Recommender systems: Random fields
Graphical recommender systems combining local dependency structure, latent factors, context, and social information.
2007-2016
AI Future Archive
Finished projects.
A record of completed programs that helped shape the current AI Future agenda, from recommender systems and conditional random fields to software analytics, generative models, and temporal anomaly detection.
Coverage
- Recommender systems and ordinal preference modeling
- Conditional random fields and structured probabilistic learning
- Restricted Boltzmann machines and disentangled representation learning
- Software analytics, GAN theory, temporal relations, and anomaly detection
Ordinal choice modelling
Models for recommendation that respect the ordinal nature of ratings instead of flattening them into generic numeric targets.
2008-2016
Conditional random fields
Work on feature selection, parameter estimation, and hierarchical CRFs for structured and partially observed data.
2004-2008
Restricted Boltzmann machines
RBM variants for mixed data, matrices, tensors, retrieval, anomaly detection, and representation learning.
2008-2016
Software analytics and automation
Institutional memory for software ecosystems, defect prediction, language modeling, and intelligent developer support.
2015-2019
Representation learning
Matrix-native deep networks, graph architectures, and information-theoretic views of disentanglement.
2017-2020
Understanding GANs
Theory and diagnostics for generalization, catastrophic forgetting, instability, and evaluation in generative models.
2017-2021
Relational and episodic structure in time
Temporal, multi-channel, and episodic reasoning across healthcare, EEG, traffic, and energy data.
2019-2023
Anomaly detection
Mixed-data, temporal, and video-centric anomaly models for ECG, surveillance, and multivariate sequential data.
2017-2021
- Telstra
- Australian Department of Defence
- Industry and academic collaborators across software, health, and data systems
- Structured learning fed into today’s reasoning systems agenda
- Memory and temporal modeling fed into current agent and video work
- Representation, anomaly, and generative modeling informed later science and health platforms