University

This is a status update on my current project for Advanced Deep Learning for Robotics. The broader question is still whether tactile information can make online adaptation more useful for in-hand manipulation. The project is inspired by RMA-style adaptation: train with access to hidden information about the environment, then learn to infer the useful parts of that information from recent history at test time. The comparison I eventually care about is simple to state: if the controller only gets proprioceptive history, how much can it adapt, and what changes when we also give it tactile or contact information? ...

This project was part of a small group benchmark on foundation models for multiplexed fluorescence microscopy. The general question was simple: if we take recent pretrained models and run them on real multiplexed microscopy data, what do their embeddings actually capture? My part focused on a channel-agnostic masked autoencoder, or CA-MAE, from the paper Masked Autoencoders for Microscopy are Scalable Learners of Cellular Biology. I did not train a new foundation model here, and I did not fine-tune the model. The work was more practical than that: take a pretrained model, make our data fit the input format, run inference on the university cluster, and then look at the resulting embeddings carefully. ...

This project started as a practical follow-up to an introductory deep reinforcement learning course. The goal was simple to state and harder to make work: train deep RL agents to solve a small stochastic pickup-and-delivery task better than a hand-written greedy baseline. The code is available on GitHub: github.com/jclotten/deep-rl-gridworld-benchmark. Why This Project Small gridworlds are often used to explain reinforcement learning because the rules are easy to understand. That does not mean they are automatically easy for deep RL. ...