In the field of quantum computing, one of the most critical challenges is quantum state tomography, a process that seeks to reconstruct the complete description of a quantum system from measurements. Traditionally, this task requires a huge number of samples, making it impractical for systems with more than a few qubits. However, recent research is showing that real quantum states do not occupy the entire Hilbert space, but live in low-dimensional manifolds that can be learned efficiently. A promising approach is to represent the quantum state using a matrix product state (MPS), a structure that allows capturing local correlations with polynomial complexity in the number of qubits. On this basis, two ways have been explored: one that uses a previous generative model and then infers a posteriori using the measurements, and another, more novel one, that trains a fixed protocol amortized estimator. The latter is especially relevant because it avoids dependence on an overly strong prior and demonstrates that true efficiency in measurement depends on how the measurements are designed, not just on the model.
The key is in the design of the measurement protocol. While random measurements are common in tomography, for an MPS state it is much more informative to measure local Pauli sets, which directly reveal the low-density arrays. When conditioning the estimator on these informative measurements, the fidelity jumps from values close to random (approximately 0.36) to more than 0.95, far exceeding the approach used only by the prior. This result is not trivial: a control with shuffled measurements shows that an estimator that does not actually use the measurement information does not pass the test, while the new model does. In addition, the amortized estimator includes a predictive uncertainty mechanism through a conformally recalibrated dropout ensemble that offers 90% coverage intervals even for observables that were never measured, something that shot-based tomography cannot provide.
For companies looking to integrate quantum computing into their processes, these advances open the door to bespoke applications in materials simulation, financial optimisation or cryptography. The ability to reconstruct quantum states with few measurements and calibrated uncertainty is essential to validate experiments in real hardware, as demonstrated in an IBM processor with five states reaching fidelities of 0.97. However, implementing these solutions requires tailored software that manages the complexity of the pipeline: from the preparation of quantum circuits to the processing of measurement data and state inference. This is where companies like Q2BSTUDIO bring their expertise in developing platforms that integrate artificial intelligence, AWS and Azure cloud services, and cybersecurity tools to ensure the integrity of quantum data.
Artificial intelligence for companies plays a dual role in this context. On the one hand, AI models, such as neural networks that learn MPS representation, allow for much more efficient tomography than traditional methods. On the other hand, the predictive uncertainty generated by those same models can be used to make informed decisions in environments where measurement is expensive or noisy. For example, an AI agent could decide which measurements to perform next based on current uncertainty, optimizing resource usage. This type of optimization is key in business intelligence services and Power BI when integrating quantum data with business dashboards to monitor the performance of quantum algorithms in real time.
The scalability of the amortized MPS method is remarkable: it maintains fidelities greater than 0.90 for 10-qubit systems and the gain from prior increases with the size of the system. Even with a bond dimension of 4, the fidelity remains at 0.88. This suggests that the approach is viable for medium-sized systems, which are precisely those that are beginning to be accessible on current quantum hardware. Polynomial parameterization allows up to 20 qubits to be contracted natively, which opens the door to practical applications in quantum chemistry or condensed matter theory. However, for these techniques to reach the market, it is necessary to have AI for companies that not only implements the algorithms, but packages them into robust solutions, with AWS and Azure cloud services that guarantee the scalability and cybersecurity necessary for production environments.
In short, fixed-protocol amortized quantum tomography represents a paradigm shift: from reliance on huge data sets to measurement design intelligence and the ability to learn efficiently. For any organization that wants to explore quantum computing as a competitive advantage, understanding these fundamentals is the first step. And to make the leap to real implementation, having a technology partner that offers both custom application development and integration of artificial intelligence and cloud services is essential. Q2BSTUDIO is prepared to accompany that journey, combining deep technical knowledge with a practical business vision.
