Victor Faraggi

M.Sc. in Computer Science

My thesis explores state-of-the-art neural network pruning techniques, empirically examining how pruned models converge to various theoretical bounds within controlled environments. Additionally, I assessed the performance and efficiency of these pruned networks in comparison to their fully dense counterparts.
Funded by the Millennium Institute Foundational Research on Data Science (IMFD) and National Center Artificial Intelligence (CENIA).
Department of Computer Science, Universidad de Chile Santiago, Chile

Computer Engineering & B.Sc. in Engineering

B.Sc. Major: Computer Science
Faculty of Physical & Mathematical Sciences, Universidad de Chile Santiago, Chile

Exchange Student in Computer Science

Enrolled in both graduate and undergraduate courses with an emphasis in Machine Learning.
Faculty of Arts & Sciences, University of Toronto Toronto, Canada

Machine Learning Engineer

Built the first version of the NLP pipelines for automated report generation using LLMs, parsing structured and unstructured data from diverse sources (administrative, legal, operational). Implemented custom evaluation metrics and sophisticated prompt engineering, directly applicable to NLP model evaluation frameworks.
Final version not available. Technologies: Python, Vector DBs, LangChain, Hugging Face Transformers, Pydantic, OpenAI API. PSINET Santiago, Chile

Machine Learning Engineer Intern

Developed and evaluated embedding-based protein language models applied to bioinformatics, providing baseline benchmarks to guide further model improvements.
Technologies: PyTorch, Protein-specific LLMs, Model Evaluation. Protera, BioTech Startup Santiago, Chile

Data Engineer

Analyzed satellite, oceanographic and industrial data in order to detect, prevent harmful algal blooms. Obtained, processed and stored data from different sources to build a recommendation system for mining companies to improve the efficiency of their desalination plants. Data was stored in GCS and processed using BigQuery.
Bloom Alert, Tech Startup Santiago, Chile

Data Science Intern

Built an analytics dashboard that presented analytics for the HR team of a major telecom in South America. Connected more than 10 different internal data-sources.
Designed a scalable data model reflective of the company's organizational structure.
WorldalyticsSantiago, Chile

Understanding Encoder-Decoder Structures in Machine Learning Using Information Measures [Accepted in Elsevier Signal Processing Journal]

Linked information-theoretical concepts to understand effects of encoder-decoder structures in ML architectures. By decomposing the effects of encoder and decoder in a separate manner, we provided theoretical bounds to the performance of models in terms of information measures.
I specifically led the efforts into empirically proving the theoretical results.
Silva, J, Faraggi, V., et al.Elsevier Signal Processing

Characterizing Probabilistic Structure in Learning Using Information Sufficiency [Accepted in IEEE Machine Learning for Signal Processing Workshop]

We use the concept of information sufficiency (IS) to represent probabilistic structures in machine learning (ML). Our main result provides a functional expression that characterizes the class of probabilistic models consistent with an IS encoder-decoder latent predictive structure. This result formally justifies the encoder-decoder forward stages many modern ML architectures adopt to learn latent (compressed) representations in data. Faraggi, V., et al.IEEE Machine Learning & Signal Processing (MLSP), London

Intersecting ML and Information Theory

Led empirical validation of information-theoretic encoder-decoder bounds.
Co-authored Elsevier Signal Processing and IEEE MLSP papers; managed review responses.
Managed lab resources and mentored new students. Research Assistant, Information Decision System LabUniversidad de Chile

Academic exchange

Improving Deep Learning Prediction of Cyclone Wind Speeds with Neural ODEs: Applied Neural ODEs to model the Gradient Wind Balance equation, which is used to estimate wind speeds in tropical cyclones. Victor Faraggi, Gary Tom, Selden Lin
Investigating Climate Downscaling with Generative Super Resolution methods: Worked on the application of generative models to downscaling problems in climate science. We performed an overview of the differences between different models. In particular, we studied the difficulties in training and the transferability of established super-resolution models. Victor Faraggi, Dhananjay Ashok, Aparna Gopalakrishnan Research ProjectsUniversity of Toronto, Canada

ProjectX 2020: ML for Climate Change, U. Toronto

Represented U.Chile in a three-month-long ML research competition, hosted by U.Toronto. The focus of our research went into studying and developing models that could be used for landslide's early warning systems. I particularly worked on data processing and model development. Mentored by three professors: Felipe Tobar (U.Chile), Jorge Perez (U.Chile) and Apurva Narayan(U. Vancouver)
Victor Faraggi, Tomas Rojas, Jose Diaz, Francisco Munoz, Kevin Pinochet, Fabian Lema  Remote

Teaching Assistant

Courses: Information Theory & Application (Electrical Eng. Dept.), Deep Learning, Data Mining, Algorithm Design and Analysis (CS Dept.), Machine Learning (Mathematical Eng. Dept.), Computing Tools for Engineering, Experimental Methods (Physics Dept.), Programming I (CS Dept.)
Universidad de Chile Santiago, Chile

IT Student Director

Organized a four man team that was tasked with maintaining and developing different technological solutions for a 50+ student organization. Coordinated and evaluated the work of my team with the ongoing needs of the whole organization.
Organized job fairs, company talks, and interview workshops connecting students and alumni with industry.
Redes Beauchef, Student Organization Santiago, Chile