A tale of two papers
Why reading actual research papers in AI matters.
Today’s diptych is a little unconventional. It explains why it is useful to not just read between the lines of a news report but read the actual lines from the source.
Left Panel
Financial Times news report on a quantum trading ‘breakthrough’. It states that the use of quantum computing led to a breakthrough in performance - “34 per cent better than traditional means in predicting how likely an order was to be filled”. Reading the news article, you would have thought that they managed to build a groundbreaking machine learning or AI method that works with quantum computing. If they did, it would indeed have been a major shift.
Right Panel
The actual paper from HSBC and IBM with a much more boring (but clear) title. Super interesting work though. I think the researchers did great work.
What did they do? They used quantum computing techniques as a method for transforming features (or attributes) associated with historical bond trading data. That’s about it for the quantum part. The rest is simple machine learning.
The only thing quantum was that they transformed around 200 features (from the bond trading data) to around 300 (quantum processed) features. Then they used these as inputs to fairly simple logistic regression, gradient boosting, random forest and simple neural network models running on normal (and not quantum) machines. That’s it.
And that 34% boost that seems earth shaking? It’s 34% compared to just feeding the original 200+ features to the same logistic regression, gradient boosting, random forest and simple neural network models. Feature transformation has been around in AI for ages, and if one had tried to transform that 200+ features with some other well known non-quantum technique, there is a possibility that the 34% gap could have been much smaller or even non-existent.
The authors of the paper were upfront and clear about this though. But you would have obtained a really wrong impression of what transpired from media reports.
And the most interesting result from the paper - much more interesting than the 34% performance gain?
Their best results came from when the features had more NOISE due to the quantum hardware. Interesting because a lot of the work on quantum computing revolves around minimizing that noise.
Interestingly enough, that’s how I would characterize media reports on this fascinating experiment by HSBC and IBM.
But the paper itself? Lots of useful insights.
hashtag#QuantumComputing hashtag#FinTech hashtag#AI hashtag#AIRiskManagement
Link to paper - https://arxiv.org/pdf/2509.17715v1