About me
I’m a machine learning researcher and engineer. I’m broadly interested in probabilistic machine learning, generative models, bayesian statistics, unsupervised and self-supervised learning and inductive priors.
I’m currently working at twig.energy, trying to remove some carbon from the power grid. Prior to that I did a PostDoc in the REAL group at the IT University of Copenhagen working with Sebastian Risi. Before joining ITU I did my PhD at the Cognitive Systems group at the Technical University of Denmark with Ole Winther. Prior to that I was a Staff Engineer and Machine Learning Team lead at Tradeshift.
“What I cannot create, I do not understand.” - Richard Feynman
Publications
- Rasmus Berg Palm, Miguel González-Duque, Shyam Sudhakaran and Sebastian Risi. Variational Neural Cellular Automata ICLR 2022
- Palm, Rasmus Berg, Elias Najarro, and Sebastian Risi. Testing the genomic bottleneck hypothesis in Hebbian meta-learning NeurIPS 2020 Workshop on Pre-registration in Machine Learning. PMLR, 2021.
- González-Duque, Miguel, Rasmus Berg Palm, and Sebastian Risi. Fast Game Content Adaptation Through Bayesian-based Player Modelling arXiv preprint arXiv:2105.08484 (2021).
- Olesen, Thor V.A.N., Dennis T.T. Nguyen, Rasmus Berg Palm and Sebastian Risi. Evolutionary Planning in Latent Space International Conference on the Applications of Evolutionary Computation. 2021.
- Palm, Rasmus Berg and Pola Schwöbel. Justitia ex Machina: The Case for Automating Morals. The Gradient, 2021.
- Grbic, Djordje, Rasmus Berg Palm, Elias Najarro, Claire Glanois and Sebastian Risi. EvoCraft: A New Challenge for Open-Endedness EvoApplications. 2021.
- González-Duque, Miguel, Rasmus Berg Palm, David Ha and Sebastian Risi. Finding Game Levels with the Right Difficulty in a Few Trials through Intelligent Trial-and-Error 2020 IEEE Conference on Games (CoG). IEEE, 2020.
- Palm, Rasmus Berg, Florian Laws, and Ole Winther. Attend, copy, parse end-to-end information extraction from documents 2019 International Conference on Document Analysis and Recognition (ICDAR). IEEE, 2019.
- Palm, Rasmus Berg. End-to-end information extraction from business documents English, PhD thesis (2018).
- Palm, Rasmus Berg, Ulrich Paquet, and Ole Winther. Recurrent Relational Networks Proceedings of the 32nd International Conference on Neural Information Processing Systems. 2018.
- Palm, Rasmus Berg, Ole Winther, and Florian Laws. Cloudscan - a configuration-free invoice analysis system using recurrent neural networks 2017 14th IAPR International Conference on Document Analysis and Recognition (ICDAR). Vol. 1. IEEE, 2017.
- Palm, Rasmus Berg, Dirk Hovy, Florian Laws and Ole Winther. End-to-End Information Extraction without Token-Level Supervision Proceedings of the Workshop on Speech-Centric Natural Language Processing. 2017.
- Palm, Rasmus Berg. Prediction as a candidate for learning deep hierarchical models of data Master thesis - Technical University of Denmark (2012).
Software
I often create software libraries and tools and I love publishing them as open source. It lets me write clean code, think about abstractions and interfaces and ship working software to users, which I enjoy.
- nanograd - The simplest and smallest possible library for autograd. Great for teaching.
- shapeguard - A tiny library, which allows you to very succinctly assert the expected shapes of tensors in a dynamic, einsum inspired way. A great tool for avoiding bugs.
- EvoStrat - A library that makes Evolutionary Strategies (ES) simple to use. It has a flexible and natural interface for ES that cleanly separates the environment, the reinforcement learning agent, the population distribution and the optimization, which allows you to use the standard pytorch
nn.Modules
for policy networks and torch.optim
optimizers for optimization.
- pymc3-quap A quadratic approximation package for PyMC3.
- pytorch-lgssm A clean Linear Gaussian State Space Model for pytorch, which supports sampling and inference using the Kalman filtering algorithm.
- Blayze A very fast and efficient Bayesian Naive Bayes classifier which perfectly incorporates new information in an online learning setting and support both Gaussian and Categorical/Multinomial features.
- DeepLearnToolbox - A now deprecated Matlab toolbox for Deep Learning. I was unwise enough to do my Masters on Deep Learning in Matlab, which didn’t have any support for it, so I made a library for MLPs, CNNs, DBN, etc, implementing everything from the ground up, including the gradients, backprop, etc. It became quite popular. I quickly moved on to Python though (yay autograd!), and deprecated it.
Teaching
I thoroughly enjoy teaching. It challenges me to understand things at a much deeper level, and I find it very rewarding.
Hosted on GitHub Pages — Theme by orderedlist