Stuart Gibson
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Stuart Gibson has a total of 46 co-authors. Here are the top co-authors by number of shared publications:
Margarita Osadchy
Computer Science Department
5 shared publications- Margarita Osadchy
Department of Computer Science, University of Haifa, Haifa, Israel
4 shared publications Alex Poppe
School of Physics and Astronomy
3 shared publications- Bart de Nijs
NanoPhotonics Centre, Cavendish Laboratory, Department of Physics, University of Cambridge, Cambridge, UK
3 shared publications Fangliang Bai
2 shared publicationsJack Griffiths
2 shared publications
Latest Publications
- Open Access
- 14 May 2026
- Published by Springer Nature in Annals of Operations Research
- Vol. 362 (1-3) , 197-218
A p-median based approach to constrained clustering
Constrained clustering is a semi-supervised method for dividing multi-attribute datasets into groups of similar elements that incorporates restrictions on how data points should or should not be clustered together. In this study, we investigate the application of the p-median model coupled with instance-level constraints, specifically must-link and cannot-link constraint, for performing constrained clustering. A Lagrangian relaxation procedure is proposed to efficiently solve large problem instances. To further speed up solution times, a simple but highly effective variable reduction technique is devised for identifying a small set of potential cluster centers. We test our modeling framework and solution approach on a set of benchmark datasets and several real-world household electricity consumption datasets. We find that high-quality solutions with small optimality gaps can be obtained with moderate computational effort. In addition, analysis of the largest household electricity consumption dataset reveals that constrained p-median clusters have less extreme variability with respect to average energy usage, a more even spread of cluster sizes, and much less overlap between high- and low-income households within the same clusters compared to classic k-means clustering.
- 14 May 2026
- Published by Springer Nature in Annals of Operations Research
- Vol. 362 (1-3) , 197-218
Open AccessA p-median based approach to constrained clustering
Constrained clustering is a semi-supervised method for dividing multi-attribute datasets into groups of similar elements that incorporates restrictions on how data points should or should not be clustered together. In this study, we investigate the application of the p-median model coupled with instance-level constraints, specifically must-link and cannot-link constraint, for performing constrained clustering. A Lagrangian relaxation procedure is proposed to efficiently solve large problem instances. To further speed up solution times, a simple but highly effective variable reduction technique is devised for identifying a small set of potential cluster centers. We test our modeling framework and solution approach on a set of benchmark datasets and several real-world household electricity consumption datasets. We find that high-quality solutions with small optimality gaps can be obtained with moderate computational effort. In addition, analysis of the largest household electricity consumption dataset reveals that constrained p-median clusters have less extreme variability with respect to average energy usage, a more even spread of cluster sizes, and much less overlap between high- and low-income households within the same clusters compared to classic k-means clustering.
- Open Access
- 12 June 2025
- Published by Wiley in Journal of Raman Spectroscopy
- Vol. 56 (12) , 1569-1578
Autoencoding Raman Spectra to Predict Analyte Concentrations
Machine learning analysis has been applied to Raman data obtained in both nuclear and biopharmaceutical industrial applications. A 785‐nm Raman instrument using a spatial heterodyne spectrometer (SHS) was used to acquire Raman spectra for the nuclear dataset, whilst a new deep UV resonant SHS system, featuring a 228.5‐nm diode‐pumped solid‐state laser, was used to capture Raman spectra of biological macromolecule samples for the biopharmaceutical dataset. A key focus is on the practical challenges faced in the design of data processing tasks and machine learning architectures due to real‐world limitations in data collection. A fully connected (FC) autoencoder is employed as part of a regression task, which generates predictions on analyte concentrations in mixed substances. The method was shown to outperform industry standard regression tools, principal component regression (PCR) and partial least squares (PLS) regression, each used as comparative benchmarks, by over 50% in a test of model precision across the nuclear and biopharmaceutical datasets investigated in this work. Advancements in the precision, speed and effectiveness of such tools are of critical importance in an industrial environment. This is driven by compelling motivations to reduce not only the costs associated with these processes but also to increase the quality of resulting products or to reduce the risks within industrial operations, where applicable.
