Public relations

I am available for speaking to tech audiences at conferences, symposiums, hackathons, and meetups. My talks are described as very crisp and niche.

Previous talks include: top-tier conferences (AAAI & ICASSP); Deep Learning Barcelona Symposiums; American Society of Human Genetics meetups; SF Tech meetups.


Bio: Margarita Geleta is an AI researcher affiliated to the Berkeley AI Research (BAIR) laboratory and the Stanford Department of Biomedical Data Science (Stanford DBDS), currently pursuing her PhD in Computer Science at the University of California, Berkeley. Previously, she interned at Amazon as an Applied Scientist at the Home Innovation Team, specializing in large-scale generative models, including the development of the state-of-the-art method for GAN inversion. She also developed XR applications for spatial audio manipulation while interning at Dolby Laboratories at the Advanced Technology Group (ATG) in R&D, and co-founded a defense tech startup backed by Founders, Inc.

Event organization

I carried out the organization of dozens tech workshops, and in-person and online career fairs with +200 participants, organization of Europe's biggest student hackathon with +700 participants, and took on positions of responsibility within two associations and the governance of Barcelona School of Informatics (FIB).

In 2023, I was the SF Techstars Community Leader & organized a 3-day startup hackathon in San Francisco, with +150 participants, 17 teams, 5 workshops, 2 talks (Doordash & Meta Research), 32 ideas pitched. Also, I was the social chair at the Computer Science Graduate Entrepreneurs organization at UC Berkeley.

In 2024, I co-organized the AI Rabbit Hole, tier 1 AI conference in San Francisco with +500 participants & hosted an AGI House hackathon with +130 participants.

Check my CV for the full list of events that I organized in the past.

I am open for event organization collaborations.

Research

Currently, I researching investigative genetic genealogy. I am interested in the application of AI/ML techniques for enhanced kinship prediction and genotype simulation conditioned on pedigrees.

My DMs are always open! I love interesting conversations and I am open for collaborations.


I am into genealogy. If you think we are distant cousins, send me an e-mail.

Through my genealogy research, not only have I gained a deep understanding of my family's past, but I have also contributed to the preservation of Pentecostal movement history in Siberia. Because of my strong efforts in genealogical research, in 2023, I have been awarded the Samuel Silver Memorial Scholarship Award and Berkeley Chapter of Sigma Xi Grant-in-Aid-of-Research. Additionally, I was awarded the 2024 Hearts to Humanity Eternal (H2H8) research grant award for the promise of being one of the early AI researchers working on creating a legacy in the field of deep learning for population genetics & investigative genetic genealogy.

Some of my papers are listed below. Check my CV for the full list.

Near Perfect GAN Inversion

Qianli Feng*, Raghudeep Gadde, Viraj Shah, Margarita Geleta, Pietro Perona, Aleix M. Martinez  [Under Review]

Imagine being able to effortlessly manipulate, edit and morph photos with just a few clicks. We have presented a method that achieves nearly perfect reconstructions of images from the latent space of Generative Adversarial Networks (GANs), while preserving strong editability capabilities.

Adversarial Learning for Feature Shift Detection and Correction

Míriam Barrabés*, Daniel Mas Montserrat*, Margarita Geleta, Xavier Giro-i-Nieto, Alexander G. Ioannidis  [NeurIPS 2023] @ New Orleans, LA & [DLBCN 2023]

Think of it as a band of discriminators, trained to spot the difference between two data distributions. Now, these discriminators do more than just point fingers: they lend a hand in both spotting and correcting those features, getting rid of the feature shifts in the dataset.

Autoencoders for Genomics

Margarita Geleta*, Daniel Mas Montserrat, [Carlos D. Bustamante], Xavier Giró-i-Nieto, Alexander G. Ioannidis  [ASHG 2021] & [DLBCN 2021] @ Barcelona, Spain

In this exhaustive study, we explore the many ways in which Variational Autoencoders (VAEs) can be used for dimensionality reduction, classification, compression, imputation and simulation of genomic data. We also dive into interesting discussions about human and canine populations, unlocking new insights into the diverse genetic makeup of our world.

Towards Robust Image-in-Audio Steganography

Jaume Ros Alonso*, Margarita Geleta*, Jordi Pons, Xavier Giro-i-Nieto  [WiCV/CVPR 2023] @ Vancouver, Canada

Want to hide your images in audio with the utmost finesse? Look no further than our enhanced steganography method! We present an improved version of a deep steganographic model for hiding images in audio.

Maestro: A Gamified Platform for Teaching AI Robustness

Margarita Geleta*, Jiacen Xu, Manikanta Loya, Junlin Wang, Sameer Singh, Zhou Li and Sergio Gago Masague   [AAAI 2023] @ Washington DC & [SIGCSE 2023]

Are you passionate about robust AI and want to put your skills to the test? I am the proud maintainer of the Maestro platform and expert tester of the framework. This platform offers a unique opportunity for adversarial AI enthusiasts to engage in a thrilling and educational domain of adversarial attacks and defenses.

PixInWav: Residual Steganography for Hiding Pixels in Audio

Margarita Geleta*, Cristina Punti, Kevin McGuinness, Jordi Pons, Cristian Canton, Xavier Giro-i-Nieto   [ICASSP 2022] @ Singapore & [WiCV/CVPR 2021] & [DLBCN 2022]

Are you ready to unlock the secrets of the art of hiding? Our pioneering work on hiding images within audio waveforms is sure to leave you in awe. It is the ultimate fusion of vision and sound, allowing you to hide information in plain sight and unlock a whole new world of possibilities.

Generative Moment Matching Networks for Genotype Simulation

Maria Perera*, Daniel Mas Montserrat, Míriam Barrabés, Margarita Geleta, Xavier Giró-i-Nieto, Alexander G. Ioannidis  [EMBC 2022] & [ASHG 2022] @ Los Angeles, CA

Discover the power of Generative Moment Matching Networks (GMMNs) in SNP simulation and privacy-preserving data sharing for biobanks. Our research shows GMMNs can mitigate population bias and improve data accuracy, unlocking new insights for genetic research.