Kipu brings a quantum preprocessing step to IBM’s machine-learning users
Rimay joins IBM’s software catalogue, letting eligible teams test quantum-derived features without replacing their usual prediction models.
Reading tools for text size, copying, printing and your reading list
Kipu Quantum is bringing its Rimay machine-learning service to IBM’s Quantum Functions Catalog. The new route gives eligible IBM users a way to try quantum processing inside an existing data-analysis workflow, rather than replace the whole system.[1][2]
Kipu announced the addition on November 12, following IBM’s November 11 developer-conference update. Rimay itself is not new; the change is access through IBM’s catalogue.[1][2][4]
The idea is to change what a conventional model learns from. Rimay sends numerical data through quantum circuits and returns transformed features, the inputs a prediction model uses. A conventional classifier can then train on those features, alone or alongside the originals.[3]
Kipu’s benchmark, published in IBM’s documentation, shows why this remains an experiment, not a general claim that quantum makes machine learning better. A hybrid random-forest model achieved 77% accuracy on a credit-risk dataset, against 76.5% for its classical counterpart. On the 200-case test set, that difference amounts to one additional correct prediction. A separate classical model reached 77.5%.[3]
Recall improved, meaning more positive cases were detected, while precision fell, meaning more positive predictions were wrong. IBM cautions that the small test requires further verification. The useful question for a data team is whether the extra quantum step improves its own results enough to justify the computing resources.[3]
Access is limited to IBM Premium, Flex and On-Prem plans; the function remains beta. Larger, repeated comparisons against strong classical models would make the evidence more convincing.[3]