Automatic species identification from images for Aotearoa
Research and Publications
As TAIAO continues to grow as a national hub for environmental data science, our expanding community and research outputs reflect a shared commitment to open collaboration, data sovereignty, and supporting the future of environmental AI.
Journal Publications
High‑quality journal articles that have been formally reviewed by experts. These papers represent TAIAO’s core scientific contributions to environmental AI, machine learning, and environmental science.
Disagreement-Based Semi-Supervised Learning for Sparsely Labeled Evolving Data Streams
Introducing SLEADE, a semi‑supervised ensemble method that uses disagreement‑based learning and unsupervised drift detection to improve classification accuracy in evolving data streams with very sparse labels.
Linear adaptive filtering for regression in data streams
A lightweight regression approach for data streams that uses recursive linear adaptive filters to deliver fast, accurate, and drift‑aware predictions without ensembles or explicit drift detectors.
A comparative study of four deep learning algorithms for predicting tree stem radius measured by dendrometer: A case study
A deep learning comparison showing that accurate tree stem radius prediction is possible even with drastically reduced dendrometer sampling frequency.
Accelerated Weka: GPU Machine Learning with Weka Workbench
A GPU‑accelerated extension of the Weka Workbench that delivers massive speedups while preserving Weka’s simplicity and accessibility.
RMIDDM: an unsupervised and interpretable concept drift detection method for data streams
RMIDDM introduces an unsupervised, interpretable method for detecting concept drift in data streams by leveraging relative mutual information to identify meaningful changes in evolving distributions.
Automatic species identification from images for Aotearoa
An adaptive region proposal network that dynamically generates oriented anchors and labels to improve accuracy and efficiency in remote sensing object detection.
Bayesian Stream Tuner: Dynamic Hyperparameter Optimization for Real-Time Data Streams
This paper introduces Bayesian Stream Tuner, a hyperparameter optimisation method designed for real‑time, non‑stationary data streams. BST adapts as the data distribution shifts, using online Bayesian modelling and drift detection to continually update hyperparameter choices.
Streaming Isolation Forest
Streaming Isolation Forest is an online anomaly‑detection method that incrementally updates isolation trees to detect anomalies in real time as data streams evolve.
Auto-Reg: A Dynamic AutoML Framework for Streaming Regression
Auto‑Reg is a dynamic AutoML system that continually updates and searches for the best regression pipelines as data streams evolve, allowing it to stay accurate and efficient even when the data changes over time.
Remote Sensing for Water Quality: A Multi-Task, Metadata-Driven Hypernetwork Approach
The paper shows that a metadata‑driven multi‑task hypernetwork can generate flexible, lake‑specific models that significantly improve water‑quality prediction across diverse and data‑limited remote‑sensing tasks.
ASML-REG: Automated Machine Learning for Data Stream Regression
ASML‑REG is an automated machine‑learning system designed for data‑stream regression that continually selects, tunes, and updates models in real time so it can stay accurate and efficient even as the incoming data evolves.
Time-Evolving Data Science and Artificial Intelligence for Advanced Open Environmental Science (TAIAO) Programme
TAIAO demonstrates how adaptive, streaming AI methods and open‑source tools can transform environmental science by enabling real‑time, scalable, and transparent modelling of Aotearoa’s dynamic environmental systems.
Conference Publications
Research presented at national and international conferences. These works share emerging ideas, early findings, and innovative methods developed by the TAIAO team.
Practical Machine Learning for Streaming Data
A tutorial‑style conference contribution that introduces practical methods, tools, and challenges in applying machine learning techniques to real‑time, evolving data streams.
Time-evolving data science and artificial intelligence for Advanced Open Environmental Science (TAIAO) programme
A high‑level overview of the TAIAO programme, outlining how time‑evolving data science and AI are being advanced to support open, impactful environmental research in Aotearoa.
Dynamic Ensemble Member Selection for Data Stream Classification
A new framework that boosts data stream classification by dynamically selecting the most effective ensemble members for each prediction.
Practical Machine Learning for Streaming Data
A tutorial‑style conference contribution that introduces practical methods, tools, and challenges in applying machine learning techniques to real‑time, evolving data streams.
Time-evolving data science and artificial intelligence for Advanced Open Environmental Science (TAIAO) programme
A high‑level overview of the TAIAO programme, outlining how time‑evolving data science and AI are being advanced to support open, impactful environmental research in Aotearoa.
Dynamic Ensemble Member Selection for Data Stream Classification
A new framework that boosts data stream classification by dynamically selecting the most effective ensemble members for each prediction.
Featured Publications
Click on to one of the images to find out more about the publications we are featuring for this month.
Linear adaptive filtering for regression in data streams
The University of Waikato
University of Canterbury
The University of Auckland
Victoria University of Wellington
MetService
Beca
