Professor Giovanni Montana

I am Professor of Data Science at the University of Warwick, jointly appointed to the Department of Statistics and Warwick Manufacturing Group (WMG). My main research interest is reinforcement learning for sequential decision-making under uncertainty in dynamic environments: how agents can learn good policies from limited or previously collected data, act safely under constraints, plan with learned models of their environment, exploit structure in large action spaces, and, in multi-agent systems, learn to communicate and coordinate. Applications include drug discovery, experimental design and industrial automation. A newer line of work addresses machine unlearning, the removal of specific training data from deployed models with verifiable guarantees. Alongside this, my group develops machine learning for medical imaging that has been validated across hospitals and released as open-source models.

I joined Warwick in 2018 from King’s College London, where I was Professor of Biostatistics and Bioinformatics in the Department of Biomedical Engineering (2013–2018), having previously been Reader in Statistics in the Mathematics Department at Imperial College London (2005–2013). Before that I completed a PhD in computational statistics at Warwick, worked as a postdoctoral researcher at the University of Chicago and as a data scientist at Bristol-Myers Squibb. Across those years I developed statistical learning methods for medical imaging, neuroimaging and genomics, addressing the challenges of high-dimensional data with complex structure.

My research has been supported by a UKRI Turing AI Acceleration Fellowship (2021–2025), which allowed me to focus on reinforcement learning methodology, and by grants from EPSRC, MRC, CRUK, Wellcome and Innovate UK, among others. I am a Chartered Statistician and Fellow of the Royal Statistical Society and have served on grant panels for EPSRC, MRC, the Royal Society and CRUK. I have taught courses from introductory statistics to PhD-level machine learning.

A full listing of publications can be found on my Google ScholarLink opens in a new window profile, and open-source software and pre-trained models are available on GitHubLink opens in a new window and HuggingFaceLink opens in a new window.

Research interests and selected publications

Across these areas the common problem is learning from high-dimensional, structured data in order to act on it, whether the action is an inference, a prediction or a sequence of decisions.

Reinforcement learning and sequential decision-making (2019–present). Work since 2019 develops reinforcement learning methods for settings where interaction is costly: learning from limited or offline data, safe decision-making under constraints, incorporating structural knowledge (e.g. relational graphs and factorised actions) to improve sample efficiency and coordination, model-based planning with learned dynamics, and multi-agent systems. This work has produced new algorithms for learning policies from fixed datasets, for planning with learned models, for exploiting the structure of large action spaces, and for communication and coordination between learning agents. Much of this methodological development has been motivated by unlocking specific applications.

• Zhu, Jin & Montana (2026) When lifelong novelty fails: coordination breakdown in decentralised MARL. Transactions on Machine Learning Research
• Neggatu, Houssineau & Montana (2026) Action-free offline-to-online RL via discretised state policies. ICLR
• Jin & Montana (2026) Partial action replacement: tackling distribution shift in offline MARL. AAAI
• Wang & Montana (2025) Retrosynthesis planning via worst-path policy optimisation in tree-structured MDPs. NeurIPS
• Zhu et al. (2025) Uncertainty-based smooth policy regularisation for reinforcement learning with few demonstrations. NeurIPS
• Neggatu, Houssineau & Montana (2025) Evaluation-time policy switching for offline reinforcement learning. AAMAS
• Utke, Houssineau & Montana (2025) Investigating relational state abstraction in collaborative MARL. AAAI
• Jin, Wei & Montana (2025) Achieving collective welfare in multi-agent reinforcement learning via suggestion sharing. Machine Learning
• Wang, Jin & Montana (2025) Learning on one mode: addressing multi-modality in offline reinforcement learning. ICLR
• Beeson, Couper & Montana (2025) Offline reinforcement learning for community-acquired pneumonia management: a feasibility study. ECML PKDD
• Hepburn, Jin & Montana (2025) State-constrained offline reinforcement learning. Transactions on Machine Learning Research
• Zhu et al. (2024) Mitigating relative over-generalization in multi-agent reinforcement learning. Transactions on Machine Learning Research
• Beeson, Ireland & Montana (2024) An investigation of offline reinforcement learning in factorisable action spaces. Transactions on Machine Learning Research
• Wang, Jin & Montana (2024) Goal-conditioned offline reinforcement learning through state space partitioning. Machine Learning
• Hepburn & Montana (2024) Model-based trajectory stitching for improved behavioural cloning and its applications. Machine Learning
• Beeson & Montana (2024) Balancing policy constraint and ensemble size in uncertainty-based offline reinforcement learning. Machine Learning
• Ireland & Montana (2024) REValueD: regularised ensemble value-decomposition for factorisable Markov decision processes. ICLR
• Utke, Houssineau & Montana (2024) Embracing relational reasoning in multi-agent actor-critic. AAMAS
• Wang et al. (2024) GOPlan: goal-conditioned offline reinforcement learning by planning with learned models. Transactions on Machine Learning Research
• Pesce & Montana (2023) Learning multi-agent coordination through connectivity-driven communication. Machine Learning
• Ireland & Montana (2022) LeNSE: learning to navigate subgraph embeddings for large-scale combinatorial optimisation. ICML
• Kilinc & Montana (2022) Reinforcement learning for robotic manipulation using simulated locomotion demonstrations. Machine Learning
• Charlesworth & Montana (2021) Solving challenging dexterous manipulation tasks with trajectory optimisation and reinforcement learning. ICML (Spotlight)
• Charlesworth & Montana (2020) PlanGAN: model-based planning with sparse rewards and multiple goals. NeurIPS
• Pesce & Montana (2020) Improving coordination in small-scale multi-agent deep reinforcement learning through memory-driven communication. Machine Learning

