MAS Emergence Safety
Formalized MAS emergence misalignment; proposed safety mitigation strategies.
Research
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.
My involvement has spanned, from conceptualizing the ideas and developing machine learning models, to providing support and insights to my colleagues.
List of Papers
Aquarium MARL Environment
Aquarium: Open-source MARL environment for predator-prey studies.
Primate Subsegment Sorting
Binary subsegment presorting improves noisy primate sound classification.
Emergent Social Dynamics
Artificial chemistry networks develop predictive models via surprise minimization.
Autoencoder Trajectory Compression
LSTM autoencoder better DP for trajectory compression (Fréchet/DTW).
Voronoi Data Augmentation
VoronoiPatches improves CNN robustness via non-linear recombination augmentation.
Organism Network Emergence
Self-replicating networks collaborate forming higher-level Organism Networks with emergent functionalities.
Extended Self-Replication
Journal extension: self-replication, noise robustness, emergence, dynamical system analysis.
RL Anomaly Detection
Perspective on anomaly detection challenges and future in reinforcement learning.
RNN Memory Limits
Investigated memory limits of RNNs in recalling uncorrelated sequences.
Tasked Self-Replication
Self-replicating networks perform tasks, exploring stabilization in artificial chemistry.
Audio Vision Transformer
Vision Transformer on spectrograms for audio classification, with data augmentation.
Primate Vocalization Classification
Deep BiLSTM classifies primate vocalizations for acoustic wildlife monitoring.
Acoustic Leak Detection
Anomaly detection models for acoustic leak detection in water networks.
Sound Anomaly Transfer
Image nets detect acoustic anomalies in machinery via spectrograms.
Anomalous Sound Features
Pretrained networks extract features for anomalous industrial sound detection.
Surgical-Mask Detection
CNN mask detection in speech using augmented spectrograms.
AV Meantime Coverage
Analyzing service coverage of parked AVs during downtime (‘meantime’).
PEOC OOD Detection
PEOC uses policy entropy for OOD detection in deep RL.
3D Primitive Segmentation
Hybrid method segments/fits primitives in large 3D point clouds.
Soccer Team Vectors
STEVE learns soccer team embeddings from match data for analysis.
Deep Audio Baselines
Deep learning audio baseline for Interspeech 2019 ComParE challenge.
Neural Self-Replication
Neural networks replicating weights, inspired by biology and artificial life.
Learned Trajectory Annotation
Unsupervised autoencoder learns spatial context from trajectory data for annotation.