Presented by: Tianyu Zhao

Date: Friday, September 4, 2026
Time: 10:00 am - 11:00 am
Place: Zoom - see below.

Location: Remote via Zoom

Join Zoom Meeting  

https://zoom.us/j/97745041226?pwd=zPP50fkg8pKbnlOuPrJFu8KAGQ8qon.1

Meeting ID: 977 4504 1226  

Passcode: 5wEEdZ

Abstract:  

Double nanohole (DNH) optical tweezers provide a label-free platform for trapping nanoscale biological objects while simultaneously monitoring their optical transmission signals in real time. In this seminar, I will present our recent work using DNH optical tweezers for the characterization of extracellular vesicles and individual proteins. First, I will discuss the classification of extracellular vesicles derived from different breast cancer cell lines using optical trapping signals combined with machine-learning methods, demonstrating that the time-dependent trapping signal contains information that can distinguish different EV populations. I will then present studies of single-protein conformational dynamics, where multi-state trapping signals are analyzed to reconstruct temperature-dependent free-energy landscapes. In addition, I will introduce a zero-power extrapolation approach for investigating protein–surface interactions near the nanoaperture and show how the observed interaction depends on protein charge. Finally, I will discuss conalbumin as an example of how biochemical changes such as iron binding can influence both surface interactions and conformational dynamics. Together, these studies demonstrate the potential of DNH optical tweezers as a versatile platform for label-free, real-time biophysical measurements at the single-particle and single-protein level.

Presented by: Solomon Ameh

Date: Tuesday, September 8, 2026
Time: 11:00 am
Place: Zoom - see below.

Zoom Details:

Meeting link: https://us04web.zoom.us/j/77409286563?pwd=JskabVMJmHfLJsq0yPbabJdPg2XqRd.1

Meeting ID: 77409286563

Passcode: 7Q0zm5

Abstract:

Supply chain management (SCM) is a dynamic field with varying definitions, diverse terminology, and limited visibility across organizations, making it difficult to represent secure chains. This thesis proposes and defines a production view modeling framework in which the supply chain is represented as a sequence of individual production steps. The framework was designed to preserve more details than producer-level representations and to support structural analysis of where important production steps occur within the chain.

Warfield's partitioning methods are used to describe structural levels, dependence, influence, and independent regions of the supply chain, providing a basis for identifying production steps that may be critical from a security perspective. The framework was evaluated using large-scale software supply chains (SSCs) constructed from the npm, PyPI, and Maven ecosystems, containing 250,944 packages and 1,063,358 dependency relationships. The evaluation focused on previously legitimate packages that were later compromised in recent supply-chain campaigns, including incidents linked to Shai-Hulud and TeamPCP. We compared the production view representation with detectors that rely on package metadata and examined what happens when metadata associated with the assessed package becomes unreliable. When package metadata was complete and unchanged, the npm metadata detector achieved an area under the receiver operating characteristic (ROC-AUC) curve of $0.992$ and an area under the precision–recall (PR-AUC) curve of $0.929$. Under adversarial manipulation, however, performance decreased substantially to $0.435$ ROC-AUC and $0.019$ PR-AUC. In comparison, production view features derived from structural position, centrality, upstream and downstream reach, and relationships with compromised or vulnerable production steps remained comparatively stable, achieving a ROC-AUC of $0.736$ and a PR-AUC of $0.142$. We also used Warfield’s partitioning to evaluate a two-stage detector that propagates risk scores from upstream Warfield’s levels, improving the results to $0.804$ ROC-AUC and $0.201$ PR-AUC.

This work bridges theoretical supply-chain concepts and practical supply-chain security by defining an interpretable framework and demonstrating its application to a large software supply-chain dataset.

Presented by: Yifeng Bie

Date: Friday, September 11, 2026
Time: 9:30 am
Place: Zoom - see below.

https://uvic.zoom.us/j/8472200636?pwd=NTVCbHhZcnJYSU80VXlkSGlsMlF3Zz09&omn=88271444099

Meeting ID: 847 220 0636
Password: 700022

Note: Please log in to Zoom via SSO and your UVic Netlink ID

Abstract: 

Spectroscopic quantification often depends on large labeled datasets and computationally intensive models, which can limit practical deployment in real-time sensing applications. This work investigates the intrinsic structure of spectral data and shows that, when the signal dominates noise, a mixture containing chemical components can be effectively represented by approximately essential principal components. This low-dimensional structure indicates that much of the observed spectral dimensionality is redundant or noise-related. By retaining only the dominant functional principal components, the most informative spectral variation can be captured while substantially reducing the dimensionality of the original data.

Based on this observation, we develop an fPCA-based linear regression model and a nearly training-free quantification approach that uses known single-component extinction spectra to construct the spectral basis directly. This reduces dependence on large training datasets and simplifies the learning process to a compact concentration-mapping problem. The proposed methods achieve competitive quantification accuracy compared with conventional approaches such as PLSR and XGBoost, with the training-free method showing particular advantages in low-sample regimes. The framework is evaluated using simulated nine-gas infrared mixtures and further validated experimentally with Orange G and crystal violet solutions, demonstrating its effectiveness under spectral overlap, noise, and experimental variability.

Presented by: Yifeng Bie

Date: Saturday, September 12, 2026
Time: 10:00 am
Place: Zoom - see below.

Zoom Meeting Link:

https://uvic.zoom.us/j/8472200636?pwd=NTVCbHhZcnJYSU80VXlkSGlsMlF3Zz09&omn=83498705938

Meeting ID: 847 220 0636

Password: 700022

Note: Please log in to Zoom via SSO and your UVic Netlink ID

Abstract: 

 This work develops an unsupervised-learning framework for tracking whispering-gallery modes in a waveguide–microdisk coupled system. Instead of relying on computationally expensive full-field overlap calculations, each eigenmode is represented using a compact set of physically interpretable features. A hierarchical procedure is used to first separate guided and radiative modes, then classify the guided modes by polarization, and finally identify mode families across different geometric configurations using a reference-based tracking strategy.

The method is evaluated on a COMSOL dataset containing 3,000 eigenmodes from 40 geometric configurations. The proposed Gap-Reference approach achieves an average accuracy of 94.4% and a weighted F1-score of 92.8%, outperforming both K-means and agglomerative clustering. These results show that preserving modal continuity across a parameter sweep is more effective for mode tracking than treating the problem as conventional global clustering.