Machine unlearning and federated learning (2025–present). A new line of work since 2025 addresses the removal of specific training data from deployed models, with methods that give exact or verifiable guarantees for distilled and federated models rather than approximate forgetting.

• Quan, Wu & Montana (2026) Exact federated continual unlearning for ridge heads on frozen foundation models. ECML PKDD
• Quan, Li & Montana (2025) Efficient verified unlearning for distillation. NeurIPS

Deep learning for medical imaging (2015–present). Work in medical imaging has driven the development of AI methods addressing distinctive deployment challenges: high-resolution images with low signal-to-noise ratios, severe class imbalance, longitudinal tracking, dataset shift from scanner and protocol variability across institutions, supervision ranging from pixel annotations to radiologist text reports, and clinical requirements for calibrated, interpretable, externally validated models. This work has produced models for automated reading and triage of chest radiographs, for lesion detection and cardiac segmentation, and for breast cancer risk prediction, several of which have been validated across institutions and released as open source.

• Wei et al. (2026) BREST-AI: three-year breast cancer risk prediction from screening mammograms. npj Breast Cancer
• Cid et al. (2024) Development and validation of open-source deep neural networks for comprehensive chest x-ray reading: a retrospective, multicentre study. The Lancet Digital Health
• Macpherson et al. (2024) Automated ranking of chest x-ray radiological finding severity in a binary label setting. MIDL
• Zhu, Liakata & Montana (2024) A multi-task transformer model for fine-grained labelling of chest x-ray reports. LREC-COLING
• Santeramo et al. (2024) Are better AI algorithms for breast cancer detection also better at predicting risk? A paired case–control study. Breast Cancer Research
• Macpherson et al. (2023) Patient reidentification from chest radiographs: an interpretable deep metric learning approach and its applications. Radiology: Artificial Intelligence
• Damiani et al. (2023) Evaluation of an AI model to assess future breast cancer risk. Radiology
• Brentnall et al. (2023) An optimization framework to guide the choice of thresholds for risk-based cancer screening. npj Digital Medicine
• Byrne et al. (2023) A persistent homology-based topological loss for CNN-based multiclass segmentation of CMR. IEEE Transactions on Medical Imaging
• Annarumma et al. (2019) Automated triaging of adult chest radiographs with deep artificial neural networks. Radiology
• Pesce et al. (2019) Learning to detect chest radiographs containing pulmonary lesions using visual attention networks. Medical Image Analysis

Statistical learning for neuroimaging and imaging genetics (2010–2018). Work in neuroimaging addressed fundamental challenges in understanding brain structure and function: extracting meaningful patterns from high-dimensional, noisy neuroimaging data (fMRI, structural MRI), understanding how brain organization varies across individuals while identifying population-level commonalities, integrating imaging data with genetic and clinical information, characterizing network topology and dynamics, and developing methods that scale to large cohorts while remaining interpretable for neuroscience discovery. This work introduced multivariate methods for relating genetic variation to brain imaging phenotypes, estimators of time-varying brain connectivity, closed-loop methods for designing neuroimaging experiments in real time, and early deep learning models for brain segmentation and brain-age prediction.

