Ali Zia, machine learning researcher

I build machine learning models that keep working when data is scarce, noisy or shifted.

Much of my work brings topology into deep learning: persistent homology, cell and simplicial complexes and hypergraphs give a network structure it can rely on when pixel-level cues fail. I pair this with hyperspectral and multimodal sensing, where I started, and apply it to crops and food, medical imaging, industrial inspection and radar.

I am a Research Fellow at La Trobe University with adjunct appointments at CSIRO and the Australian National University. I co-lead the CSIRO ML Workshop Series, supervise PhD and honours students, and received my PhD from Griffith University for work on 3D computer vision from hyperspectral images.

One point cloud at four scales. As ε grows, edges and triangles fill in; small loops die quickly while the large one persists. That persistence is the signal topological deep learning keeps. Colour follows the visible spectrum, a nod to hyperspectral imaging.

News

Oct 2026Code for TopoOT, TopoTTA, MAFR and MM-HCAN is now public.
Oct 2026TopoOT accepted at NeurIPS 2026 as a main-conference poster. See you in Sydney.
Oct 2026“Learnable Persistent Topology for Multi-View Radar Semantic Segmentation” accepted at the NeurIPS 2026 workshop on Symmetry and Geometry in Neural Representations.
2026Our tutorial Topological Deep Learning for Robust Representation Learning was accepted at ACML 2026 (Melbourne, 1 December), with Usman Ali and Abdelwahed Khamis.
2026TopoTTA published in IEEE TPAMI.
2026Papers at ECCV 2026 (structurally consistent multi-view radar segmentation) and IEEE JBHI (expert fusion for blastocyst assessment and IVF decision support).
2026MM-HCAN published in the IEEE Sensors Journal; HyperEcho, on higher-order representations for automotive radar, at IEEE PerCom 2026.
2026Seed funding as lead for SpectraResist-AI, detecting herbicide resistance in ryegrass, and a La Trobe Research Infrastructure grant for a high-speed camera.
2025MAFR and Split-Fuse-Transport at DICTA 2025; Hierarchical Spatial Mamba at ACPR 2025; BioNet and NeFF at WACV 2025.
2024Our review of topological deep learning appeared in Artificial Intelligence Review, and TopoX in the Journal of Machine Learning Research.
2023Runner-up paper award at DICTA 2023 for contrastive-aware ViT for weakly supervised segmentation.

Research

components (H₀)
loops (H₁)

Points far from the diagonal are features that persist across scales.

Topological deep learning

Networks that reason over shape as well as signal: persistence-based pseudo-labels, optimal transport between persistence diagrams, and message passing on hypergraphs and cell complexes. This includes a widely read review of the field and the TopoX software suite.

Hyperspectral and multimodal sensing

Hundreds of narrow bands reveal what RGB cannot: protein content in chickpea flour, contamination in meat, early disease in sugarcane and herbicide resistance in ryegrass. I work on weak-signal unmixing, spectral 3D vision and fusion with lidar and point clouds.

Anomaly segmentation results on a bottle and a transistor: input image, ground truth, anomaly heat map, threshold baseline, TTT4AS and TopoTTA. TopoTTA's masks are closest to the ground truth.

Reliable AI under real conditions

Test-time adaptation and out-of-distribution robustness for models deployed where the data moves: industrial anomaly segmentation, radar perception, motor fault diagnosis and embryo assessment for IVF.

See all research projects

Featured work

NeurIPS 2026

Topology-Aware Optimal Transport for Source-Free Test-Time Adaptation in Anomaly Segmentation

A. Zia, U. Ali, U. Ramzan, A. Rehman, A. Khamis, W. Xiang

TopoOT chains optimal transport across persistence diagrams from different thresholds and filtrations, then trains a light head online on the features that stay stable. Up to +24.1% mean F1 over the next best method on 2D benchmarks.

arXiv Code

IEEE TPAMI 2026

Learning Topology-Aware Representations via Test-Time Adaptation for Anomaly Segmentation

A. Zia, U. Ali, A. Rehman, U. Ramzan, K. Han, M. Faheem, S. Qureshi, W. Xiang

TopoTTA replaces threshold heuristics with persistent homology on anomaly maps, producing pseudo-labels that keep defects connected. About 15% higher F1 on average across five 2D and 3D benchmarks.

Paper arXiv Project page Code

IEEE Sensors Journal 2026

Hypergraph Contrastive Sensor Fusion for Multimodal Fault Diagnosis in Induction Motors

U. Ali, A. Zia, W. Ali, U. Ramzan, A. Rehman, M. T. Chaudhry, W. Xiang

MM-HCAN builds hypergraphs within and across sensor modalities and trains them contrastively, diagnosing bearing, stator and rotor faults together. Up to 99.82% accuracy with strong cross-domain transfer.

arXiv Code

DICTA 2025

2D–3D Feature Fusion via Cross-Modal Latent Synthesis and Attention-Guided Restoration for Industrial Anomaly Detection

U. Ali, A. Zia, A. Rehman, U. Ramzan, Z. Hassan, T. Sattar, J. Wang, W. Xiang

MAFR fuses RGB and point clouds in one latent space and finds anomalies by how badly each modality restores. Mean I-AUROC of 0.972 on MVTec 3D-AD.