- 12 June 2025
- Published by Wiley in Journal of Raman Spectroscopy
- Vol. 56 (12) , 1569-1578
Open AccessAutoencoding Raman Spectra to Predict Analyte Concentrations
Machine learning analysis has been applied to Raman data obtained in both nuclear and biopharmaceutical industrial applications. A 785‐nm Raman instrument using a spatial heterodyne spectrometer (SHS) was used to acquire Raman spectra for the nuclear dataset, whilst a new deep UV resonant SHS system, featuring a 228.5‐nm diode‐pumped solid‐state laser, was used to capture Raman spectra of biological macromolecule samples for the biopharmaceutical dataset. A key focus is on the practical challenges faced in the design of data processing tasks and machine learning architectures due to real‐world limitations in data collection. A fully connected (FC) autoencoder is employed as part of a regression task, which generates predictions on analyte concentrations in mixed substances. The method was shown to outperform industry standard regression tools, principal component regression (PCR) and partial least squares (PLS) regression, each used as comparative benchmarks, by over 50% in a test of model precision across the nuclear and biopharmaceutical datasets investigated in this work. Advancements in the precision, speed and effectiveness of such tools are of critical importance in an industrial environment. This is driven by compelling motivations to reduce not only the costs associated with these processes but also to increase the quality of resulting products or to reduce the risks within industrial operations, where applicable.
- Open Access
- 22 May 2025
- Published by American Chemical Society (ACS) in ACS Nano
- Vol. 19 (21) , 20120-20127
Resolving Molecular Perturbations Near Undercoordinated Metals
Metal surfaces can act as efficient heterogeneous catalysts, but their underlying mechanisms are often poorly understood. This is due to the highly transient nature of the underpinning interactions occurring at the single-molecule level, making these difficult to resolve by using traditional analysis techniques. Here, we present a methodology to study metal-molecule interactions near undercoordinated binding sites using single-molecule surface-enhanced Raman spectroscopy (SERS). We demonstrate how machine learning can identify the metal-induced molecular perturbations by recognizing concurrent frequency wandering in vibrational energies, and we compare these peak displacements to extensive DFT modeling to reveal what interactions are occurring. This allows us to resolve how molecules are deformed as they interact with binding sites on metal surfaces. The work provides rare insight into the dynamics and behavior of molecules at catalytically active interfaces and can aid in the rational design of heterogeneous catalysts.
- 22 May 2025
- Published by American Chemical Society (ACS) in ACS Nano
- Vol. 19 (21) , 20120-20127
Open AccessResolving Molecular Perturbations Near Undercoordinated Metals
Metal surfaces can act as efficient heterogeneous catalysts, but their underlying mechanisms are often poorly understood. This is due to the highly transient nature of the underpinning interactions occurring at the single-molecule level, making these difficult to resolve by using traditional analysis techniques. Here, we present a methodology to study metal-molecule interactions near undercoordinated binding sites using single-molecule surface-enhanced Raman spectroscopy (SERS). We demonstrate how machine learning can identify the metal-induced molecular perturbations by recognizing concurrent frequency wandering in vibrational energies, and we compare these peak displacements to extensive DFT modeling to reveal what interactions are occurring. This allows us to resolve how molecules are deformed as they interact with binding sites on metal surfaces. The work provides rare insight into the dynamics and behavior of molecules at catalytically active interfaces and can aid in the rational design of heterogeneous catalysts.
- Open Access
- 3 April 2025
- Published by Elsevier in Pattern Recognition Letters
- Vol. 192, 136-143
Extended excitation backprop with gradient weighting: A general visualization solution for understanding heterogeneous face recognition
- 3 April 2025
- Published by Elsevier in Pattern Recognition Letters
- Vol. 192, 136-143
Open AccessExtended excitation backprop with gradient weighting: A general visualization solution for understanding heterogeneous face recognition
- 27 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. 46 (5) , 3845-3861
Reconstructing Randomly Masked Spectra Helps DNNs Identify Discriminant Wavenumbers
Nondestructive detection methods, based on vibrational spectroscopy, are vitally important in a wide range of applications including industrial chemistry, pharmacy and national defense. Recently, deep learning has been introduced into vibrational spectroscopy showing great potential. Different from images, text, etc. that offer large labeled data sets, vibrational spectroscopic data is very limited, which requires novel concepts beyond transfer and meta learning. To tackle this, we propose a task-enhanced augmentation network (TeaNet). The key component of TeaNet is a reconstruction module that inputs randomly masked spectra and outputs reconstructed samples that are similar to the original ones, but include additional variations learned from the domain. These augmented samples are used to train the classification model. The reconstruction and prediction parts are trained simultaneously, end-to-end with back-propagation. Results on both synthetic and real-world datasets verified the superiority of the proposed method. In the most difficult synthetic scenarios TeaNet outperformed CNN by 17%. We visualized and analysed the neuron responses of TeaNet and CNN, and found that TeaNet's ability to identify discriminant wavenumbers was excellent compared to CNN. Our approach is general and can be easily adapted to other domains, offering a solution to more accurate and interpretable few-shot learning.
