Abstract:Quantum machine learning is often highlighted as one of the most promising practical applications for which quantum computers could provide a computational advantage. However, a major obstacle to the widespread use of quantum machine learning models in practice is that these models, even once trained, still require access to a quantum computer in order to be evaluated on new data. To solve this issue, we introduce a new class of quantum models where quantum resources are only required during training, while the deployment of the trained model is classical. Specifically, the training phase of our models ends with the generation of a 'shadow model' from which the classical deployment becomes possible. We prove that: i) this class of models is universal for classically-deployed quantum machine learning; ii) it does have restricted learning capacities compared to 'fully quantum' models, but nonetheless iii) it achieves a provable learning advantage over fully classical learners, contingent on widely-believed assumptions in complexity theory. These results provide compelling evidence that quantum machine learning can confer learning advantages across a substantially broader range of scenarios, where quantum computers are exclusively employed during the training phase. By enabling classical deployment, our approach facilitates the implementation of quantum machine learning models in various practical contexts.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is the deployment difficulties of quantum machine - learning models in practical applications. Specifically, traditional quantum machine - learning models need to rely on quantum computers during both the training and deployment stages, which greatly limits their wide application because the availability of quantum computers is currently very limited. To solve this problem, the author proposes a new class of quantum models. These models only require quantum resources during the training stage and can be completely evaluated using classical computers during the deployment stage. ### Core Contributions of the Paper 1. **Introduction of "Shadow Model"**: - The author proposes a new class of quantum models called "Shadow Model". The characteristic of this model is that it uses a quantum computer to generate a "shadow", that is, a classical representation during the training stage, so that the trained model can be evaluated on a classical computer. 2. **Proof of the Universality and Advantages of the Shadow Model**: - The author proves that this type of shadow model is universal for quantum machine - learning with classical deployment, and although their learning ability is somewhat limited, they can still provide learning advantages over fully classical models in some cases. 3. **Solving Key Open Problems**: - The paper solves two important open problems: 1. Can the shadow model show quantum advantages over fully classical models in some tasks? 2. Are there quantum models for which shadow models cannot be efficiently generated? ### Specific Implementation Methods - **Flipped Model**: - The author defines a flipped model \( f_\theta(x)=\text{Tr}[\rho(\theta)O(x)] \), where \( \rho(\theta) \) is a parameterized quantum state and \( O(x) \) is an observable encoding data. In this way, it is more convenient to construct a shadow model. - **Shadow Tomography**: - Using shadow tomography technology, classical information can be extracted from the quantum state \( \rho(\theta) \) to approximately calculate the output of the model on a classical computer. ### Experimental and Theoretical Support - **Theoretical Proof**: - Based on complexity - theory assumptions, the author proves that in some learning tasks, the shadow model can show quantum advantages. - **Experimental Verification**: - Through specific examples and experiments, the effectiveness and feasibility of the shadow model are verified. ### Conclusion This paper effectively solves the problem of quantum machine - learning models relying on quantum computers in practical applications by introducing the shadow model, enabling quantum machine - learning models to be applied in more scenarios. At the same time, through strict theoretical analysis and experimental verification, the author shows the possibility of the shadow model having quantum advantages in some tasks.

Shadows of quantum machine learning

Quantum machine learning beyond kernel methods

Quantum machine learning: a classical perspective

Quantum enhancements for deep reinforcement learning in large spaces

Single-shot quantum machine learning

Advantage of Quantum Machine Learning from General Computational Advantages

Quantum Machine Learning: Fad or Future?

ShadowNet for Data-Centric Quantum System Learning

Quantum machine learning: from physics to software engineering

Power of data in quantum machine learning

Classical versus quantum models in machine learning: insights from a finance application

Is Quantum Advantage the Right Goal for Quantum Machine Learning?

Quantum Machine Learning For Classical Data

Quantum Machine Learning: Performance and Security Implications in Real-World Applications

Entanglement-induced provable and robust quantum learning advantages

Machine Learning for Quantum Computing Specialists

Quantum-Train: Rethinking Hybrid Quantum-Classical Machine Learning in the Model Compression Perspective

Machine Learning for Practical Quantum Error Mitigation

Quantum Error Mitigated Classical Shadows

Machine Learning in the Quantum Age: Quantum vs. Classical Support Vector Machines

Exponential separations between classical and quantum learners