arXiv Code

Selected publications

2026

Learning Topology-Aware Representations via Test-Time Adaptation for Anomaly SegmentationA. Zia, U. Ali, A. Rehman, U. Ramzan, K. Han, M. Faheem, S. Qureshi, W. XiangIEEE TPAMIpaper arXiv code
Topology-Aware Optimal Transport for Source-Free Test-Time Adaptation in Anomaly SegmentationA. Zia, U. Ali, U. Ramzan, A. Rehman, A. Khamis, W. XiangNeurIPSarXiv code
Learning Structurally Consistent Representations for Multi-view Radar Semantic SegmentationA. Zia, M. U. Ramzan, A. Khamis, U. Ali, A. RehmanECCV
Expert Fusion Network for Automated Blastocyst Morphology and IVF Decision SupportA. Zia, S. Qureshi, F. Ansarizadeh, M. Fozan, E. Kamawal, S. Anwer, et al.IEEE JBHI
Hypergraph Contrastive Sensor Fusion for Multimodal Fault Diagnosis in Induction MotorsU. Ali, A. Zia, W. Ali, U. Ramzan, A. Rehman, M. T. Chaudhry, W. XiangIEEE Sensors JournalarXiv code
HyperEcho: Structured Higher Order Representation for Automotive Radar Semantic SegmentationA. Khamis, M. U. Ramzan, U. Ali, A. ZiaIEEE PerCom
A Review on Vision-Centric Coarse to Fine-Grained Animal Action RecognitionA. Zia, R. Sharma, A. Khamis, U. Ali, X. Li, M. Husnain, N. Shafi, S. Anwar, et al.AI ReviewarXiv
WS-Net: Weak-Signal Representation Learning and Gated Abundance Reconstruction for Hyperspectral Unmixing via State-Space and Weak Signal Attention FusionZ. Long, A. Zia, G. Fu, V. Rolland, J. ZhouarXiv preprintarXiv
Representation Geometry as a Diagnostic for Out-of-Distribution RobustnessA. Zia, F. HazratianarXiv preprintarXiv

2025

2D–3D Feature Fusion via Cross-Modal Latent Synthesis and Attention-Guided Restoration for Industrial Anomaly DetectionU. Ali, A. Zia, A. Rehman, U. Ramzan, Z. Hassan, T. Sattar, J. Wang, W. XiangDICTAarXiv code
Hierarchical Spatial Mamba Framework for Point Cloud ClassificationY. Sun, A. Zia, Z. Long, Z. Qiu, W. Xiang, J. ZhouACPR
BioNet and NeFF: Crop Biomass Prediction from Point Clouds to Drone ImageryX. Li, Z. Hayder, A. Zia, C. Cassidy, S. Liu, W. Stiller, E. Stone, W. Conaty, et al.WACVarXiv
Advances in Machine Learning for Wetland Classification: A Comprehensive Survey of Methods and ApplicationsD. Effah, A. Zia, M. Awrangjeb, Y. Gao, K. SarpongAI Review
Unlocking Chickpea Flour Potential: AI-Powered Prediction for Quality Assessment and Compositional CharacterisationA. Zia, M. Husnain, S. Buck, J. Richetti, E. Hulm, J. P. Ral, V. Rolland, X. SiraultCurr. Res. Food Sci.

2024

Topological Deep Learning: A Review of an Emerging ParadigmA. Zia, A. Khamis, J. Nichols, U. B. Tayab, Z. Hayder, et al.AI Reviewpaper arXiv
TopoX: A Suite of Python Packages for Machine Learning on Topological DomainsM. Hajij, M. Papillon, F. Frantzen, J. Agerberg, I. AlJabea, R. Ballester, et al., including A. ZiaJMLRpaper arXiv code
A New Hyperspectral Unmixing Benchmark for Weak Signal Meat Contamination DetectionZ. Long, A. Zia, J. Nelis, V. Rolland, J. ZhouDICTA

2023

ICML 2023 Topological Deep Learning Challenge: Design and ResultsM. Papillon, M. Hajij, A. Myers, H. Jenne, J. Mathe, T. Papamarkou, et al., including A. ZiaPMLRpaper arXiv
CVB: A Video Dataset of Cattle Visual BehaviorsA. Zia, R. Sharma, R. Arablouei, G. Bishop-Hurley, J. McNally, N. Bagnall, et al.CVPR-WarXiv
Contrastive-Aware ViT for Weakly Supervised Semantic SegmentationP. Pativada, Z. Hayder, A. Zia, N. BarnesDICTA Runner-up paper award

Earlier

Exploring Chromatic Aberration and Defocus Blur for Relative Depth Estimation from Monocular Hyperspectral ImageA. Zia, J. Zhou, Y. GaoIEEE TIPpaper code
Spectral-Spatial Scale Invariant Feature Transform for Hyperspectral ImagesS. L. Al-Khafaji, J. Zhou, A. Zia, A. W. C. LiewIEEE TIP
3D Reconstruction from Hyperspectral ImagesA. Zia, J. Liang, J. Zhou, Y. GaoWACVpaper code
3D Plant Modelling via Hyperspectral ImagingJ. Liang, A. Zia, J. Zhou, X. SiraultICCV-W

All 63 publications, with search and filters

Code

TopoOTNeurIPS 2026

Topology-aware optimal transport for source-free test-time anomaly segmentation.

TopoTTAIEEE TPAMI 2026

Persistent-homology pseudo-labels for test-time adaptation in anomaly segmentation.

MM-HCANIEEE Sensors Journal 2026

Hypergraph contrastive fusion of vibration signals and spectrograms for motor fault diagnosis.

MAFRDICTA 2025

RGB and point cloud fusion with attention-guided restoration for industrial anomaly detection.

Plus 7 hyperspectral computer vision repositories in MATLAB from my PhD and related projects.

All code github.com/alizia369