- 27 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. 46 (5) , 3845-3861
Reconstructing Randomly Masked Spectra Helps DNNs Identify Discriminant Wavenumbers
Nondestructive detection methods, based on vibrational spectroscopy, are vitally important in a wide range of applications including industrial chemistry, pharmacy and national defense. Recently, deep learning has been introduced into vibrational spectroscopy showing great potential. Different from images, text, etc. that offer large labeled data sets, vibrational spectroscopic data is very limited, which requires novel concepts beyond transfer and meta learning. To tackle this, we propose a task-enhanced augmentation network (TeaNet). The key component of TeaNet is a reconstruction module that inputs randomly masked spectra and outputs reconstructed samples that are similar to the original ones, but include additional variations learned from the domain. These augmented samples are used to train the classification model. The reconstruction and prediction parts are trained simultaneously, end-to-end with back-propagation. Results on both synthetic and real-world datasets verified the superiority of the proposed method. In the most difficult synthetic scenarios TeaNet outperformed CNN by 17%. We visualized and analysed the neuron responses of TeaNet and CNN, and found that TeaNet's ability to identify discriminant wavenumbers was excellent compared to CNN. Our approach is general and can be easily adapted to other domains, offering a solution to more accurate and interpretable few-shot learning.
- Open Access
- 26 October 2023
- Published by American Chemical Society (ACS) in The Journal of Physical Chemistry Letters
- Vol. 14 (43) , 9793
Correction to “Mapping Atomic-Scale Metal–Molecule Interactions: Salient Feature Extraction through Autoencoding of Vibrational Spectroscopy Data”
- 26 October 2023
- Published by American Chemical Society (ACS) in The Journal of Physical Chemistry Letters
- Vol. 14 (43) , 9793
Open AccessCorrection to “Mapping Atomic-Scale Metal–Molecule Interactions: Salient Feature Extraction through Autoencoding of Vibrational Spectroscopy Data”
- Open Access
- 18 August 2023
- Published by American Chemical Society (ACS) in The Journal of Physical Chemistry Letters
- Vol. 14 (34) , 7603-7610
Mapping Atomic-Scale Metal–Molecule Interactions: Salient Feature Extraction through Autoencoding of Vibrational Spectroscopy Data
Atomic-scale features, such as step edges and adatoms, play key roles in metal-molecule interactions and are critically important in heterogeneous catalysis, molecular electronics, and sensing applications. However, the small size and often transient nature of atomic-scale structures make studying such interactions challenging. Here, by combining single-molecule surface-enhanced Raman spectroscopy with machine learning, spectra are extracted of perturbed molecules, revealing the formation dynamics of adatoms in gold and palladium metal surfaces. This provides unique insight into atomic-scale processes, allowing us to resolve where such metallic protrusions form and how they interact with nearby molecules. Our technique paves the way to tailor metal-molecule interactions on an atomic level and assists in rational heterogeneous catalyst design.
- 18 August 2023
- Published by American Chemical Society (ACS) in The Journal of Physical Chemistry Letters
- Vol. 14 (34) , 7603-7610
Open AccessMapping Atomic-Scale Metal–Molecule Interactions: Salient Feature Extraction through Autoencoding of Vibrational Spectroscopy Data
Atomic-scale features, such as step edges and adatoms, play key roles in metal-molecule interactions and are critically important in heterogeneous catalysis, molecular electronics, and sensing applications. However, the small size and often transient nature of atomic-scale structures make studying such interactions challenging. Here, by combining single-molecule surface-enhanced Raman spectroscopy with machine learning, spectra are extracted of perturbed molecules, revealing the formation dynamics of adatoms in gold and palladium metal surfaces. This provides unique insight into atomic-scale processes, allowing us to resolve where such metallic protrusions form and how they interact with nearby molecules. Our technique paves the way to tailor metal-molecule interactions on an atomic level and assists in rational heterogeneous catalyst design.
