My name is Elvis, I am a co-founder of mimic, a Swiss startup working on scalable AI models for robotic manipulation. I am also a Doctoral Student in Machine Learning at ETH Zurich, a Doctoral Fellow at the ETH AI Center, part of Prof. Benjamin Grewe’s group at the Institute of Neuroinformatics and Prof. Robert Katzschmann’s Soft Robotics Lab.
Previously, I received a Master’s in Data Science at ETH Zurich and a Bachelor’s in Computer Science at the University of Milan. I was also an intern at Oracle Labs.
Research
At mimic we are developing recipes for scalable training of generative models of robotic behaviors, and run these models on our own robotic hand hardware.
As for my PhD: I work on Meta-Learning, Multimodality and Robotics. My current research focuses on how sample-efficient learning architectures, pre-training and multi-modal conditioning can be leveraged to train agents that solve complex task such as dexterous robotic manipulation. I am interested in methods for extracting knowledge from pre-trained foundation models that can be useful for training downstream agents - see recent work on repurposing Vision-Language Models as reward functions. I also worked on an application of Diffusion Models for generating neural network parameters (because, why not?).
In the past, I did theoretical research in Batched Bayesian Optimization with Determinantal Point Processes - the kind of research involving proofs of Bayesian bounds on regret. I also worked on neural network-based Differentiable Simulation of a Soft Robotic Fish.
See here a complete list of my publications.
Other Stuff
Learn more about me here. For my hot takes and reading recommendations on deep learning and miscellaneous topics, see my links page.
Contact
You can send me an email at elvis.nava@ai.ethz.ch or DM me on Twitter @elvisnavah.
Latest News
May 2024: Me and my co-founders announced our startup mimic. We raised $2.5M in our pre-seed equity round from Founderful, another.vc, Tiny VC and angel investors. Our fundraising round was featured in Forbes and Sifted.
January 2024: A paper I collaborated on, “Vision-Language Models are Zero-Shot Reward Models for Reinforcement Learning” was accepted to ICLR 2024.
August 2023: My first-author paper “Meta-Learning via Classifier(-free) Diffusion Guidance” was accepted to Transactions on Machine Learning Research.