Mu-DisCoCat: A Variational Pipeline for Compositional Generalization on Quantum Processors
Mu-DisCoCat uses variational quantum circuits to improve compositional generalization on noisy near-term quantum hardware.
The paper introduces Mu-DisCoCat, a multimodal variational quantum framework that maps compositional distributional semantics onto variational quantum circuits for compositional concept generalization. It learns stable object representations from single-object image-text pairs, then fixes them to learn relations in multi-object scenes. In classical simulation, Uhlmann state fidelity produced higher relational out-of-distribution accuracy than the evaluated CLIP baseline. On IBM fake backends, IQM FakeAphrodite, and the IBM Marrakesh processor, noisy hardware results remained strongly correlated with simulated fidelities.
- Mu-DisCoCat learns object representations first, then relations between them.
- Classical simulation beat the evaluated CLIP baseline on relational out-of-distribution accuracy.
- Tests used the destructive SWAP test on IBM and IQM noisy emulators and IBM Marrakesh.
- Hardware results stayed positively correlated with simulated fidelities despite device noise.
- Tags
- quantum-machine-learningdiscocatvariational-quantum-circuitscompositional-generalizationclipibmmultimodal
- Products
- IBM Marrakesh
- AI models
- CLIP
From the original article
1 min read at arxiv.orgAchieving compositional concept generalization (CoCoGen), the ability to understand novel situations by recombining learned primitives, remains a fundamental challenge in artificial intelligence. Compositional semantic models such as Compositional Distributional Semantics (DisCoCat) offer solutions by generalising vectors to tensors, but suffer from scaling bottlenecks when learning the tensors.
By Mina Abbaszadeh · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.