- Open Access
- 25 October 2021
- Published by Center for Open Science in PsyArXiv
Active Recognition Attempts Induce Updating of Face Memories
While successful retrieval typically strengthens memories, errors made during retrieval attempts can become encoded and bias subsequent remembering. It is not clear however whether such updating occurs during recognition of faces based on their visual properties. We investigated recognition-induced updating of face memories across three experiments, by comparing the effects of active recognition attempts against two control tasks that also exposed participants to erroneous face information while they were not trying to remember. Importantly, we used computer generated facial images drawn from locations in a multidimensional “face space” to match the degree of error that was introduced by the different tasks, enabling us to isolate the role of active retrieval processes in face memory updating. Participants first learned a series of target faces. Next, target faces were shown mixed with similar distractor faces and participants either actively tried to recognize the targets, or tried to encode one of the faces, or selected the face they thought was most distinctive. We then tested participants’ recognition memory for targets in a surprise final test, and measured to what extent their recognition errors on the final test were biased by their responses on the prior task. Across the three experiments, final recognition bias was significantly enhanced after active recognition attempts and was larger following recognition attempts compared to either control task. The findings extend on prior demonstrations that retrieval-induced updating occurs for semantically rich, complex memories by showing that engagement of active retrieval processes during visually-based face recognition can also induce updating.
- 25 October 2021
- Published by Center for Open Science in PsyArXiv
Open AccessActive Recognition Attempts Induce Updating of Face Memories
While successful retrieval typically strengthens memories, errors made during retrieval attempts can become encoded and bias subsequent remembering. It is not clear however whether such updating occurs during recognition of faces based on their visual properties. We investigated recognition-induced updating of face memories across three experiments, by comparing the effects of active recognition attempts against two control tasks that also exposed participants to erroneous face information while they were not trying to remember. Importantly, we used computer generated facial images drawn from locations in a multidimensional “face space” to match the degree of error that was introduced by the different tasks, enabling us to isolate the role of active retrieval processes in face memory updating. Participants first learned a series of target faces. Next, target faces were shown mixed with similar distractor faces and participants either actively tried to recognize the targets, or tried to encode one of the faces, or selected the face they thought was most distinctive. We then tested participants’ recognition memory for targets in a surprise final test, and measured to what extent their recognition errors on the final test were biased by their responses on the prior task. Across the three experiments, final recognition bias was significantly enhanced after active recognition attempts and was larger following recognition attempts compared to either control task. The findings extend on prior demonstrations that retrieval-induced updating occurs for semantically rich, complex memories by showing that engagement of active retrieval processes during visually-based face recognition can also induce updating.
- Open Access
- 22 June 2021
- Published by IOP Publishing in Journal of Physics: Condensed Matter
- Vol. 33 (32) , 324002
Principal component analysis of diffuse magnetic neutron scattering: a theoretical study
We present a theoretical study of the potential of principal component (PC) analysis to analyse magnetic diffuse neutron scattering data on quantum materials. To address this question, we simulate the scattering function S q for a model describing a cluster magnet with anisotropic spin–spin interactions under different conditions of applied field and temperature. We find high dimensionality reduction and that the algorithm can be trained with surprisingly small numbers of simulated observations. Subsequently, observations can be projected onto the reduced-dimensionality space defined by the learnt PCs. Constant-field temperature scans correspond to trajectories in this space which show characteristic bifurcations at the critical fields corresponding to ground-state phase boundaries. Such plots allow the ground-state phase diagram to be accurately determined from finite-temperature measurements.
- 22 June 2021
- Published by IOP Publishing in Journal of Physics: Condensed Matter
- Vol. 33 (32) , 324002
Open AccessPrincipal component analysis of diffuse magnetic neutron scattering: a theoretical study
We present a theoretical study of the potential of principal component (PC) analysis to analyse magnetic diffuse neutron scattering data on quantum materials. To address this question, we simulate the scattering function S q for a model describing a cluster magnet with anisotropic spin–spin interactions under different conditions of applied field and temperature. We find high dimensionality reduction and that the algorithm can be trained with surprisingly small numbers of simulated observations. Subsequently, observations can be projected onto the reduced-dimensionality space defined by the learnt PCs. Constant-field temperature scans correspond to trajectories in this space which show characteristic bifurcations at the critical fields corresponding to ground-state phase boundaries. Such plots allow the ground-state phase diagram to be accurately determined from finite-temperature measurements.
- 1 September 2020
- Published by Elsevier in Neurocomputing
- Vol. 406, 169-181
LSHR-Net: A hardware-friendly solution for high-resolution computational imaging using a mixed-weights neural network
- 1 September 2020
- Published by Elsevier in Neurocomputing
- Vol. 406, 169-181
LSHR-Net: A hardware-friendly solution for high-resolution computational imaging using a mixed-weights neural network