MAS Emergence Safety
Formalized MAS emergence misalignment; proposed safety mitigation strategies.
Here you’ll find a curated overview of the papers where I have played a pivotal role, either as the first author or as a contributing author further down the authorship line. My involvement has spanned a variety of activities, from conceptualizing the initial ideas and developing machine learning models, to providing support and insights to my colleagues, or rigorously reviewing and refining the work.
Formalized MAS emergence misalignment; proposed safety mitigation strategies.
Aquarium: Open-source MARL environment for predator-prey studies.
Binary subsegment presorting improves noisy primate sound classification.
Artificial chemistry networks develop predictive models via surprise minimization.
VoronoiPatches improves CNN robustness via non-linear recombination augmentation.
Self-replicating networks collaborate forming higher-level Organism Networks with emergent functionalities.
Journal extension: self-replication, noise robustness, emergence, dynamical system analysis.
Perspective on anomaly detection challenges and future in reinforcement learning.
Self-replicating networks perform tasks, exploring stabilization in artificial chemistry.
Vision Transformer on spectrograms for audio classification, with data augmentation.
Deep BiLSTM classifies primate vocalizations for acoustic wildlife monitoring.
Anomaly detection models for acoustic leak detection in water networks.
Image nets detect acoustic anomalies in machinery via spectrograms.
Pretrained networks extract features for anomalous industrial sound detection.
CNN mask detection in speech using augmented spectrograms.
Analyzing service coverage of parked AVs during downtime (‘meantime’).
PEOC uses policy entropy for OOD detection in deep RL.
Hybrid method segments/fits primitives in large 3D point clouds.
STEVE learns soccer team embeddings from match data for analysis.
Deep learning audio baseline for Interspeech 2019 ComParE challenge.
Neural networks replicating weights, inspired by biology and artificial life.
Unsupervised autoencoder learns spatial context from trajectory data for annotation.