects
light-weight Speech Activity Detection
Speaker verification (includes code additions to Kaldi: NDA, Clustering, DCF calculations
Teaching
Applied Data Science MAST30034, The Univ. of Melbourne, 2022
Image Processing ELEN90076, The Univ. of Melbourne, 2017
Notes
Kernel interpretation of self-attention Mechanism in Transformer Architectures
Back-propagation: a dynamic programming perspective
Asymptotic difference between ML and empirical ML
An interesting example on the short-comings of ML
Canonical correlation coefficients as a measure of affine similarity
Note on the significance of Gaussian distributions from an ML perspective
Chi-squared distributions, statistical measures, and information theory (in progress)
Notes on fundamentals of Variational Bayes
Notes on Akaike’s Information Criterion
Calculating the Cummulative Match Curve (code)
Cross-validation from an information-theoretic perspective (in progress)
Note on CCA: geometric interpretations
Interesting Reading Material
Other
Online Stack Exchange activity