• Lorenz et al. (2018) Dissociating frontoparietal brain networks with neuroadaptive Bayesian optimization. Nature Communications
• Krishnan et al. (2017) Machine learning shows association between genetic variability in PPARG and cerebral connectivity in preterm infants. PNAS
• Cole et al. (2017) Predicting brain age with deep learning from raw imaging data results in a reliable and heritable biomarker. NeuroImage
• Monti et al. (2017) Real-time estimation of dynamic functional connectivity networks. Human Brain Mapping
• Monti, Anagnostopoulos & Montana (2017) Learning population and subject-specific brain connectivity networks via mixed neighborhood selection. The Annals of Applied Statistics
• Chung et al. (2016) Characterising brain network topologies: a dynamic analysis approach using heat kernels. NeuroImage
• Lorenz et al. (2016) The Automatic Neuroscientist: a framework for optimizing experimental design with closed-loop real-time fMRI. NeuroImage
• de Brébisson & Montana (2015) Deep neural networks for anatomical brain segmentation. CVPR Workshops
• Monti et al. (2014) Estimating time-varying brain connectivity networks from functional MRI time series. NeuroImage
• Silver et al. (2012) Identification of gene pathways implicated in Alzheimer’s disease using longitudinal imaging phenotypes with sparse regression. NeuroImage
• Vounou et al. (2010) Discovering genetic associations with high-dimensional neuroimaging phenotypes: a sparse reduced-rank regression approach. NeuroImage

High-dimensional and structured sparse models for genomics (2009–2015). Earlier work in genomics and multi-omics addressed challenges inherent to molecular data: extreme dimensionality where features vastly outnumber samples, rich biological structure encoded in pathway databases and regulatory networks, substantial technical and biological variability, non-Gaussian measurement distributions, and the need to integrate heterogeneous data types (genomics, transcriptomics, methylation, imaging) to understand complex phenotypes. This work developed penalised regression and matrix factorisation methods that use known biological structure, tests of association for heterogeneous data types, and models for time-course molecular data.

• Wang, Yuan & Montana (2015) Sparse multi-view matrix factorization: a multivariate approach to multiple tissue comparisons. Bioinformatics
• Ruan, Young & Montana (2015) Differential analysis of biological networks. BMC Bioinformatics
• Wang & Montana (2014) The graph-guided group lasso for genome-wide association studies. In Regularization, Optimization, Kernels, and Support Vector Machines
• Wang, Curry & Montana (2014) Network-guided regression for detecting associations between DNA methylation and gene expression. Bioinformatics
• Cozzini et al. (2014) A Bayesian mixture of lasso regressions with t-errors. Computational Statistics & Data Analysis
• Minas, Curry & Montana (2013) A distance-based test of association between paired heterogeneous genomic data. Bioinformatics
• Silver et al. (2013) Pathways-driven sparse regression identifies pathways and genes associated with high-density lipoprotein cholesterol in two Asian cohorts. PLoS Genetics
• Silver, Montana & Alzheimer's Disease Neuroimaging Initiative (2012) Fast identification of biological pathways associated with a quantitative trait using group lasso with overlaps. Statistical Applications in Genetics and Molecular Biology
• Berk, Ebbels & Montana (2011) A statistical framework for biomarker discovery in metabolomic time course data. Bioinformatics
• Minas, Waddell & Montana (2011) Distance-based differential analysis of gene curves. Bioinformatics

Selected media coverage

• AI as good as doctors at checking X-rays - study (BBC, 2023)
• World’s most advanced robotic hand is approaching human-level dexterity (Digital Trends, 2020)
• How AI could speed X-ray processing (BBC, 2019)
• See how this Artificial Intelligence can predict an abnormal chest X-ray (Forbes, 2019)
• This AI could hold the key to decoding human intelligence (Wired, 2017)
• The eyes don't have it: nose is most likely feature to be inherited (The Telegraph, 2017)
• Like mother, like daughter: face scanning technology reveals the features you're most likely to inherit from your parents (MailOnline, 2017)
• Deep-learning machine uses MRI scans to determine your brain age (MIT Technology Reviews, 2016)
• Predicting your brain age from MRI scans (NVIDIA Developer, 2016)

Contact details: g.montana@warwick.ac.uk