Machine Learning and AI
Model Management
Detecting performance drift in Machine Learning models is a crucial and necessary task, but it's also very challenging. Often, data drift is the cause, and, as data labels are often difficult or expensive to obtain, an approach that avoids the need for labels is desirable.
Luckily, an IBM Research team has proposed a new feature-space drift detection method based on feature space rules (data slices) and provides good evidence of its effectiveness.
Their paper Machine Learning Model Drift Detection Via Weak Data Slices is available on arXiv
Quantum Computing
Quantum Optimisation
Quantum Annealing is an excellent tool in solving optimisation problems.
However, real world examples are still quite hard to find.
In October 2023, D-Wave broadcast a webinar on the subject, showing how Quantum Annealing could be used in resource allocation problems.
Significantly, the webinar included a discussion of when a quantum approach could outperform a classical optimisation solution.
The D-Wave folk have kindly made the recording available on YouTube: Optimizing resource allocation in manufacturing: Job shop scheduling using Quantum Computing.
Quantum Machine Learning
While QML algorithms are the subject of active research, QML has suffered from having few equivalents to such datasets as the MNIST database of handwritten digits, the Hello World of Machine Learning.
To help address this problem a small team funded by the University of Sydney have created the QDataSet. From their paper QDataset: Quantum Datasets for Machine Learning: "to address this gap in QML research by presenting a comprehensive QML dataset as a dedicated resource designed for researchers across classical and quantum computation to develop and train hybrid classical-quantum algorithms for use in theoretical and applied settings"
If you just want to access the data, it's here: Perrier E., Youssry A., Ferrie C. (2021). QDataSet: Quantum Datasets for Machine Learning (version 1.0.0). DOI: https://doi.org/10.5281/zenodo.5202814
Updated Tools
Following the milestone release of IBM's Qiskit v1.0 in February 2024, the ever improving library is now at metaversion 1.2.4. The significant changes are explained in the release notes.
Improved Technology
The control of multiple qubits is a significant issue for current quantum computing solution. This breakthrough provides a way to initialise a whole array of qubits using a single field approach: Quantum Computation Protocol for Dressed Spins in a Global Field