Which of the following is NOT a limitation of the support vector machine (SVM) technique?

Choosing C as the answer is correct because while support vector machines (SVM) can sometimes have performance issues in specific domains, it's not universally true that their accuracy is poor relative to neural networks across the board. SVM can be very effective in certain situations, especially when the data is well-separated and not too complex. In contrast, neural networks excel in tasks involving large amounts of data, especially with unstructured data such as images and text. This means that while they may outperform SVM in these scenarios, it doesn't imply that SVM lacks accuracy entirely across all data types and domains.

The other options highlight genuine limitations faced by SVMs. For instance, sensitivity to noise means that if there are outliers or irreparable errors in the training data, it can greatly affect the outcome of the model. The requirement for extensive computational resources often arises in high-dimensional spaces due to the nature of the SVM algorithm. Additionally, when faced with very complex datasets or overlapping classes, SVM may falter compared to more flexible models, which is indicative of its limitations. Thus, option C stands out as the only one that does not represent a definitive limitation of SVM.